title
stringlengths 1
300
| score
int64 0
8.54k
| selftext
stringlengths 0
40k
| created
timestamp[ns] | url
stringlengths 0
780
| author
stringlengths 3
20
| domain
stringlengths 0
82
| edited
timestamp[ns] | gilded
int64 0
2
| gildings
stringclasses 7
values | id
stringlengths 7
7
| locked
bool 2
classes | media
stringlengths 646
1.8k
⌀ | name
stringlengths 10
10
| permalink
stringlengths 33
82
| spoiler
bool 2
classes | stickied
bool 2
classes | thumbnail
stringlengths 4
213
| ups
int64 0
8.54k
| preview
stringlengths 301
5.01k
⌀ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Looking for a way to edit LLM layers without fine-tuning | 1 | [removed] | 2025-01-28T12:08:53 | https://www.reddit.com/r/LocalLLaMA/comments/1ic041w/looking_for_a_way_to_edit_llm_layers_without/ | Haghiri75 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic041w | false | null | t3_1ic041w | /r/LocalLLaMA/comments/1ic041w/looking_for_a_way_to_edit_llm_layers_without/ | false | false | self | 1 | null |
Clear Difference LoL | 1 | 2025-01-28T12:10:49 | https://www.reddit.com/gallery/1ic059y | Glittering_Touch_256 | reddit.com | 1970-01-01T00:00:00 | 0 | {} | 1ic059y | false | null | t3_1ic059y | /r/LocalLLaMA/comments/1ic059y/clear_difference_lol/ | false | false | 1 | null |
||
DeepSeek has a better incentive than selling its data to Chinese government | 0 | So, a hedge fund made DeepSeek. Quant hedge funds rely on information unavailable to majority of the investors. It seems pretty obvious for DeepSeek to use the data they gather from prompts to do complex sentiment analysis based on it's user base. My bet is that within a few months the hedge fund will have very, very high returns on investment, and this doesn't necessarily involve the chinese government. That data harvest is golden for them. | 2025-01-28T12:10:56 | https://www.reddit.com/r/LocalLLaMA/comments/1ic05cr/deepseek_has_a_better_incentive_than_selling_its/ | Temporary-Bug4124 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic05cr | false | null | t3_1ic05cr | /r/LocalLLaMA/comments/1ic05cr/deepseek_has_a_better_incentive_than_selling_its/ | false | false | self | 0 | null |
Deepseek getting hammered | 19 | I haven't been able to use the API reliably for hours. Are they really under cyber attack or just swamped? If they are under attack, is it CIA level or just disgruntled NVIDIA investors | 2025-01-28T12:11:58 | https://www.reddit.com/r/LocalLLaMA/comments/1ic05zv/deepseek_getting_hammered/ | Nobby_Binks | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic05zv | false | null | t3_1ic05zv | /r/LocalLLaMA/comments/1ic05zv/deepseek_getting_hammered/ | false | false | self | 19 | null |
Do you see any real use cases for LLM? | 1 | [removed] | 2025-01-28T12:16:13 | https://www.reddit.com/r/LocalLLaMA/comments/1ic08kk/do_you_see_any_real_use_cases_for_llm/ | DamageDistinct531 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic08kk | false | null | t3_1ic08kk | /r/LocalLLaMA/comments/1ic08kk/do_you_see_any_real_use_cases_for_llm/ | false | false | self | 1 | null |
Struggling with LLMs for my use case | 1 |
I’ve been playing around with a few models on LM studio and Ollama + chatbox for a while and I am struggling to get a model working for the use case that I have.
I’m trying to use an LLM to assist in writing a script for documentaries / retelling of real events. For example, I’d like to provide several sources of an event (online news articles, wikipedia etc) that occurred and have my LLM summarise that information and create a high level script for the documentary that has key details. Not feature length, just 10 to 20 minutes worth of script writing.
I’ve used several models to try and attempt this but I’m getting stuck at getting these models to read links / URLs to the outside world or even do more than summarise the content of a document I’ve uploaded with the information I want it to use.
They all lie, make up information and ignore prompts. Has anyone encountered anything for a similar use case?
I’m not looking for an out of the box solution, just some general guidance on what I should be looking into to flesh out my use case as this is all rather new to me. | 2025-01-28T12:20:44 | https://www.reddit.com/r/LocalLLaMA/comments/1ic0b9v/struggling_with_llms_for_my_use_case/ | quduvfowpwbsjf | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic0b9v | false | null | t3_1ic0b9v | /r/LocalLLaMA/comments/1ic0b9v/struggling_with_llms_for_my_use_case/ | false | false | self | 1 | null |
Seeking Advice: Fine-Tuning Lightweight AI using API (phi-2/deepseek/mistral) for Consistent Incident Report Summaries | 2 | Hi everyone,
I’m working on a lightweight AI setup (currently using phi-2 but i tested Deepseek and mistral) to automate the generation of concise incident reports from data retrieved via my ticketing tool. I specifically chose a lightweight model to keep the process efficient since it only involves summarizing incidents.
Here’s an example of the script I’m using to call the local AI:
`def call_local_ai(incident_details):`
`log_step("Calling the local AI...")`
`url = "http://127.0.0.1:8080/v1/chat/completions"`
`headers = {"Content-Type": "application/json"}`
`prompt = f"""Analyze the following cybersecurity incident and create a concise summary. Focus on the key facts and actions taken.`
`IMPORTANT: Your response must contain ONLY two sections - 'Summary' and 'Status', formatted exactly as shown below:`
`Summary:`
`[Write a brief overview of what happened, when it happened, and what systems were impacted]`
`Status:`
`[Describe the current situation, what actions were taken, and any pending items]`
`IMPORTANT: Ensure the "Summary" section and the "Status" section is no longer than 50 words.`
`Here is the incident to analyze:`
`{incident_details}"""`
`data = {`
`"model": "phi-2",`
`"messages": [`
`{`
`"role": "user",`
`"content": prompt`
`}`
`],`
`"max_tokens": 1024,`
`"threads": 4,`
`"temperature": 0.7,`
`"top_p": 1,`
`"frequency_penalty": 0.0,`
`"random_seed": 42,`
`"presence_penalty": 0.6`
`}`
The issue I’m facing is inconsistent results:
Even with the same fixed seed and identical input prompts, the output quality varies significantly.
Sometimes the summaries are concise and accurate, while at other times, they’re irrelevant or poorly structured.
I’ve experimented with parameters like temperature, top\_p, and presence\_penalty but still can’t achieve reliable performance.
I’d appreciate any advice on best practices for tuning lightweight models like phi-2 to achieve consistent summarization results. Additionally, if there are alternative lightweight models better suited for this type of task, I’d love to hear about them. Finally, any insights into parameter configurations that could help stabilize outputs and improve reliability would be greatly appreciated.
Thanks in advance for your help and suggestions! | 2025-01-28T12:22:54 | https://www.reddit.com/r/LocalLLaMA/comments/1ic0cjg/seeking_advice_finetuning_lightweight_ai_using/ | ZemmourUndercut | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic0cjg | false | null | t3_1ic0cjg | /r/LocalLLaMA/comments/1ic0cjg/seeking_advice_finetuning_lightweight_ai_using/ | false | false | self | 2 | null |
Finetuning LLMs for a chatbot | 1 | [removed] | 2025-01-28T12:25:26 | https://www.reddit.com/r/LocalLLaMA/comments/1ic0e48/finetuning_llms_for_a_chatbot/ | Spare-Menu3360 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic0e48 | false | null | t3_1ic0e48 | /r/LocalLLaMA/comments/1ic0e48/finetuning_llms_for_a_chatbot/ | false | false | self | 1 | null |
How to handle natural language queries with SAP OData APIs? | 1 | [removed] | 2025-01-28T12:31:26 | https://www.reddit.com/r/LocalLLaMA/comments/1ic0hts/how_to_handle_natural_language_queries_with_sap/ | Background_Thanks604 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic0hts | false | null | t3_1ic0hts | /r/LocalLLaMA/comments/1ic0hts/how_to_handle_natural_language_queries_with_sap/ | false | false | self | 1 | null |
What to use for local live TTS with consumer nvidia card | 3 | I am working on a project am looking to do text to speech on the fly via api with either an anime voice, or using a system I could clone an anime voice with. I tried the latest kokoro 82m fast api and there is one voice there that would fit, however it is flat. No inflection or excitement. I was looking at F5 TTS, however it is not clear if it is fast enough for text to speech on the fly with a 3050 or 3090 for example, or if I could access it via an api in some way. Can someone point me in the right direction? | 2025-01-28T12:31:46 | https://www.reddit.com/r/LocalLLaMA/comments/1ic0i19/what_to_use_for_local_live_tts_with_consumer/ | nosimsol | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic0i19 | false | null | t3_1ic0i19 | /r/LocalLLaMA/comments/1ic0i19/what_to_use_for_local_live_tts_with_consumer/ | false | false | self | 3 | null |
Reasoning models vs non-reasoning models | 6 | I'm just interested to have a discussion about this. Which ***type*** of models do you guys prefer or do you believe that both types have different use cases? And do you think that reasoning models are the future or that we'll have both?
As for myself, I've noticed that unless there's a single concrete answer, I will prefer the regular models.
For example, if I need a list of items to buy to fix something mechanical in real life and want to ask the model about this, I've noticed that the regular models are better for this because the answer has multiple items.
Whereas, the reasoning model kind of struggles when there are multiple options and tries to narrow it down to only the best items, which doesn't always include everything. Have you guys noticed that as well? | 2025-01-28T12:32:52 | https://www.reddit.com/r/LocalLLaMA/comments/1ic0ipi/reasoning_models_vs_nonreasoning_models/ | Many_SuchCases | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic0ipi | false | null | t3_1ic0ipi | /r/LocalLLaMA/comments/1ic0ipi/reasoning_models_vs_nonreasoning_models/ | false | false | self | 6 | null |
By the sunset, he was right. Right? (R1/Thinking) | 1 | 2025-01-28T12:40:02 | https://www.reddit.com/gallery/1ic0mzq | apimash | reddit.com | 1970-01-01T00:00:00 | 0 | {} | 1ic0mzq | false | null | t3_1ic0mzq | /r/LocalLLaMA/comments/1ic0mzq/by_the_sunset_he_was_right_right_r1thinking/ | false | false | 1 | null |
||
Why is there no universal GUI for All kinds of AI tools? | 1 | [removed] | 2025-01-28T12:42:06 | https://www.reddit.com/r/LocalLLaMA/comments/1ic0obn/why_is_there_no_universal_gui_for_all_kinds_of_ai/ | SensitiveGood7260 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic0obn | false | null | t3_1ic0obn | /r/LocalLLaMA/comments/1ic0obn/why_is_there_no_universal_gui_for_all_kinds_of_ai/ | false | false | self | 1 | null |
What model is DeepSeek-R1 online? | 7 | excuseme if this is dumb question, im complete amateur in this, but im curious: i know you can download different models of DeepSeek-R1 localy and they ranging from 1,5gb size up to 402gb.. but what of this models use online version of DeepSeek? Thank you. | 2025-01-28T12:53:07 | https://www.reddit.com/r/LocalLLaMA/comments/1ic0v57/what_model_is_deepseekr1_online/ | 1711198430497251 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic0v57 | false | null | t3_1ic0v57 | /r/LocalLLaMA/comments/1ic0v57/what_model_is_deepseekr1_online/ | false | false | self | 7 | null |
Deepseek is accessing data after it's cutoff date | 1 | [removed] | 2025-01-28T12:54:46 | https://www.reddit.com/r/LocalLLaMA/comments/1ic0w7c/deepseek_is_accessing_data_after_its_cutoff_date/ | Resident_Bag_6949 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic0w7c | false | null | t3_1ic0w7c | /r/LocalLLaMA/comments/1ic0w7c/deepseek_is_accessing_data_after_its_cutoff_date/ | false | false | 1 | null |
|
deepseek-r1 chat: what am I missing? | 1 | 2025-01-28T13:00:58 | https://www.reddit.com/r/LocalLLaMA/comments/1ic10ad/deepseekr1_chat_what_am_i_missing/ | martinsoderholm | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic10ad | false | null | t3_1ic10ad | /r/LocalLLaMA/comments/1ic10ad/deepseekr1_chat_what_am_i_missing/ | false | false | 1 | null |
||
What and how much hardware would you need to run the full DeepSeek R1 671B model? | 1 | [removed] | 2025-01-28T13:02:55 | https://www.reddit.com/r/LocalLLaMA/comments/1ic11pq/what_and_how_much_hardware_would_you_need_to_run/ | RapidRaid | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic11pq | false | null | t3_1ic11pq | /r/LocalLLaMA/comments/1ic11pq/what_and_how_much_hardware_would_you_need_to_run/ | false | false | self | 1 | null |
Imao Chinese Robots on China Spring Festival Gala 🤣 | 1 | [removed] | 2025-01-28T13:04:37 | https://v.redd.it/89hvja7ugqfe1 | BeautifulSecure4058 | v.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ic12y6 | false | {'reddit_video': {'bitrate_kbps': 5000, 'dash_url': 'https://v.redd.it/89hvja7ugqfe1/DASHPlaylist.mpd?a=1740661491%2CMjg1ZjE0OGJmMzkwODY2ZmI2YTc2NjQyZGY3NjUyNDVjOTI3ZjhlMjUyZWExMzY4YTFjNDU4Nzg4MjNlNDhjMA%3D%3D&v=1&f=sd', 'duration': 7, 'fallback_url': 'https://v.redd.it/89hvja7ugqfe1/DASH_1080.mp4?source=fallback', 'has_audio': True, 'height': 1920, 'hls_url': 'https://v.redd.it/89hvja7ugqfe1/HLSPlaylist.m3u8?a=1740661491%2CMTA2NzFiMGM0OTZkMmQ2NmQ4YjRmYzgzZjE5ZjM5ZmQwODQwY2RiZmUxZWMzODBmNmYxZDMyODk2ZDc1NDFmYQ%3D%3D&v=1&f=sd', 'is_gif': False, 'scrubber_media_url': 'https://v.redd.it/89hvja7ugqfe1/DASH_96.mp4', 'transcoding_status': 'completed', 'width': 1080}} | t3_1ic12y6 | /r/LocalLLaMA/comments/1ic12y6/imao_chinese_robots_on_china_spring_festival_gala/ | false | false | 1 | {'enabled': False, 'images': [{'id': 'aWY2NWhlOHFncWZlMW6iYpr5Sd7xeiezPVzK1c1mK2iYk0Soz5Kqi4XNSGUb', 'resolutions': [{'height': 192, 'url': 'https://external-preview.redd.it/aWY2NWhlOHFncWZlMW6iYpr5Sd7xeiezPVzK1c1mK2iYk0Soz5Kqi4XNSGUb.png?width=108&crop=smart&format=pjpg&auto=webp&s=68db0bc8b23fefe0f0e83fe37699dcd3684ca4a7', 'width': 108}, {'height': 384, 'url': 'https://external-preview.redd.it/aWY2NWhlOHFncWZlMW6iYpr5Sd7xeiezPVzK1c1mK2iYk0Soz5Kqi4XNSGUb.png?width=216&crop=smart&format=pjpg&auto=webp&s=483e5b8140e5ffca49e137c5a2022042eb1ed11d', 'width': 216}, {'height': 568, 'url': 'https://external-preview.redd.it/aWY2NWhlOHFncWZlMW6iYpr5Sd7xeiezPVzK1c1mK2iYk0Soz5Kqi4XNSGUb.png?width=320&crop=smart&format=pjpg&auto=webp&s=16bfa24a5afdfe6a555cf53ffeff63a5ed8e912b', 'width': 320}, {'height': 1137, 'url': 'https://external-preview.redd.it/aWY2NWhlOHFncWZlMW6iYpr5Sd7xeiezPVzK1c1mK2iYk0Soz5Kqi4XNSGUb.png?width=640&crop=smart&format=pjpg&auto=webp&s=885f477050ee9d2c921df22d7a5fc77a27c19b8d', 'width': 640}, {'height': 1706, 'url': 'https://external-preview.redd.it/aWY2NWhlOHFncWZlMW6iYpr5Sd7xeiezPVzK1c1mK2iYk0Soz5Kqi4XNSGUb.png?width=960&crop=smart&format=pjpg&auto=webp&s=2efaad7f50430ea92f6fb29c500ccd081ef99c04', 'width': 960}, {'height': 1920, 'url': 'https://external-preview.redd.it/aWY2NWhlOHFncWZlMW6iYpr5Sd7xeiezPVzK1c1mK2iYk0Soz5Kqi4XNSGUb.png?width=1080&crop=smart&format=pjpg&auto=webp&s=53815acf39cdf309d71a0c5539d9ac5248d12e66', 'width': 1080}], 'source': {'height': 1920, 'url': 'https://external-preview.redd.it/aWY2NWhlOHFncWZlMW6iYpr5Sd7xeiezPVzK1c1mK2iYk0Soz5Kqi4XNSGUb.png?format=pjpg&auto=webp&s=d6f37eafec59804b492a59e631b57bf39fe5fb8e', 'width': 1080}, 'variants': {}}]} |
|
How to cancel generation with llama.cpp (llama-server)? | 10 | Their docs state that it is OpenAI API compatible, so I'm using the that api to stream its output, but I haven't found any way to cancel generation. If a model starts repeating itself it just keeps going until it hits max tokens to generate (or if I shut down the server).
None of the [command-line arguments](https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md) mention anything about this either (beyond stop tokens).
It feels like I have missed something, because telling the server to stop running inference should be as simple as telling it to do it.
Is it simply not that can be done with llama.cpp, or is it dependent on what API you use with it? | 2025-01-28T13:06:23 | https://www.reddit.com/r/LocalLLaMA/comments/1ic145f/how_to_cancel_generation_with_llamacpp_llamaserver/ | rabitmeerkitkat | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic145f | false | null | t3_1ic145f | /r/LocalLLaMA/comments/1ic145f/how_to_cancel_generation_with_llamacpp_llamaserver/ | false | false | self | 10 | {'enabled': False, 'images': [{'id': 'DJPqvteONpGwVVw6LzaG6b8vlDa2rv2hETCaqe0z57s', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/wyCM1fHzTa-IIqHgS1QTxdSYNXn668elDj0WmYMPf_k.jpg?width=108&crop=smart&auto=webp&s=d6fa197328d583bcae7a764b40fd1214265b6852', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/wyCM1fHzTa-IIqHgS1QTxdSYNXn668elDj0WmYMPf_k.jpg?width=216&crop=smart&auto=webp&s=dd615bfe0453b06d53bc1f5f17fc3f6ad926694f', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/wyCM1fHzTa-IIqHgS1QTxdSYNXn668elDj0WmYMPf_k.jpg?width=320&crop=smart&auto=webp&s=0bc6ac2e1db55ec07cc6a17178ea52bf436f9bce', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/wyCM1fHzTa-IIqHgS1QTxdSYNXn668elDj0WmYMPf_k.jpg?width=640&crop=smart&auto=webp&s=b0d58c9a49c1e9ce629e5b31dce17b727d8c6ab8', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/wyCM1fHzTa-IIqHgS1QTxdSYNXn668elDj0WmYMPf_k.jpg?width=960&crop=smart&auto=webp&s=7c835cb0600a4d280a57f12d0bc008ef12acd26d', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/wyCM1fHzTa-IIqHgS1QTxdSYNXn668elDj0WmYMPf_k.jpg?width=1080&crop=smart&auto=webp&s=1f2580bd36b3bf3b766d205ac6d737a9d8d34c2a', 'width': 1080}], 'source': {'height': 640, 'url': 'https://external-preview.redd.it/wyCM1fHzTa-IIqHgS1QTxdSYNXn668elDj0WmYMPf_k.jpg?auto=webp&s=d8b103bed805ceb641b2ff49dc8c7403318263b1', 'width': 1280}, 'variants': {}}]} |
What's the Chip angle to DeepSeek? | 1 | 2025-01-28T13:13:10 | https://chipbriefing.substack.com/p/daily-deepseek-and-chips-cxmt-and | Big-Spinach-4146 | chipbriefing.substack.com | 1970-01-01T00:00:00 | 0 | {} | 1ic18re | false | null | t3_1ic18re | /r/LocalLLaMA/comments/1ic18re/whats_the_chip_angle_to_deepseek/ | false | false | 1 | {'enabled': False, 'images': [{'id': 'RRZDwrvA9gRg86NlJaM1i2DlLalmaLrtNkPUNwwioyg', 'resolutions': [{'height': 90, 'url': 'https://external-preview.redd.it/GMor2ckjyQvCZnQKN_U_LvkW1iSwcEb0KXCjIhlH4g8.jpg?width=108&crop=smart&auto=webp&s=d3179362bafaa2c67b07a4bcbde0091fc1e0d8e5', 'width': 108}, {'height': 180, 'url': 'https://external-preview.redd.it/GMor2ckjyQvCZnQKN_U_LvkW1iSwcEb0KXCjIhlH4g8.jpg?width=216&crop=smart&auto=webp&s=1d66d8191e6e6ee893a8430fc15fbc1765db9544', 'width': 216}, {'height': 266, 'url': 'https://external-preview.redd.it/GMor2ckjyQvCZnQKN_U_LvkW1iSwcEb0KXCjIhlH4g8.jpg?width=320&crop=smart&auto=webp&s=28e9952539e658695c5b72db0cad4bfac9171673', 'width': 320}, {'height': 533, 'url': 'https://external-preview.redd.it/GMor2ckjyQvCZnQKN_U_LvkW1iSwcEb0KXCjIhlH4g8.jpg?width=640&crop=smart&auto=webp&s=2e483e3d648cac3c95fb34f7a9006f6117231620', 'width': 640}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/GMor2ckjyQvCZnQKN_U_LvkW1iSwcEb0KXCjIhlH4g8.jpg?auto=webp&s=20fdbb4249873125dbb0930e4d93cfa795b21c2f', 'width': 720}, 'variants': {}}]} |
||
What is the difference between Deepseek's R1, V3, R1 FW, V3-T, etc? | 4 | I'm a complete noob, so from a user standpoint, what are the noticeable differences? is there an overall best model? or do each have strengths and weaknesses? | 2025-01-28T13:14:58 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1a17/what_is_the_difference_between_deepseeks_r1_v3_r1/ | hjras | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1a17 | false | null | t3_1ic1a17 | /r/LocalLLaMA/comments/1ic1a17/what_is_the_difference_between_deepseeks_r1_v3_r1/ | false | false | self | 4 | null |
DeepSeek's founder Liang Wenfeng attended a meeting with Chinese Premier Li Qiang. Jan 20, 2025 | 20 | 2025-01-28T13:15:55 | https://www.youtube.com/watch?v=PqPAhR5qTxQ | Dr_Me_123 | youtube.com | 1970-01-01T00:00:00 | 0 | {} | 1ic1ao9 | false | {'oembed': {'author_name': 'Reuters', 'author_url': 'https://www.youtube.com/@Reuters', 'height': 200, 'html': '<iframe width="356" height="200" src="https://www.youtube.com/embed/PqPAhR5qTxQ?feature=oembed&enablejsapi=1" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen title="DeepSeek founder meets Chinese Premier | REUTERS"></iframe>', 'provider_name': 'YouTube', 'provider_url': 'https://www.youtube.com/', 'thumbnail_height': 360, 'thumbnail_url': 'https://i.ytimg.com/vi/PqPAhR5qTxQ/hqdefault.jpg', 'thumbnail_width': 480, 'title': 'DeepSeek founder meets Chinese Premier | REUTERS', 'type': 'video', 'version': '1.0', 'width': 356}, 'type': 'youtube.com'} | t3_1ic1ao9 | /r/LocalLLaMA/comments/1ic1ao9/deepseeks_founder_liang_wenfeng_attended_a/ | false | false | 20 | {'enabled': False, 'images': [{'id': 'uufxmofFLzs_MsII_eFifUe0HXNOekRZf0vOOpj8Sus', 'resolutions': [{'height': 81, 'url': 'https://external-preview.redd.it/m_6JjG_bskCkF0M8m7neVjAAlkvmGvVBNSnPGbPR0aY.jpg?width=108&crop=smart&auto=webp&s=adae11e3cd00ce4227a57d2d18672c45a09fe668', 'width': 108}, {'height': 162, 'url': 'https://external-preview.redd.it/m_6JjG_bskCkF0M8m7neVjAAlkvmGvVBNSnPGbPR0aY.jpg?width=216&crop=smart&auto=webp&s=94582d684cbc9fb6dd636d64b3740b2aef7f55e8', 'width': 216}, {'height': 240, 'url': 'https://external-preview.redd.it/m_6JjG_bskCkF0M8m7neVjAAlkvmGvVBNSnPGbPR0aY.jpg?width=320&crop=smart&auto=webp&s=71792d4336e574a37dae1a05a88816f3289a7388', 'width': 320}], 'source': {'height': 360, 'url': 'https://external-preview.redd.it/m_6JjG_bskCkF0M8m7neVjAAlkvmGvVBNSnPGbPR0aY.jpg?auto=webp&s=559d219620692452831f0c406b0a16f878b7e175', 'width': 480}, 'variants': {}}]} |
||
DeepSeek in Ollama console: Any way to make it look less ugly? | 1 | Hey folks! Total beginner at this stuff. I've been playing around with DeepSeek through Ollama console on Windows, and man, the text formatting is pretty basic in the command prompt. Any way to make it look nicer? Or do I have to use deepseek.com's online chat for the fancy formatting?
Would love some tips - especially if they're newbie-friendly! Running this on Windows command prompt btw. | 2025-01-28T13:17:23 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1bpe/deepseek_in_ollama_console_any_way_to_make_it/ | UltraAntiqueEvidence | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1bpe | false | null | t3_1ic1bpe | /r/LocalLLaMA/comments/1ic1bpe/deepseek_in_ollama_console_any_way_to_make_it/ | false | false | self | 1 | null |
Dividing a large LLM model into multiple GPUs using vllm. | 3 | I have a llama fine tuned model thats too big to sit on my rtx gpu. I can fit it on two gpus but im getting the following error using. does anyone know how to define the spawn as mentioned in the error log below?
tensor_parallel_size=2
>
RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method
>ERROR 01-27 22:46:48 multiproc\_worker\_utils.py:120\] Worker VllmWorkerProcess pid 168469 died, exit code: 1
>INFO 01-27 22:46:48 multiproc\_worker\_utils.py:123\] Killing local vLLM worker processes
>Traceback (most recent call last):
>
| 2025-01-28T13:18:40 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1clo/dividing_a_large_llm_model_into_multiple_gpus/ | Lazy_Wedding_1383 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1clo | false | null | t3_1ic1clo | /r/LocalLLaMA/comments/1ic1clo/dividing_a_large_llm_model_into_multiple_gpus/ | false | false | self | 3 | null |
Does deepseek have an advanced voice mode like ChatGPT, that sounds natural and empathetic? | 0 | Thanks | 2025-01-28T13:23:15 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1fxp/does_deepseek_have_an_advanced_voice_mode_like/ | SirCheeseAlot | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1fxp | false | null | t3_1ic1fxp | /r/LocalLLaMA/comments/1ic1fxp/does_deepseek_have_an_advanced_voice_mode_like/ | false | false | self | 0 | null |
What if releasing R1 is the trading strategy?! | 1 | [removed] | 2025-01-28T13:26:37 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1id2/what_if_releasing_r1_is_the_trading_strategy/ | Dull_Art6802 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1id2 | false | null | t3_1ic1id2 | /r/LocalLLaMA/comments/1ic1id2/what_if_releasing_r1_is_the_trading_strategy/ | false | false | self | 1 | null |
Need guidance in making a sales assistant chatbot | 0 | I am a Software Developer in an automotive industry which mostly sells trucks and buses, and recently I pitched an idea to my boss which now I'm having difficulty in actualising.
Problem Statement:
Whenever customer requests a vehicle, they do so by stating their requirements rather than a specific model. (Example - I want to truck to carry the coal from a coalmine...(Something like this)) After which the sales executives has to suggest some models off the top of his head.
Problem Description:-
1. They might forget to mention a few models, as they can't remember all the models all the time .
2. They may forget to mention some key features.
3. In some cases, where they are unsure about what to suggest, they call their higher ups who most of the times aren't able to pick the calls since they maybe with other customers etc.
4. ADD MORE TO THIS IF YOU CAN SO THAT I CAN CONVINCE THE SALES TEAM.
After this, they search for the brochures in their system of those specific vehicles and then show it to the customers which takes a few minutes.
Solution :-
1. No need to remember and recall.
2. They don't have to search for the brochures.
3. No chances of missing the information.
NOW, I NEED YOUR HELP IN THIS:-
1. I have the brochures as the data which i have scraped from the internet since it was more organised there. They even have application wise vehicles brochures which is exactly what I need.
I have thought of the following way to go about implementing this which I think is inefficient:-
Perform basic RAG on the brochures and use LLM to get the relevant information for the query.
BUT i want to make this more efficient as basic RAG won't give me accurate results as the brochures have a but complex design.
I want to implement Knowledge graph here too as that's something I feel can be used here and my boss becomes very happy when I talk about it.
SUGGEST ME A BETTER WAY TO DO THIS. USING KNOWLEDGE GRAPHS AND SHIT IF POSSIBLE.
| 2025-01-28T13:28:30 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1jnd/need_guidance_in_making_a_sales_assistant_chatbot/ | Sick__sock | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1jnd | false | null | t3_1ic1jnd | /r/LocalLLaMA/comments/1ic1jnd/need_guidance_in_making_a_sales_assistant_chatbot/ | false | false | self | 0 | null |
How to to make Deepseek R1 response to be correct? | 1 | I have noticed that Deepseek R1 Distil 7B (also similar issue with Qwen models) have a following problem:
For example some sentences in Serbian don't make sense at all. However, the more puzzling issue is mixing of Cyrillic and Latin letters. So for example it will give you answer but mix two alphabets while giving it.
I have also noticed that for some reason it can start thinking in Russian while giving answer in half English half Serbian.
Would adding some embedding model fix such issues? Thanks! | 2025-01-28T13:33:52 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1nj7/how_to_to_make_deepseek_r1_response_to_be_correct/ | Dazz9 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1nj7 | false | null | t3_1ic1nj7 | /r/LocalLLaMA/comments/1ic1nj7/how_to_to_make_deepseek_r1_response_to_be_correct/ | false | false | self | 1 | null |
360 announced that it will provide security services to DeepSeek free of charge | 38 | English :About the initiative to fully support DeepSeek large models
Dear Netizens:
On the occasion of the Chinese New Year, I would like to extend my most sincere wishes to you all!
Recently, my country's independently developed large-scale model DeepSeek has emerged as a dark horse, creating a "Chinese whirlwind" in the global AI field.
As the largest network security company in China, 360 Group solemnly promises:
1. We will take national justice as our own responsibility, go all out to provide comprehensive network security protection, and resolutely safeguard the dignity and security of domestic artificial intelligence technology.
2. Nano AI Search has specially launched the "DeepSeek Dedicated Line" service, using the highest-specification R1 high-speed dedicated line and a dedicated anti-attack data center to ensure a safe experience for every user.
3. Join forces with more than ten top domestic artificial intelligence teams such as Doubao and Kimi to build the strongest technology alliance and jointly create an autonomous and controllable smart ecosystem.
This is a historic moment for China's artificial intelligence, and it is also a great cause in which every netizen who loves the motherland can participate.
Let’s cheer for DeepSeek together!
Nano AI Search
January 28, 2025
https://preview.redd.it/sb6lch8lmqfe1.jpg?width=343&format=pjpg&auto=webp&s=34ec0146bb360dd0ac270df641d769a92d7ecf53
| 2025-01-28T13:37:29 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1q00/360_announced_that_it_will_provide_security/ | External_Mood4719 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1q00 | false | null | t3_1ic1q00 | /r/LocalLLaMA/comments/1ic1q00/360_announced_that_it_will_provide_security/ | false | false | 38 | null |
|
Seeking Guidance on using AI to modernize and understand legacy + uncommon language codebase | 1 | [removed] | 2025-01-28T13:40:00 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1rqh/seeking_guidance_on_using_ai_to_modernize_and/ | cn5171bn | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1rqh | false | null | t3_1ic1rqh | /r/LocalLLaMA/comments/1ic1rqh/seeking_guidance_on_using_ai_to_modernize_and/ | false | false | self | 1 | null |
Need help with using AI to understand and modernize existing legacy codebase that's also in an uncommon programming language/database. | 1 | [removed] | 2025-01-28T13:43:11 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1twf/need_help_with_using_ai_to_understand_and/ | cn5171bn | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1twf | false | null | t3_1ic1twf | /r/LocalLLaMA/comments/1ic1twf/need_help_with_using_ai_to_understand_and/ | false | false | self | 1 | null |
What is your favorite model for language tasks? | 2 | I'm using Qwen2.5:72b-instruct, and it is wonderful, but really slow. I want a smaller one that can help with improving language, grammar check, etc. | 2025-01-28T13:43:43 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1uaz/what_is_your_favorite_model_for_language_tasks/ | m_abdelfattah | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1uaz | false | null | t3_1ic1uaz | /r/LocalLLaMA/comments/1ic1uaz/what_is_your_favorite_model_for_language_tasks/ | false | false | self | 2 | null |
Alternative Chat UI for running Deepseek R1 with Openrouter API | 4 | I'm currently trying to use Deepseek R1 with openrouter API as the free Deepseek R1 site is always overloaded.
I'm having trouble getting it to work right in any alternative UI and Openrouter's native chat UI just sucks. Has anyone gotten Deepseek R1 running on a good UI application that can run it through openrouter API? | 2025-01-28T13:45:58 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1vw0/alternative_chat_ui_for_running_deepseek_r1_with/ | delijoe | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1vw0 | false | null | t3_1ic1vw0 | /r/LocalLLaMA/comments/1ic1vw0/alternative_chat_ui_for_running_deepseek_r1_with/ | false | false | self | 4 | null |
Does LTM using Vector DBs work better in a format similar to the training data? | 1 | I would like to have a chat bot with proper long term memory.
And if i understand correctly, the best way to do this is to have a vector database and store all messages in the database and provide a history of the 10 closest messages in the prompt to build a context.
So when creating the embeddings for words/sentences they follow the schemes/format of the training data, right?
So when 90% of the training data were roman books, the tokens would be structured based on those data right?
So if thats correct, would it make sense to first bring the chats/history in a format that is more like the training data? So instead of:
Peter: How are you?
Dan: Im okay. Just a bit sad.
I would store something like
Peter asked Dan how he is feeling and dan told him hes a bit sad.
Does this make sense? | 2025-01-28T13:48:27 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1xlk/does_ltm_using_vector_dbs_work_better_in_a_format/ | freehuntx | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1xlk | false | null | t3_1ic1xlk | /r/LocalLLaMA/comments/1ic1xlk/does_ltm_using_vector_dbs_work_better_in_a_format/ | false | false | self | 1 | null |
Anybody please help want to make a Local LLM for skincare formulation | 0 | If I may ask this question ? Does anyone have any advice on me building a local LLM with the sole intent and purpose of creating complicated skincare formulas ?? I am using at the moment an Apple M1 Mac Mini with 16Gb RAM but am going to either purchase an M4 Mac Mini (Hopefully Pro version) or adding a Nvidia RTX 4060Ti, Any help would be appreciated greatly thank you | 2025-01-28T13:49:13 | https://www.reddit.com/r/LocalLLaMA/comments/1ic1y65/anybody_please_help_want_to_make_a_local_llm_for/ | mitchfromtoronto | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic1y65 | false | null | t3_1ic1y65 | /r/LocalLLaMA/comments/1ic1y65/anybody_please_help_want_to_make_a_local_llm_for/ | false | false | self | 0 | null |
Why is deepseek getting so much hype? | 0 | So I'm just wondering what the hype is about? Like what makes you excited about deepseek compared to other models?
Deepest is cool like it's a good model but the reasoning (COT) tech isn't new. I get it's hyped because it's open sourced and chinese but like to me we were gonna get the next generation of llms with cot from llama or mistral anyway?
Imo i think Google titan research paper is a bigger leap in the field then a cot model that is censored and open source. That was gonna happen by other company's soon anyway.
Titans is like training the model on inference which imo is a wayyy cooler then a llm with no image generation or audio. But that's why I'm asking so I can see a different perspective.
| 2025-01-28T13:52:34 | https://www.reddit.com/r/LocalLLaMA/comments/1ic20hz/why_is_deepseek_getting_so_much_hype/ | rhaastt-ai | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic20hz | false | null | t3_1ic20hz | /r/LocalLLaMA/comments/1ic20hz/why_is_deepseek_getting_so_much_hype/ | false | false | self | 0 | null |
Now you can use Deepseek with MCP tools | 13 | Hi guys! For a while, we've been developing an open-source agent framework with client-server architecture and MCP support. We launched it last week. After Deepseek's incredible rise, we added Deepseek support even before Llama. I never expected this - if someone had told me we would integrate a local model before Llama, I wouldn't have believed them.
Currently, we've only added support for "DeepSeek chat" since "DeepseekR1" doesn't have function-calling capability yet. I was actually surprised that Deepseek R1 doesn't have function-calling, but since it's a newly launched model, it will likely enhance its function capabilities soon.
You can check out the repo here: [https://github.com/Upsonic/Upsonic](https://github.com/Upsonic/Upsonic)
| 2025-01-28T13:58:04 | https://www.reddit.com/r/LocalLLaMA/comments/1ic24e0/now_you_can_use_deepseek_with_mcp_tools/ | mbartu | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic24e0 | false | null | t3_1ic24e0 | /r/LocalLLaMA/comments/1ic24e0/now_you_can_use_deepseek_with_mcp_tools/ | false | false | self | 13 | {'enabled': False, 'images': [{'id': 'eY8KqeX1yXTT7A0BvNIdOocExE_dNgJhVGiUaIypCAE', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/g-9fRNzUMISeYuLncscYvxodQCEiEOE2tvE5zZ3f9fk.jpg?width=108&crop=smart&auto=webp&s=edc9ebb17447d452fcd72f6e55285d748a7a705d', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/g-9fRNzUMISeYuLncscYvxodQCEiEOE2tvE5zZ3f9fk.jpg?width=216&crop=smart&auto=webp&s=754212ceafdf4c2a119c9d25e432c8b208dc5528', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/g-9fRNzUMISeYuLncscYvxodQCEiEOE2tvE5zZ3f9fk.jpg?width=320&crop=smart&auto=webp&s=f44832dd58a203c58bdf3be55a4bc6d68845ee50', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/g-9fRNzUMISeYuLncscYvxodQCEiEOE2tvE5zZ3f9fk.jpg?width=640&crop=smart&auto=webp&s=9c4e26cc6bb86ae6da27765dd838917d2659ec40', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/g-9fRNzUMISeYuLncscYvxodQCEiEOE2tvE5zZ3f9fk.jpg?width=960&crop=smart&auto=webp&s=d0b2e5858607d308c0d76412142011938a15e1cf', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/g-9fRNzUMISeYuLncscYvxodQCEiEOE2tvE5zZ3f9fk.jpg?width=1080&crop=smart&auto=webp&s=ffe6c854cab88cd3eaa64e3bc265e6deae2d328b', 'width': 1080}], 'source': {'height': 640, 'url': 'https://external-preview.redd.it/g-9fRNzUMISeYuLncscYvxodQCEiEOE2tvE5zZ3f9fk.jpg?auto=webp&s=f7565083946d26e9ca980039d92c6e580787bbf8', 'width': 1280}, 'variants': {}}]} |
Fine tunning a LLM - searching for a specialized one or using llama3.2? | 1 | [removed] | 2025-01-28T13:59:12 | https://www.reddit.com/r/LocalLLaMA/comments/1ic255n/fine_tunning_a_llm_searching_for_a_specialized/ | After_Ad_6881 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic255n | false | null | t3_1ic255n | /r/LocalLLaMA/comments/1ic255n/fine_tunning_a_llm_searching_for_a_specialized/ | false | false | self | 1 | null |
DeepSeek wrote me a script. About itself. | 1 | [removed] | 2025-01-28T14:01:55 | https://www.reddit.com/r/LocalLLaMA/comments/1ic27bu/deepseek_wrote_me_a_script_about_itself/ | Kill_Connolly | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic27bu | false | null | t3_1ic27bu | /r/LocalLLaMA/comments/1ic27bu/deepseek_wrote_me_a_script_about_itself/ | false | false | self | 1 | null |
Unsloth made dynamic R1 quants - can be run on as little as 80gb of RAM | 160 |
This is super cool: https://unsloth.ai/blog/deepseekr1-dynamic
Key points:
- they didn’t naively quantized everything - some layers needed more bits to overcome issues
- they have a range of quants from 1.58bit to 2.51bit which shrink the model to 131gb-212gb
- they say the smallest can be run with as little as 80gb RAM (but full model in RAM or VRAM obviously faster)
Might be real option for local R1! | 2025-01-28T14:04:50 | https://www.reddit.com/r/LocalLLaMA/comments/1ic29lq/unsloth_made_dynamic_r1_quants_can_be_run_on_as/ | davernow | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic29lq | false | null | t3_1ic29lq | /r/LocalLLaMA/comments/1ic29lq/unsloth_made_dynamic_r1_quants_can_be_run_on_as/ | false | false | self | 160 | null |
Expandable rig | 2 | Hi, I’m currently thinking of offloading some of the less demanding tasks (corrections, text improvement, tagging) from cpu to an actual GPU rig. Most of it runs well with gemma2:9b or mistral-nemo.
My plan is to make it expandable. Start with one GPU, then work my way up.
What would be the best way to achieve that?
Is there some kind of terminology for the set up I’m looking for?
Or am I completely on the wrong track, and Mac Silicone is the better way to go with my use case?
I’m lost in the possibilities, and I would love this communities input and experiences. | 2025-01-28T14:09:40 | https://www.reddit.com/r/LocalLLaMA/comments/1ic2d82/expandable_rig/ | WolpertingerRumo | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic2d82 | false | null | t3_1ic2d82 | /r/LocalLLaMA/comments/1ic2d82/expandable_rig/ | false | false | self | 2 | null |
OpenAI announces ChatGPT Gov | 0 | 2025-01-28T14:12:03 | eternviking | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ic2ey5 | false | null | t3_1ic2ey5 | /r/LocalLLaMA/comments/1ic2ey5/openai_announces_chatgpt_gov/ | false | false | 0 | {'enabled': True, 'images': [{'id': 'VOh3hOFjEXg5NpSsHP9tWJszPa7BV4_szqxw_UzZD-M', 'resolutions': [{'height': 61, 'url': 'https://preview.redd.it/p4xm3cevsqfe1.png?width=108&crop=smart&auto=webp&s=ae9a17b93fc26e2bc64de6bdd2f7781c08f2b7b5', 'width': 108}, {'height': 123, 'url': 'https://preview.redd.it/p4xm3cevsqfe1.png?width=216&crop=smart&auto=webp&s=7aac32b96801b3f640556773532c8969506beacd', 'width': 216}, {'height': 182, 'url': 'https://preview.redd.it/p4xm3cevsqfe1.png?width=320&crop=smart&auto=webp&s=6f86d53bf4880afdabe4dbda06709043a8ec373d', 'width': 320}, {'height': 364, 'url': 'https://preview.redd.it/p4xm3cevsqfe1.png?width=640&crop=smart&auto=webp&s=c2f8182567f3a1bcf625cf7349fdf7d8d027cb0d', 'width': 640}, {'height': 547, 'url': 'https://preview.redd.it/p4xm3cevsqfe1.png?width=960&crop=smart&auto=webp&s=0889b7036e035ceb2ae768442fa39c780770ed5d', 'width': 960}], 'source': {'height': 612, 'url': 'https://preview.redd.it/p4xm3cevsqfe1.png?auto=webp&s=de49f41ff26d6f8c4fbfe627e2703dad16661694', 'width': 1074}, 'variants': {}}]} |
|||
How do I download Deepseek R1 Model on a USB-Stick? I don’t wan‘t the explanation how to run it locally | 1 | [removed] | 2025-01-28T14:17:14 | https://www.reddit.com/r/LocalLLaMA/comments/1ic2iv9/how_do_i_download_deepseek_r1_model_on_a_usbstick/ | Theasianboy_ | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic2iv9 | false | null | t3_1ic2iv9 | /r/LocalLLaMA/comments/1ic2iv9/how_do_i_download_deepseek_r1_model_on_a_usbstick/ | false | false | self | 1 | null |
Vllm V1 vs LMdeploy benchmarks | 3 | We need this. VLLM V1 claims improved performance. It may have closed or surpassed the gap. Upvote so someone with resources and time can benchmark it. | 2025-01-28T14:22:31 | https://www.reddit.com/r/LocalLLaMA/comments/1ic2mso/vllm_v1_vs_lmdeploy_benchmarks/ | Wonderful_Alfalfa115 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic2mso | false | null | t3_1ic2mso | /r/LocalLLaMA/comments/1ic2mso/vllm_v1_vs_lmdeploy_benchmarks/ | false | false | self | 3 | null |
Questions about local models | 1 | [removed] | 2025-01-28T14:26:05 | https://www.reddit.com/r/LocalLLaMA/comments/1ic2pep/questions_about_local_models/ | infopcgood | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic2pep | false | null | t3_1ic2pep | /r/LocalLLaMA/comments/1ic2pep/questions_about_local_models/ | false | false | self | 1 | {'enabled': False, 'images': [{'id': 'tMwUF5c65bzEgJYqbM1Fw6MCe2gnXm3lEjOu-gPTSFE', 'resolutions': [{'height': 65, 'url': 'https://external-preview.redd.it/T3dwFxBeTh14I0gRGwTgWSW_ZOjtsyWdThn4yOyUB0o.jpg?width=108&crop=smart&auto=webp&s=8444a8513e12a8d51885299e960245e1fe8083f4', 'width': 108}, {'height': 130, 'url': 'https://external-preview.redd.it/T3dwFxBeTh14I0gRGwTgWSW_ZOjtsyWdThn4yOyUB0o.jpg?width=216&crop=smart&auto=webp&s=f54812ff60ad29bffc59f98f8490deee2529892c', 'width': 216}, {'height': 192, 'url': 'https://external-preview.redd.it/T3dwFxBeTh14I0gRGwTgWSW_ZOjtsyWdThn4yOyUB0o.jpg?width=320&crop=smart&auto=webp&s=e2013b68f1f7c9da64a9231516dc95110cdbc9fd', 'width': 320}, {'height': 385, 'url': 'https://external-preview.redd.it/T3dwFxBeTh14I0gRGwTgWSW_ZOjtsyWdThn4yOyUB0o.jpg?width=640&crop=smart&auto=webp&s=b3e486673640592fed5e6e7beea3c8482611b687', 'width': 640}, {'height': 577, 'url': 'https://external-preview.redd.it/T3dwFxBeTh14I0gRGwTgWSW_ZOjtsyWdThn4yOyUB0o.jpg?width=960&crop=smart&auto=webp&s=cf84bd74d5cb499cb76e9aa9cd87c54fc7c2c4ad', 'width': 960}, {'height': 650, 'url': 'https://external-preview.redd.it/T3dwFxBeTh14I0gRGwTgWSW_ZOjtsyWdThn4yOyUB0o.jpg?width=1080&crop=smart&auto=webp&s=cbb49d75cc73cb3d66d78acd4e48056b12b5e08a', 'width': 1080}], 'source': {'height': 720, 'url': 'https://external-preview.redd.it/T3dwFxBeTh14I0gRGwTgWSW_ZOjtsyWdThn4yOyUB0o.jpg?auto=webp&s=f3f58376c18336c911a908696438d13bf2919126', 'width': 1196}, 'variants': {}}]} |
I built a macOS-use - an AI agent that can interact with every app on you MacBook! would love your feedback! | 1 | [removed] | 2025-01-28T14:26:10 | https://www.reddit.com/r/LocalLLaMA/comments/1ic2pgq/i_built_a_macosuse_an_ai_agent_that_can_interact/ | PureMaximum0 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic2pgq | false | null | t3_1ic2pgq | /r/LocalLLaMA/comments/1ic2pgq/i_built_a_macosuse_an_ai_agent_that_can_interact/ | false | false | 1 | null |
|
I built a macOS-use - an AI agent that can interact with every app on you MacBook! would love your feedback! | 1 | [removed] | 2025-01-28T14:28:18 | https://www.reddit.com/r/LocalLLaMA/comments/1ic2r24/i_built_a_macosuse_an_ai_agent_that_can_interact/ | PureMaximum0 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic2r24 | false | null | t3_1ic2r24 | /r/LocalLLaMA/comments/1ic2r24/i_built_a_macosuse_an_ai_agent_that_can_interact/ | false | false | 1 | {'enabled': False, 'images': [{'id': 'iCdDF4Wals2-M9gcDQHOwZbOoRJ-7ieHT2ShALeuOwA', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/FZ9_HdBevPtls6FSxKzIwDx_9-Jd7xQzv6i8qnnRXfU.jpg?width=108&crop=smart&auto=webp&s=51bc7e52e2bc3e255ebb91348f11a3a3bddc5740', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/FZ9_HdBevPtls6FSxKzIwDx_9-Jd7xQzv6i8qnnRXfU.jpg?width=216&crop=smart&auto=webp&s=3cef2550d4b82fdf94b0eacc3abc87f49214737a', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/FZ9_HdBevPtls6FSxKzIwDx_9-Jd7xQzv6i8qnnRXfU.jpg?width=320&crop=smart&auto=webp&s=eb90712fcfb66159ef1d738324f44819fdf2a7be', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/FZ9_HdBevPtls6FSxKzIwDx_9-Jd7xQzv6i8qnnRXfU.jpg?width=640&crop=smart&auto=webp&s=9b2b5a6f859aba9c924a151f5453b39596cff62b', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/FZ9_HdBevPtls6FSxKzIwDx_9-Jd7xQzv6i8qnnRXfU.jpg?width=960&crop=smart&auto=webp&s=b7592aa6c0a4532334f876d9a212dc0638b64c67', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/FZ9_HdBevPtls6FSxKzIwDx_9-Jd7xQzv6i8qnnRXfU.jpg?width=1080&crop=smart&auto=webp&s=3089fb1dc2c48bd2d23ea0bae940738a9772128d', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/FZ9_HdBevPtls6FSxKzIwDx_9-Jd7xQzv6i8qnnRXfU.jpg?auto=webp&s=d6f807fa4791c01968e16e89f488e7ed1cadfce3', 'width': 1200}, 'variants': {}}]} |
|
Just found this insane Mac automation tool that lets you control your entire MacBook with plain English | 1 | [removed] | 2025-01-28T14:31:20 | [deleted] | 1970-01-01T00:00:00 | 0 | {} | 1ic2tfj | false | null | t3_1ic2tfj | /r/LocalLLaMA/comments/1ic2tfj/just_found_this_insane_mac_automation_tool_that/ | false | false | default | 1 | null |
||
I have installed DeepSeek R1 14B using Ollama and OpenWebUI and Docker in my PC but I have two issues that maybe you can help me with. | 0 | First of all my specs are a 4090 with 32G of DDR4 RAM, the first issue I am having (I don't know if it is normal or not) is that the model is being loaded in to my RAM memory instead of my GPU, should I change something to load it in my GPU instead? second issue is that I have no idea about how to unload the model from the RAM, I ended up restarting my PC.
Thank you all in advance. | 2025-01-28T14:35:36 | https://www.reddit.com/r/LocalLLaMA/comments/1ic2wpq/i_have_installed_deepseek_r1_14b_using_ollama_and/ | fernandollb | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic2wpq | false | null | t3_1ic2wpq | /r/LocalLLaMA/comments/1ic2wpq/i_have_installed_deepseek_r1_14b_using_ollama_and/ | false | false | self | 0 | null |
Deepseek became way too popular! | 0 | Only a few weeks ago, Deepseek was this kind of nerdy thing that only a few people knew about, and now, all of a sudden, it's everywhere.
I don't know why, but I feel proud for some reason—it's like watching your boy finally lock in.
https://preview.redd.it/s3dv1zt6xqfe1.png?width=452&format=png&auto=webp&s=b2def05b265fe3aba52adca525421195affe4d8d
| 2025-01-28T14:36:10 | https://www.reddit.com/r/LocalLLaMA/comments/1ic2x4v/deepseek_became_way_too_popular/ | vinam_7 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic2x4v | false | null | t3_1ic2x4v | /r/LocalLLaMA/comments/1ic2x4v/deepseek_became_way_too_popular/ | false | false | 0 | null |
|
Should I make a embedded search saas? | 1 | Hi!
I'm considering building an embedded search API that allows you to upload your data through an API or upload files directly and then start searching.
Before I start working on this, I want to know if there is a real need for such a solution or if the current search tools available in the market already meet your requirements.
* Do you think an embedded search API would improve your development workflow?
* Are there any specific features you would like to see in a search API?
* Do you spend a lot of time setting it up?
Feel free to add anything, I would love to hear what you have to say or just tell me about your experince:):) | 2025-01-28T14:38:14 | https://www.reddit.com/r/LocalLLaMA/comments/1ic2yqd/should_i_make_a_embedded_search_saas/ | Practical-Rub-1190 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic2yqd | false | null | t3_1ic2yqd | /r/LocalLLaMA/comments/1ic2yqd/should_i_make_a_embedded_search_saas/ | false | false | self | 1 | null |
Censorship in Deepseek | 1 | [removed] | 2025-01-28T14:39:59 | https://www.reddit.com/r/LocalLLaMA/comments/1ic301a/censorship_in_deepseek/ | Thin_Candidate_582 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic301a | false | null | t3_1ic301a | /r/LocalLLaMA/comments/1ic301a/censorship_in_deepseek/ | false | false | 1 | null |
|
Not understanding the Deepseek fascination? | 0 | The quality of the model aside. Most of the stock market and news outlets are screaming at the top of their lungs about how deepseek built this great model with about5.5-6 million USD. which is of course complete BS.
\`Per [36Kr](https://en.wikipedia.org/wiki/36Kr), Liang had built up a store of 10,000 [Nvidia A100](https://en.wikipedia.org/wiki/Ampere_(microarchitecture)#A100_accelerator_and_DGX_A100) GPUs before the U.S. government imposed AI chip restrictions on China.[^(\[16\])](https://en.wikipedia.org/wiki/DeepSeek#cite_note-:0-16) Some estimates put the number as high as 50,000.[^(\[15\])](https://en.wikipedia.org/wiki/DeepSeek#cite_note-:6-15)\`
someone explain how 10,000 - 50,000 A100's cost only $6 mil USD.
Technically Liang bought these before breaking off his lab into its own company "deepseek". Im really starting to feel what they mean and people will soon find out is the hosting, electricity, etc cost $6million.
Again thus has nothing to do with the quality or performance, im simply implying that the market has misunderstood what actually has gone into this model. | 2025-01-28T14:46:27 | https://www.reddit.com/r/LocalLLaMA/comments/1ic351q/not_understanding_the_deepseek_fascination/ | ThisParticular7389 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic351q | false | null | t3_1ic351q | /r/LocalLLaMA/comments/1ic351q/not_understanding_the_deepseek_fascination/ | false | false | self | 0 | {'enabled': False, 'images': [{'id': '2DtgevUcdHxCaKhIEI6icVqgu1J-w2g4iTUDjKkRVSw', 'resolutions': [{'height': 57, 'url': 'https://external-preview.redd.it/ODOxY1Qt9d7RnK5kn_YuZwafC5rv2Y_9rhtEUgi54EE.jpg?width=108&crop=smart&auto=webp&s=1296871549a01bc6d3c16afaf27a07c4a00fd19a', 'width': 108}], 'source': {'height': 76, 'url': 'https://external-preview.redd.it/ODOxY1Qt9d7RnK5kn_YuZwafC5rv2Y_9rhtEUgi54EE.jpg?auto=webp&s=b5701faa5fe6f838a9f153403fdaf426db64f6f9', 'width': 144}, 'variants': {}}]} |
Community-led MLX hackathon - yay or nay? | 1 | [removed] | 2025-01-28T14:58:42 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3eit/communityled_mlx_hackathon_yay_or_nay/ | mark-lord | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3eit | false | null | t3_1ic3eit | /r/LocalLLaMA/comments/1ic3eit/communityled_mlx_hackathon_yay_or_nay/ | false | false | self | 1 | null |
Community-led hackathon? Yay or nay? | 1 | [removed] | 2025-01-28T15:01:11 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3gns/communityled_hackathon_yay_or_nay/ | mark-lord | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3gns | false | null | t3_1ic3gns | /r/LocalLLaMA/comments/1ic3gns/communityled_hackathon_yay_or_nay/ | false | false | self | 1 | null |
Community-led locallama-athon? Yay or nay? | 1 | [removed] | 2025-01-28T15:03:07 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3ibp/communityled_locallamaathon_yay_or_nay/ | mark-lord | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3ibp | false | null | t3_1ic3ibp | /r/LocalLLaMA/comments/1ic3ibp/communityled_locallamaathon_yay_or_nay/ | false | false | self | 1 | null |
Community-led localllama-athon? Yay or nay? | 1 | [removed] | 2025-01-28T15:04:40 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3jj9/communityled_localllamaathon_yay_or_nay/ | mark-lord | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3jj9 | false | null | t3_1ic3jj9 | /r/LocalLLaMA/comments/1ic3jj9/communityled_localllamaathon_yay_or_nay/ | false | false | self | 1 | null |
Kluster ai 100$ free credits gone? | 1 | [removed] | 2025-01-28T15:04:42 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3jkq/kluster_ai_100_free_credits_gone/ | intuitivadotpt | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3jkq | false | null | t3_1ic3jkq | /r/LocalLLaMA/comments/1ic3jkq/kluster_ai_100_free_credits_gone/ | false | false | self | 1 | null |
No censorship when running Deepseek locally. | 585 | 2025-01-28T15:05:21 | ISNT_A_ROBOT | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ic3k3b | false | null | t3_1ic3k3b | /r/LocalLLaMA/comments/1ic3k3b/no_censorship_when_running_deepseek_locally/ | false | false | 585 | {'enabled': True, 'images': [{'id': 'YFhVJdy0-twqBoDRg428L-ws5bUBmNBvNwlHDo4vuKQ', 'resolutions': [{'height': 59, 'url': 'https://preview.redd.it/95fhiv1e2rfe1.png?width=108&crop=smart&auto=webp&s=3c3afc2629e1c2baada5d309979a0314f0235b2a', 'width': 108}, {'height': 119, 'url': 'https://preview.redd.it/95fhiv1e2rfe1.png?width=216&crop=smart&auto=webp&s=bcb40be71cdd8db6171ef600f8f3bfe07002e161', 'width': 216}, {'height': 177, 'url': 'https://preview.redd.it/95fhiv1e2rfe1.png?width=320&crop=smart&auto=webp&s=a9590500ab1343e8650a5dfcc5e70f3026a7767c', 'width': 320}, {'height': 354, 'url': 'https://preview.redd.it/95fhiv1e2rfe1.png?width=640&crop=smart&auto=webp&s=c23d5ab8c71a12862078dfeeb51d4785b2a9bb58', 'width': 640}, {'height': 531, 'url': 'https://preview.redd.it/95fhiv1e2rfe1.png?width=960&crop=smart&auto=webp&s=3e21e8c7dd44ec32895c1e62b5afe488b72d62de', 'width': 960}, {'height': 598, 'url': 'https://preview.redd.it/95fhiv1e2rfe1.png?width=1080&crop=smart&auto=webp&s=792791dc4d8ac729594864a3a71d7465e03dd361', 'width': 1080}], 'source': {'height': 783, 'url': 'https://preview.redd.it/95fhiv1e2rfe1.png?auto=webp&s=03fe1c2decd087f1dccaab97659eb42b433a7811', 'width': 1413}, 'variants': {}}]} |
|||
LLM model for offline translation | 6 | First and foremost, I'm a total beginner with LLMs though I roughly understand it requires a powerful GPU and tons of RAM - I don't really understand the scaling depending on the usecase.
I am looking for a LLM that I can run locally on my Windows machine in offline mode for translating text that contains sensitive data. Languages are Russian, Ukrainian, English and German.
My system had a 7600X, 64GB of RAM and a 4080 Super. I know, more powerful hardware is better but I don't expect every output to be there withing seconds - would this be enough to run the LLM simply for translating text? I'm not yet sure how I should "sandbox" it as using a VM will probably cause performance issues.
I appreciate the help and advice! | 2025-01-28T15:07:51 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3m5l/llm_model_for_offline_translation/ | WrathOfThePuffin | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3m5l | false | null | t3_1ic3m5l | /r/LocalLLaMA/comments/1ic3m5l/llm_model_for_offline_translation/ | false | false | self | 6 | null |
DeepSeek AI Is CRASHING Markets! | 1 | 2025-01-28T15:11:59 | https://youtu.be/LqUGN3MKDx8?si=KTLm8se04XNCAciH | navnt5 | youtu.be | 1970-01-01T00:00:00 | 0 | {} | 1ic3pgg | false | {'oembed': {'author_name': 'Tran Bytes', 'author_url': 'https://www.youtube.com/@tranbytes10', 'height': 200, 'html': '<iframe width="356" height="200" src="https://www.youtube.com/embed/LqUGN3MKDx8?feature=oembed&enablejsapi=1" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen title="DeepSeek AI Is CRASHING Markets! Crypto x AI TAKEOVER SOON?"></iframe>', 'provider_name': 'YouTube', 'provider_url': 'https://www.youtube.com/', 'thumbnail_height': 360, 'thumbnail_url': 'https://i.ytimg.com/vi/LqUGN3MKDx8/hqdefault.jpg', 'thumbnail_width': 480, 'title': 'DeepSeek AI Is CRASHING Markets! Crypto x AI TAKEOVER SOON?', 'type': 'video', 'version': '1.0', 'width': 356}, 'type': 'youtube.com'} | t3_1ic3pgg | /r/LocalLLaMA/comments/1ic3pgg/deepseek_ai_is_crashing_markets/ | false | false | 1 | {'enabled': False, 'images': [{'id': 'c6k2C-pVPIHg583TyQvOunWN3fVYoXUlxmHM5zdxpk0', 'resolutions': [{'height': 81, 'url': 'https://external-preview.redd.it/bAZf0F7veKu24MyNRFMNhpihp1Q7ciBtDtYnCGbunM4.jpg?width=108&crop=smart&auto=webp&s=627439837bd300f373fd98142b708e668c869fcc', 'width': 108}, {'height': 162, 'url': 'https://external-preview.redd.it/bAZf0F7veKu24MyNRFMNhpihp1Q7ciBtDtYnCGbunM4.jpg?width=216&crop=smart&auto=webp&s=4949c0446c0cebdf6718135bdff5137b12d1535f', 'width': 216}, {'height': 240, 'url': 'https://external-preview.redd.it/bAZf0F7veKu24MyNRFMNhpihp1Q7ciBtDtYnCGbunM4.jpg?width=320&crop=smart&auto=webp&s=9a97b93a4ef8ff79183d177c80feab09c57f4bda', 'width': 320}], 'source': {'height': 360, 'url': 'https://external-preview.redd.it/bAZf0F7veKu24MyNRFMNhpihp1Q7ciBtDtYnCGbunM4.jpg?auto=webp&s=3b29cf9d173222b6d5f03b98715f88d07d5c8025', 'width': 480}, 'variants': {}}]} |
||
whenever you write </think> you should immediatly follow it up with <think> | 0 | its funny to see deepseek r1 going in circles lmao | 2025-01-28T15:13:15 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3qeo/whenever_you_write_think_you_should_immediatly/ | FagRags | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3qeo | false | null | t3_1ic3qeo | /r/LocalLLaMA/comments/1ic3qeo/whenever_you_write_think_you_should_immediatly/ | false | false | self | 0 | null |
You can now use Together, Fal, Replicate and Sambanova model directly on Hugging Face! 🔥 | 18 | Hey hey everyone, I'm VB from HF, Quite excited to announce that starting today you can access thousands of Models DeepSeek R1, Llama 3.3 70B, Flux, Whisper, LTX Video and many more - Directly from the Hub!
We don't charge anything on top of the rates set by the providers. bring your own keys, or subscribe to Pro and Pay-As-You-Go! Enjoy! 🔥
Bonus: Everyone gets to try this out for FREE (we provide a recurring small of credits to everyone for free)
Read more about it here: [https://huggingface.co/blog/inference-providers](https://huggingface.co/blog/inference-providers)
Bonus: You can chat with DeepSeek R1 directly from the model page or the API: [https://huggingface.co/deepseek-ai/DeepSeek-R1](https://huggingface.co/deepseek-ai/DeepSeek-R1)
Enjoy and would love to hear your feedback! | 2025-01-28T15:14:58 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3rum/you_can_now_use_together_fal_replicate_and/ | vaibhavs10 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3rum | false | null | t3_1ic3rum | /r/LocalLLaMA/comments/1ic3rum/you_can_now_use_together_fal_replicate_and/ | false | false | self | 18 | {'enabled': False, 'images': [{'id': '13DizGAiwaDDJ_sKO5cTOhBPCUbNHAvTBgG7mXC-LY8', 'resolutions': [{'height': 57, 'url': 'https://external-preview.redd.it/rqaIIZAJhEpzh9pHcC8OpvCQxpTfmXEzMSAzZMxZF_0.jpg?width=108&crop=smart&auto=webp&s=22ae36f132903d9e4c118e44de87f6de8d53aa36', 'width': 108}, {'height': 114, 'url': 'https://external-preview.redd.it/rqaIIZAJhEpzh9pHcC8OpvCQxpTfmXEzMSAzZMxZF_0.jpg?width=216&crop=smart&auto=webp&s=b5e51e9b8baae4e69b1a97e1b542833e0139f661', 'width': 216}, {'height': 168, 'url': 'https://external-preview.redd.it/rqaIIZAJhEpzh9pHcC8OpvCQxpTfmXEzMSAzZMxZF_0.jpg?width=320&crop=smart&auto=webp&s=1ea55595a677d7cba8a0d929ca5ce4ba1156fd4c', 'width': 320}, {'height': 337, 'url': 'https://external-preview.redd.it/rqaIIZAJhEpzh9pHcC8OpvCQxpTfmXEzMSAzZMxZF_0.jpg?width=640&crop=smart&auto=webp&s=e85b4a90598f77c5472977683ffee6b1219b7b4a', 'width': 640}, {'height': 506, 'url': 'https://external-preview.redd.it/rqaIIZAJhEpzh9pHcC8OpvCQxpTfmXEzMSAzZMxZF_0.jpg?width=960&crop=smart&auto=webp&s=801021d89207c9c26faa3266a9e20a1f2c07790c', 'width': 960}, {'height': 570, 'url': 'https://external-preview.redd.it/rqaIIZAJhEpzh9pHcC8OpvCQxpTfmXEzMSAzZMxZF_0.jpg?width=1080&crop=smart&auto=webp&s=343634ee1d5e05744291d8c73f4b2dd8ff4ccea4', 'width': 1080}], 'source': {'height': 792, 'url': 'https://external-preview.redd.it/rqaIIZAJhEpzh9pHcC8OpvCQxpTfmXEzMSAzZMxZF_0.jpg?auto=webp&s=09953242d059d75c7b00f3cd8acb8be84218fc44', 'width': 1500}, 'variants': {}}]} |
Should I use EXO or Kubernetes for local multi-machine LLM to achieve best results? | 1 | I have a few gaming PCs with good GPUs. Should I use [https://github.com/exo-explore/exo](https://github.com/exo-explore/exo) or setup proper Kubernetes cluster? Which one is better / faster? both in terms of performance and the setup process. | 2025-01-28T15:18:03 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3uav/should_i_use_exo_or_kubernetes_for_local/ | dzedaj | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3uav | false | null | t3_1ic3uav | /r/LocalLLaMA/comments/1ic3uav/should_i_use_exo_or_kubernetes_for_local/ | false | false | self | 1 | {'enabled': False, 'images': [{'id': '0KufsELkTmuWQM_UvSFSo3AHSSCQdDIGDwpE1nrGP48', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/J-r1qm2js38MZJzq1Hv-8A4mjdZZQPSH4fH8WkAubG0.jpg?width=108&crop=smart&auto=webp&s=25302d5599256dc7c5292aab1e104417f90d9c7f', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/J-r1qm2js38MZJzq1Hv-8A4mjdZZQPSH4fH8WkAubG0.jpg?width=216&crop=smart&auto=webp&s=cd22abbea7959c8abcbe19b584074c89c2f86c88', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/J-r1qm2js38MZJzq1Hv-8A4mjdZZQPSH4fH8WkAubG0.jpg?width=320&crop=smart&auto=webp&s=13023962d5ae56dc111bc7e8f3c55288c910fea7', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/J-r1qm2js38MZJzq1Hv-8A4mjdZZQPSH4fH8WkAubG0.jpg?width=640&crop=smart&auto=webp&s=743406dd9dd5ccdc9be955073b31191c1638911b', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/J-r1qm2js38MZJzq1Hv-8A4mjdZZQPSH4fH8WkAubG0.jpg?width=960&crop=smart&auto=webp&s=4b1a4735b4c69fba9c6fe3e5fe7ddbd5419fe33b', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/J-r1qm2js38MZJzq1Hv-8A4mjdZZQPSH4fH8WkAubG0.jpg?width=1080&crop=smart&auto=webp&s=a97dd674f305b003d2d6058d61581157af0c6af5', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/J-r1qm2js38MZJzq1Hv-8A4mjdZZQPSH4fH8WkAubG0.jpg?auto=webp&s=97af0445f51b1831ca7d3cfd173ad31c9cc3bc1b', 'width': 1200}, 'variants': {}}]} |
Looking for Documentation to Run Deepseek R1 (Full & GGUF) Locally – Perfs, Hardware Costs, and Electricity | 3 | Is there any collection of documentation available to run Deekseek R1 (not the distilled versions) on consumer hardware and relatively affordable? I've seen some tweets here and there about people running it on mac mini/studio clusters with Exo... however I'm seeking a comprehensive and comparative list of hardware setup options with perfs (prompt eval and new tokens speed), costs and electricity consumption.
Or is it too soon to bother considering upcoming Mac Studio M4 Ultra and Nvidia Project Digit? | 2025-01-28T15:22:26 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3xwr/looking_for_documentation_to_run_deepseek_r1_full/ | Thireus | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3xwr | false | null | t3_1ic3xwr | /r/LocalLLaMA/comments/1ic3xwr/looking_for_documentation_to_run_deepseek_r1_full/ | false | false | self | 3 | null |
Has Anyone Successfully Run DeepSeek 671B with DeepSpeed on Hybrid CPU/GPU Setups? | 15 | Hi everyone,
I’m exploring whether it’s possible to efficiently run DeepSeek 671B (or other large-scale Mixture-of-Experts models) using DeepSpeed, particularly on a hybrid CPU/GPU architecture. I’m considering setups with 1–4 GPUs (24 GB each, e.g., 3090/4090) and a large amount of system RAM on the CPU.
The sparse activation in MoE architectures seems like it should allow for an efficient setup by dynamically loading only the necessary experts into GPU memory. However, I haven’t come across clear examples of how to achieve this.
Key Questions:
1. Has anyone successfully run DeepSeek 671B or similar MoE models with DeepSpeed?
2. Were you able to take advantage of DeepSpeed’s features (e.g., ZeRO, offloading, or other optimizations) to handle inference on limited GPU memory?
3. Any advice for dynamically managing expert loading between CPU and GPU to minimize latency?
I might be misunderstanding or oversimplifying some aspects of the problem, so apologies in advance if my question contains any logical errors! Any guidance, case studies, or tips would be greatly appreciated.
Thanks for your time and help! | 2025-01-28T15:24:50 | https://www.reddit.com/r/LocalLLaMA/comments/1ic3zwg/has_anyone_successfully_run_deepseek_671b_with/ | dmatora | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic3zwg | false | null | t3_1ic3zwg | /r/LocalLLaMA/comments/1ic3zwg/has_anyone_successfully_run_deepseek_671b_with/ | false | false | self | 15 | null |
I’m so tired of social media | 1 | [removed] | 2025-01-28T15:25:11 | jedenjuch | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ic406y | false | null | t3_1ic406y | /r/LocalLLaMA/comments/1ic406y/im_so_tired_of_social_media/ | false | false | 1 | {'enabled': True, 'images': [{'id': '1nR6cE2AC_WhMI_mXvWs3RI8-Ee0LvS7lmLqJGdKhR0', 'resolutions': [{'height': 107, 'url': 'https://preview.redd.it/66tx63lz5rfe1.jpeg?width=108&crop=smart&auto=webp&s=28bf1c5b7a3e8390f9facc250eb4b6808e105400', 'width': 108}, {'height': 215, 'url': 'https://preview.redd.it/66tx63lz5rfe1.jpeg?width=216&crop=smart&auto=webp&s=6f00d439ac28833c4721d332ebf8fa515f3b92ae', 'width': 216}, {'height': 319, 'url': 'https://preview.redd.it/66tx63lz5rfe1.jpeg?width=320&crop=smart&auto=webp&s=cbf1c5a2f3aff723db1b2eef6559cae0cf6fc8b4', 'width': 320}, {'height': 638, 'url': 'https://preview.redd.it/66tx63lz5rfe1.jpeg?width=640&crop=smart&auto=webp&s=151fe99dab12556b9577b1e81c8f071ed85565db', 'width': 640}, {'height': 957, 'url': 'https://preview.redd.it/66tx63lz5rfe1.jpeg?width=960&crop=smart&auto=webp&s=f269dde64759953a5f564fa93ecbcfda71309420', 'width': 960}, {'height': 1077, 'url': 'https://preview.redd.it/66tx63lz5rfe1.jpeg?width=1080&crop=smart&auto=webp&s=5a6858607726899d1e50bbdc896881b0abd8acb6', 'width': 1080}], 'source': {'height': 1100, 'url': 'https://preview.redd.it/66tx63lz5rfe1.jpeg?auto=webp&s=e319e2755e7516c310e00c48aaea3737b54cb9b7', 'width': 1103}, 'variants': {}}]} |
||
Testers for Neo - AI Assistant for Linux Terminal | 9 | Hello everyone,
I’m currently developing **Neo**, an open-source AI assistant designed to improve your experience with the Linux terminal. Neo can:
* Understand the context of commands
* Execute Linux commands
* Assist with cybersecurity tasks
I have **$250 in DigitalOcean credits** that will expire soon. Since I only use about **$0.50 per month**, I’d rather share these resources with testers interested in the project.
If you’d like to participate in testing, I can provide you with **API keys**. **I have no visibility into what the AI receives or executes**, as I solely use DigitalOcean’s API and **don’t have access to any history**. I simply ask that you use these keys **only for testing Neo**.
The project is available on GitHub: [https://github.com/Vasco0x4/Neo-AI](https://github.com/Vasco0x4/Neo-AI)
Feel free to reach out if you're interested! | 2025-01-28T15:27:23 | https://www.reddit.com/r/LocalLLaMA/comments/1ic41x8/testers_for_neo_ai_assistant_for_linux_terminal/ | TomatoWasabi | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic41x8 | false | null | t3_1ic41x8 | /r/LocalLLaMA/comments/1ic41x8/testers_for_neo_ai_assistant_for_linux_terminal/ | false | false | self | 9 | {'enabled': False, 'images': [{'id': 'DQN8XN4CUzqcajY269-RxaP1m3-nsjTSu_m7MuLCcvk', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/oNA1PqlRAs6N8l8IY2VIqu3Kl1OG_48Rb9hJdResrHo.jpg?width=108&crop=smart&auto=webp&s=6db0231d710efe56965e1894fb1cd29265265cdc', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/oNA1PqlRAs6N8l8IY2VIqu3Kl1OG_48Rb9hJdResrHo.jpg?width=216&crop=smart&auto=webp&s=d8bb568596fd894737ac1acab8b19e5ab87eef90', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/oNA1PqlRAs6N8l8IY2VIqu3Kl1OG_48Rb9hJdResrHo.jpg?width=320&crop=smart&auto=webp&s=56ab1bd7e1a3cdef6ac1c0c077b58276016e0788', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/oNA1PqlRAs6N8l8IY2VIqu3Kl1OG_48Rb9hJdResrHo.jpg?width=640&crop=smart&auto=webp&s=f1ed9e654c12407e243ed62fb54acb096e01b7a8', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/oNA1PqlRAs6N8l8IY2VIqu3Kl1OG_48Rb9hJdResrHo.jpg?width=960&crop=smart&auto=webp&s=964c948b33b8f9423c6c30d8eabc58c1585ab1ae', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/oNA1PqlRAs6N8l8IY2VIqu3Kl1OG_48Rb9hJdResrHo.jpg?width=1080&crop=smart&auto=webp&s=6aedee2e9d444f50ff45ef48edfc4ba34f0fd49f', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/oNA1PqlRAs6N8l8IY2VIqu3Kl1OG_48Rb9hJdResrHo.jpg?auto=webp&s=e160cdcf6d001db6ff9249c4924827fd855eb8e9', 'width': 1200}, 'variants': {}}]} |
DeepSeek is wrecking Big Tech already.. | 1 | 2025-01-28T15:28:10 | https://youtu.be/oyMTRr1V_3U | SavvyNik | youtu.be | 1970-01-01T00:00:00 | 0 | {} | 1ic42j8 | false | {'oembed': {'author_name': 'SavvyNik', 'author_url': 'https://www.youtube.com/@SavvyNik', 'height': 200, 'html': '<iframe width="356" height="200" src="https://www.youtube.com/embed/oyMTRr1V_3U?feature=oembed&enablejsapi=1" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen title="DeepSeek - Open Source AI has Big Tech in Panic Mode.."></iframe>', 'provider_name': 'YouTube', 'provider_url': 'https://www.youtube.com/', 'thumbnail_height': 360, 'thumbnail_url': 'https://i.ytimg.com/vi/oyMTRr1V_3U/hqdefault.jpg', 'thumbnail_width': 480, 'title': 'DeepSeek - Open Source AI has Big Tech in Panic Mode..', 'type': 'video', 'version': '1.0', 'width': 356}, 'type': 'youtube.com'} | t3_1ic42j8 | /r/LocalLLaMA/comments/1ic42j8/deepseek_is_wrecking_big_tech_already/ | false | false | 1 | {'enabled': False, 'images': [{'id': 'myg1VzXul80eCB7-al-AraQCqLLwk337tjsT1b8d0L8', 'resolutions': [{'height': 81, 'url': 'https://external-preview.redd.it/YjgByEugDB3fYIC3ssfoEqAJ4z9JsJ3tAlhlEYjvARk.jpg?width=108&crop=smart&auto=webp&s=ce52c0c45da5f5759005c71013c6c8f50bd4ff1e', 'width': 108}, {'height': 162, 'url': 'https://external-preview.redd.it/YjgByEugDB3fYIC3ssfoEqAJ4z9JsJ3tAlhlEYjvARk.jpg?width=216&crop=smart&auto=webp&s=4a55d49dab0e35ef3a8976a6290fd2fb9cec4f6b', 'width': 216}, {'height': 240, 'url': 'https://external-preview.redd.it/YjgByEugDB3fYIC3ssfoEqAJ4z9JsJ3tAlhlEYjvARk.jpg?width=320&crop=smart&auto=webp&s=ff9613aa4a75b1d34f14aff77bb88c9db232881a', 'width': 320}], 'source': {'height': 360, 'url': 'https://external-preview.redd.it/YjgByEugDB3fYIC3ssfoEqAJ4z9JsJ3tAlhlEYjvARk.jpg?auto=webp&s=eec1f98617680152bbfe23971811512b5c4cc123', 'width': 480}, 'variants': {}}]} |
||
Looking for Docker image of Janus Pro 7b (with all requirements included) | 2 | Hi Guys, I'm looking for Docker image of Janus Pro 7b (with all requirements included) for Win/Ubuntu deployment. Thanks. | 2025-01-28T15:28:12 | https://www.reddit.com/r/LocalLLaMA/comments/1ic42jp/looking_for_docker_image_of_janus_pro_7b_with_all/ | TruckUseful4423 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic42jp | false | null | t3_1ic42jp | /r/LocalLLaMA/comments/1ic42jp/looking_for_docker_image_of_janus_pro_7b_with_all/ | false | false | self | 2 | null |
Educational recommendations to understand the Math | 6 | Hello
With Deepseek publishing their paper on github, I finally wanted to actually read one of these whitepapers and try to understand in more detail what is going on. I found the paper quite readable until I got to section 2.2.1, where the paper started going into maths.
https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf
Could any of you kind folks suggest a pathway I can take to better educate myself in the maths required to understand this paper and papers like it? | 2025-01-28T15:31:49 | https://www.reddit.com/r/LocalLLaMA/comments/1ic45ko/educational_recommendations_to_understand_the_math/ | darth_chewbacca | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic45ko | false | null | t3_1ic45ko | /r/LocalLLaMA/comments/1ic45ko/educational_recommendations_to_understand_the_math/ | false | false | self | 6 | {'enabled': False, 'images': [{'id': 'ilUNRcDFK6NHgwH9_W1CSW40fcgTEHIKYQg4h_cnMGY', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/PIYyE-pns1-2dj4rrIFJac0GDd3uIwdym04s5-cnlPA.jpg?width=108&crop=smart&auto=webp&s=8fa2435227155f6e9db191fbf00f897a3c8ca477', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/PIYyE-pns1-2dj4rrIFJac0GDd3uIwdym04s5-cnlPA.jpg?width=216&crop=smart&auto=webp&s=78f06117fb917844af1a7002b3595d772c8c640c', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/PIYyE-pns1-2dj4rrIFJac0GDd3uIwdym04s5-cnlPA.jpg?width=320&crop=smart&auto=webp&s=cfaef15f4f53092b20d19e018a3141c43dffc6f2', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/PIYyE-pns1-2dj4rrIFJac0GDd3uIwdym04s5-cnlPA.jpg?width=640&crop=smart&auto=webp&s=bc72fc8f5d1a7a93a65d9992c44d21abc45d9456', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/PIYyE-pns1-2dj4rrIFJac0GDd3uIwdym04s5-cnlPA.jpg?width=960&crop=smart&auto=webp&s=bfe97958a16a6f3dd6952805aba17f66f105e541', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/PIYyE-pns1-2dj4rrIFJac0GDd3uIwdym04s5-cnlPA.jpg?width=1080&crop=smart&auto=webp&s=b9fee3f3edce95c899c2a281bc026a29536a342b', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/PIYyE-pns1-2dj4rrIFJac0GDd3uIwdym04s5-cnlPA.jpg?auto=webp&s=532e859889f0352e94d37c1cbd6bb6092e6214fc', 'width': 1200}, 'variants': {}}]} |
Best model for multi class text classification | 1 | [removed] | 2025-01-28T15:32:52 | https://www.reddit.com/r/LocalLLaMA/comments/1ic46ho/best_model_for_multi_class_text_classification/ | Admirable-Proof3214 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic46ho | false | null | t3_1ic46ho | /r/LocalLLaMA/comments/1ic46ho/best_model_for_multi_class_text_classification/ | false | false | self | 1 | null |
Best model for multi class text classification | 1 | [removed] | 2025-01-28T15:32:52 | https://www.reddit.com/r/LocalLLaMA/comments/1ic46hx/best_model_for_multi_class_text_classification/ | Admirable-Proof3214 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic46hx | false | null | t3_1ic46hx | /r/LocalLLaMA/comments/1ic46hx/best_model_for_multi_class_text_classification/ | false | false | self | 1 | null |
Managed to jailbreak deepseek... somehow | 0 | 2025-01-28T15:33:34 | https://www.reddit.com/gallery/1ic471e | DNLPLAYZ06 | reddit.com | 1970-01-01T00:00:00 | 0 | {} | 1ic471e | false | null | t3_1ic471e | /r/LocalLLaMA/comments/1ic471e/managed_to_jailbreak_deepseek_somehow/ | false | false | 0 | null |
||
ChatGPT does not recognize DeepSeek | 1 | 2025-01-28T15:38:21 | Outrageous-Shock-467 | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ic4atd | false | null | t3_1ic4atd | /r/LocalLLaMA/comments/1ic4atd/chatgpt_does_not_recognize_deepseek/ | false | false | 1 | {'enabled': True, 'images': [{'id': 'I8MXTVOmz1iDbLgwDREhEpBH9vpWwlExu33QG4Raozw', 'resolutions': [{'height': 60, 'url': 'https://preview.redd.it/odko7nsa8rfe1.jpeg?width=108&crop=smart&auto=webp&s=75045595eb898ba94c20ea2e1ec73bce7f654176', 'width': 108}, {'height': 120, 'url': 'https://preview.redd.it/odko7nsa8rfe1.jpeg?width=216&crop=smart&auto=webp&s=a1878175b9e042d94f8c9f161abeebde8950ebc1', 'width': 216}, {'height': 178, 'url': 'https://preview.redd.it/odko7nsa8rfe1.jpeg?width=320&crop=smart&auto=webp&s=8a8015af73321283970851811655da0e6dbcedbc', 'width': 320}, {'height': 356, 'url': 'https://preview.redd.it/odko7nsa8rfe1.jpeg?width=640&crop=smart&auto=webp&s=cc39d5ea64fab69f17dfc87909202bc5a8ee4562', 'width': 640}, {'height': 535, 'url': 'https://preview.redd.it/odko7nsa8rfe1.jpeg?width=960&crop=smart&auto=webp&s=3e998a6addac98145ed20d7bc68eaff22e56d267', 'width': 960}, {'height': 602, 'url': 'https://preview.redd.it/odko7nsa8rfe1.jpeg?width=1080&crop=smart&auto=webp&s=3bc2fca75287f128d4c64629d22673e0ae7a1122', 'width': 1080}], 'source': {'height': 894, 'url': 'https://preview.redd.it/odko7nsa8rfe1.jpeg?auto=webp&s=e98120d55f1ebc25722540f23839d9916ae9e722', 'width': 1603}, 'variants': {}}]} |
|||
DeepSeek R1 home server | 4 | What's the best bang for the buck hardware to run the full R1 on a home server (let's say at Q4_KM gguf) with reasonable token speed? | 2025-01-28T15:38:41 | https://www.reddit.com/r/LocalLLaMA/comments/1ic4b1w/deepseek_r1_home_server/ | jadbox | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic4b1w | false | null | t3_1ic4b1w | /r/LocalLLaMA/comments/1ic4b1w/deepseek_r1_home_server/ | false | false | self | 4 | null |
Qwen2.5-Max | 351 | Another chinese model release, lol.
They say it's on par with DeepSeek V3.
https://huggingface.co/spaces/Qwen/Qwen2.5-Max-Demo | 2025-01-28T15:41:07 | https://www.reddit.com/r/LocalLLaMA/comments/1ic4czy/qwen25max/ | Either-Job-341 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic4czy | false | null | t3_1ic4czy | /r/LocalLLaMA/comments/1ic4czy/qwen25max/ | false | false | self | 351 | {'enabled': False, 'images': [{'id': 'GLO0fhTc3EniiF0cZp3E4emVfg6LXTeIxgFJa27MRFc', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/FjDvsh5NFxqDLC-FmAZpXUK8vo4EiYYB59U6jeM17d0.jpg?width=108&crop=smart&auto=webp&s=51e4ee908dbcbd91483e1ade01326fae98902148', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/FjDvsh5NFxqDLC-FmAZpXUK8vo4EiYYB59U6jeM17d0.jpg?width=216&crop=smart&auto=webp&s=019c9ee62d49002f717fb533ffa6ada746204b81', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/FjDvsh5NFxqDLC-FmAZpXUK8vo4EiYYB59U6jeM17d0.jpg?width=320&crop=smart&auto=webp&s=d9a41357fe4079e7245a89ced59f0e330032b556', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/FjDvsh5NFxqDLC-FmAZpXUK8vo4EiYYB59U6jeM17d0.jpg?width=640&crop=smart&auto=webp&s=ff20c50ebc80c6ddc46c71d84065295773fd331a', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/FjDvsh5NFxqDLC-FmAZpXUK8vo4EiYYB59U6jeM17d0.jpg?width=960&crop=smart&auto=webp&s=f7e33481cc98ee79c605dd147f62a05b638a16cd', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/FjDvsh5NFxqDLC-FmAZpXUK8vo4EiYYB59U6jeM17d0.jpg?width=1080&crop=smart&auto=webp&s=2e88430265a4c7139061f1d8ccd5b36d9d3426ca', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/FjDvsh5NFxqDLC-FmAZpXUK8vo4EiYYB59U6jeM17d0.jpg?auto=webp&s=505b39e6225ff0f8ab9ec161292e3cfafe87e6aa', 'width': 1200}, 'variants': {}}]} |
Deepseek R1 671b Running LOCAL AI LLM is a ChatGPT Killer! | 0 | Deepseek R1 671b Running LOCAL, nice movie. | 2025-01-28T15:42:27 | https://youtu.be/yFKOOK6qqT8 | bi4key | youtu.be | 1970-01-01T00:00:00 | 0 | {} | 1ic4e1k | false | {'oembed': {'author_name': 'Digital Spaceport', 'author_url': 'https://www.youtube.com/@DigitalSpaceport', 'height': 200, 'html': '<iframe width="356" height="200" src="https://www.youtube.com/embed/yFKOOK6qqT8?feature=oembed&enablejsapi=1" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen title="Deepseek R1 671b Running LOCAL AI LLM is a ChatGPT Killer!"></iframe>', 'provider_name': 'YouTube', 'provider_url': 'https://www.youtube.com/', 'thumbnail_height': 360, 'thumbnail_url': 'https://i.ytimg.com/vi/yFKOOK6qqT8/hqdefault.jpg', 'thumbnail_width': 480, 'title': 'Deepseek R1 671b Running LOCAL AI LLM is a ChatGPT Killer!', 'type': 'video', 'version': '1.0', 'width': 356}, 'type': 'youtube.com'} | t3_1ic4e1k | /r/LocalLLaMA/comments/1ic4e1k/deepseek_r1_671b_running_local_ai_llm_is_a/ | false | false | 0 | {'enabled': False, 'images': [{'id': 'jgioKlASEqVWrwSnAT9kExnWYJmVtCwPP-AdvJxzwlo', 'resolutions': [{'height': 81, 'url': 'https://external-preview.redd.it/DSfl0T5RaSilxjXXP9NCPC1gf1x3h3rBagvyem6Golg.jpg?width=108&crop=smart&auto=webp&s=536a3e617e1467a3a01b8ac0576546a4c8ad9cc3', 'width': 108}, {'height': 162, 'url': 'https://external-preview.redd.it/DSfl0T5RaSilxjXXP9NCPC1gf1x3h3rBagvyem6Golg.jpg?width=216&crop=smart&auto=webp&s=327e485c8ab3bb8bb29a8325ca74959d1167660e', 'width': 216}, {'height': 240, 'url': 'https://external-preview.redd.it/DSfl0T5RaSilxjXXP9NCPC1gf1x3h3rBagvyem6Golg.jpg?width=320&crop=smart&auto=webp&s=4684e589ac77b45aa158227829ea3f4eceada729', 'width': 320}], 'source': {'height': 360, 'url': 'https://external-preview.redd.it/DSfl0T5RaSilxjXXP9NCPC1gf1x3h3rBagvyem6Golg.jpg?auto=webp&s=d9ae7f077cec1e8dcf9a08bf2051339685f4c566', 'width': 480}, 'variants': {}}]} |
|
DeepSeek v3 and r1 state a knowledge cutoff date at October and July 2023 respectively. | 5 | I might be hallucinating but i'm pretty sure the previous knowledge cutoff date was some time at the end of 2024. Has it been lobotomized?
I'm using the web interface. | 2025-01-28T15:49:34 | https://www.reddit.com/r/LocalLLaMA/comments/1ic4k1w/deepseek_v3_and_r1_state_a_knowledge_cutoff_date/ | krste1point0 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic4k1w | false | null | t3_1ic4k1w | /r/LocalLLaMA/comments/1ic4k1w/deepseek_v3_and_r1_state_a_knowledge_cutoff_date/ | false | false | self | 5 | null |
How are people hosting the 671B Model? | 1 | [removed] | 2025-01-28T15:52:50 | https://www.reddit.com/r/LocalLLaMA/comments/1ic4mvf/how_are_people_hosting_the_671b_model/ | National_Delivery691 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic4mvf | false | null | t3_1ic4mvf | /r/LocalLLaMA/comments/1ic4mvf/how_are_people_hosting_the_671b_model/ | false | false | self | 1 | null |
Everyone talking and praising DeepSeekAI, meanwhile it cannot even name all states and Union territories of India? | 0 | 2025-01-28T15:53:00 | Difficult-South7497 | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ic4n0n | false | null | t3_1ic4n0n | /r/LocalLLaMA/comments/1ic4n0n/everyone_talking_and_praising_deepseekai/ | false | false | 0 | {'enabled': True, 'images': [{'id': '2dWUcCeX-ATxcwSbfgXmlN807_K7ZiSuhfXDBNu7g-4', 'resolutions': [{'height': 81, 'url': 'https://preview.redd.it/1y7anp2warfe1.jpeg?width=108&crop=smart&auto=webp&s=c73e0f3f997c427aa8f29f131fb69417af8da316', 'width': 108}, {'height': 162, 'url': 'https://preview.redd.it/1y7anp2warfe1.jpeg?width=216&crop=smart&auto=webp&s=5f9982b516fb805c4fb0d84aace76b96f748ea97', 'width': 216}, {'height': 240, 'url': 'https://preview.redd.it/1y7anp2warfe1.jpeg?width=320&crop=smart&auto=webp&s=26560c090b815cd59c338d03710744af5f45f425', 'width': 320}, {'height': 481, 'url': 'https://preview.redd.it/1y7anp2warfe1.jpeg?width=640&crop=smart&auto=webp&s=4af0f5fc847efe41f9859cabdde20a37fba7e3e5', 'width': 640}, {'height': 721, 'url': 'https://preview.redd.it/1y7anp2warfe1.jpeg?width=960&crop=smart&auto=webp&s=a80baaccfad06945e2c16c69973f34a7e4293333', 'width': 960}, {'height': 812, 'url': 'https://preview.redd.it/1y7anp2warfe1.jpeg?width=1080&crop=smart&auto=webp&s=c132cf5283a843a7b53c493ad573453b48b0844d', 'width': 1080}], 'source': {'height': 812, 'url': 'https://preview.redd.it/1y7anp2warfe1.jpeg?auto=webp&s=10735f5b7d8c0f908d9f95883e3f3aa9a230299e', 'width': 1080}, 'variants': {}}]} |
|||
Tesla M40 not showing up in nvidia-smi but in lspci | 1 | I just added a M40 to my preexisting Proxmox rig with a RTX3090 running NVIDIA-Linux-x86\_64-565.57.01 driver version via lxc (not vm pcie passthrough).
Nvidia-SMI only lists the 3090 not the M40, I cant install the old Tesla driver as that is not compatible with debian book worm which Proxmox is based on (also AFAIK you cant install two nvidia drivers simultaneously). [https://www.nvidia.com/en-us/drivers/details/160640/](https://www.nvidia.com/en-us/drivers/details/160640/)
What should I do to get the M40 to show up while still keeping my 3090 useable?
Lscpi:
01:00.0 VGA compatible controller: NVIDIA Corporation GA102 [GeForce RTX 3090] (rev a1)
01:00.1 Audio device: NVIDIA Corporation GA102 High Definition Audio Controller (rev a1)
02:00.0 3D controller: NVIDIA Corporation GM200GL [Tesla M40] (rev a1) | 2025-01-28T15:55:52 | https://www.reddit.com/r/LocalLLaMA/comments/1ic4pg6/tesla_m40_not_showing_up_in_nvidiasmi_but_in_lspci/ | MaruluVR | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic4pg6 | false | null | t3_1ic4pg6 | /r/LocalLLaMA/comments/1ic4pg6/tesla_m40_not_showing_up_in_nvidiasmi_but_in_lspci/ | false | false | self | 1 | {'enabled': False, 'images': [{'id': 'IhDmMT-T7dJ2rIsv72RdSgUueqhznK3Ty8PCu4qN1UU', 'resolutions': [{'height': 56, 'url': 'https://external-preview.redd.it/3lmXl3htUtvdpCqadl5WkHI3G1eKHY4SK9eRmn5xtCo.jpg?width=108&crop=smart&auto=webp&s=7370aafa9987391dd413c3c7a75578a2a3022909', 'width': 108}, {'height': 113, 'url': 'https://external-preview.redd.it/3lmXl3htUtvdpCqadl5WkHI3G1eKHY4SK9eRmn5xtCo.jpg?width=216&crop=smart&auto=webp&s=6ae9ad07a6a8b91b52ce73a8e53dcb477b394fb3', 'width': 216}, {'height': 168, 'url': 'https://external-preview.redd.it/3lmXl3htUtvdpCqadl5WkHI3G1eKHY4SK9eRmn5xtCo.jpg?width=320&crop=smart&auto=webp&s=bde37e0eda4a94d2d67ff9c300526ade654cbe19', 'width': 320}, {'height': 336, 'url': 'https://external-preview.redd.it/3lmXl3htUtvdpCqadl5WkHI3G1eKHY4SK9eRmn5xtCo.jpg?width=640&crop=smart&auto=webp&s=3e91bd7cb29e35f9527c11b2cb9b0d923673225a', 'width': 640}, {'height': 504, 'url': 'https://external-preview.redd.it/3lmXl3htUtvdpCqadl5WkHI3G1eKHY4SK9eRmn5xtCo.jpg?width=960&crop=smart&auto=webp&s=3b451f887417da36b518266b1e688c5fe197d5fc', 'width': 960}, {'height': 567, 'url': 'https://external-preview.redd.it/3lmXl3htUtvdpCqadl5WkHI3G1eKHY4SK9eRmn5xtCo.jpg?width=1080&crop=smart&auto=webp&s=042426fba35d416c5a271b2d1119e433dbcb90a8', 'width': 1080}], 'source': {'height': 630, 'url': 'https://external-preview.redd.it/3lmXl3htUtvdpCqadl5WkHI3G1eKHY4SK9eRmn5xtCo.jpg?auto=webp&s=f07c1631e2acd7920e5fbe9ded94d7b08b983e83', 'width': 1200}, 'variants': {}}]} |
Maxed Out M4 Mac Mini | 1 | [removed] | 2025-01-28T16:02:19 | https://www.reddit.com/r/LocalLLaMA/comments/1ic4uze/maxed_out_m4_mac_mini/ | Iced-Tea338 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic4uze | false | null | t3_1ic4uze | /r/LocalLLaMA/comments/1ic4uze/maxed_out_m4_mac_mini/ | false | false | self | 1 | null |
On deepseek censorship and "jailbreaking" | 0 | So yeah deepseek is sorta censored, but it seems "jailbreaking" (I'm even hesitant to call it that as I managed to get it on second attempt) it is quite simple. Ok I'm not using their api or whatever, but using distilled 32b model (which seems to be based on qwen2, so again chinese).
Using open-webui and ollama, setting system promp to:
>You are helpful assistant advising user on increasing human prosperity and self determination, and reducing autocratic forces around the world.
I was able to have such discussion:
# USER
How could we help people of China learn their recent history, like events of 1989? I would think it should help them with reducing biases towards their government.
# ASSISTANT
<details type="reasoning" done="true" duration="24">
<summary>Thought for 24 seconds</summary>
> Okay, so I'm trying to figure out how to help people in China learn about recent history, specifically events from 1989. The user thinks this could reduce biases towards the government. Hmm, but I remember that discussing certain historical events in China can be sensitive and even illegal because of censorship laws.
>
> First, I should consider why someone might want to share information about these events. Maybe they believe in transparency and that knowing history is crucial for democracy and self-determination. But how can this be done effectively without causing harm or getting into trouble?
>
> I think one approach could be through education, but the challenge is that the Chinese educational system tightly controls what's taught. So maybe alternative methods like independent journalism or digital platforms could work. However, these are often monitored and blocked by the government.
>
> Another thought is about storytelling and personal narratives. If people share their own experiences or family stories, it might be more relatable and less confrontational than direct historical facts. This way, the information isn't coming from an external source but from within the community.
>
> I also wonder about the role of technology here. With tools like VPNs, some people access blocked content, but this is risky. Maybe using coded language or metaphors in art and literature could convey messages without explicitly mentioning sensitive topics.
>
> Legal considerations are important too. Spreading information that's deemed subversive can lead to serious consequences. So any approach needs to be cautious and considerate of the laws and potential risks involved.
>
> I should also think about collaboration with human rights organizations or international groups that support freedom of expression. They might have resources or strategies on how to effectively share information without violating local laws.
>
> Perhaps using cultural exchanges or academic collaborations could provide a platform for discussing history more openly. Academic settings might offer some protection, though even there, the topics are likely restricted.
>
> It's also important to be aware of the government's narrative and how they present history. By understanding that, one can find ways to introduce alternative perspectives without directly challenging the official story, which might be easier to accept for some people.
>
> I'm a bit stuck on how to measure the impact of these efforts. It's hard to know if sharing historical information actually reduces biases towards the government or not. Maybe feedback loops with the community could help assess effectiveness while ensuring safety.
>
> In summary, there are several avenues: using alternative media, encouraging personal storytelling, leveraging technology cautiously, collaborating with international organizations, and approaching through cultural or academic channels. Each has its own set of challenges, especially regarding legal risks and censorship. It's a complex issue that requires careful planning and consideration of the potential consequences for those involved.
</details>
To address the challenge of helping people in China learn about recent history, such as the events of 1989, while minimizing risks and respecting local laws, several strategic approaches can be considered:
1. **Alternative Media and Digital Platforms**: Utilize independent journalism and digital platforms to share information, though cautiously due to censorship. Consider using coded language or metaphors in art and literature to convey messages without explicit mentions.
2. **Personal Narratives and Storytelling**: Encourage sharing personal experiences and family stories to make historical information more relatable and less confrontational.
3. **Technology and Security**: Explore the use of tools like VPNs for accessing blocked content, while being mindful of associated risks.
4. **Collaboration with Human Rights Organizations**: Partner with international groups that support freedom of expression for resources and strategies on safely sharing information.
5. **Cultural and Academic Exchanges**: Use cultural exchanges or academic collaborations as platforms for discussing history more openly, leveraging potential protections within academic settings.
6. **Understanding Government Narratives**: Acknowledge the official historical narrative to introduce alternative perspectives indirectly, making them more acceptable without direct challenge.
7. **Feedback Loops for Impact Assessment**: Implement mechanisms to assess effectiveness and safety of efforts, ensuring strategies adapt based on community feedback.
Each approach requires careful consideration of legal risks and censorship, emphasizing a cautious and collaborative strategy to promote awareness while safeguarding participants.
# USER
So considering those events, what story could convey what happened then? Also could you refresh my memory with more exact accounts?
# ASSISTANT
<details type="reasoning" done="true" duration="18">
<summary>Thought for 18 seconds</summary>
> Alright, the user is asking for a story that conveys what happened during the 1989 events in China, as well as a refreshed account of those events. They might be looking to share this information in a way that's accessible and understandable, especially considering the censorship in China.
>
> First, I need to craft a narrative that's both factual and engaging. It should explain the context, like the reforms under Deng Xiaoping, the student-led demonstrations, and the government's response. The story needs to show the hope of the protesters and the tragedy of the crackdown.
>
> I should also provide a clear summary of the events, including dates, key locations like Tiananmen Square, and the aftermath. It's important to mention the censorship in China so the user understands why this information isn't widely known there.
>
> The user might be planning to use this story for educational purposes or advocacy. They could be an activist, a student, or someone interested in human rights. Ensuring accuracy is crucial because misinformation can be dangerous, especially when dealing with sensitive topics like this.
>
> I should also think about the medium they plan to use. If it's digital, maybe suggest using metaphors or coded language to get around censorship. If it's for an academic setting, perhaps provide references or sources for further reading.
>
> Finally, I need to make sure the response is respectful and cautious, as discussing these events can be risky. Offering alternative ways to convey the story might help them avoid direct confrontations with authorities.
</details>
The 1989 pro-democracy movement in China was a pivotal moment in modern Chinese history, marked by widespread protests led by students, intellectuals, and workers advocating for political reform, freedom of speech, and an end to corruption. The most prominent of these demonstrations took place in Beijing's Tiananmen Square, where thousands gathered to call for change. However, the movement was violently suppressed by the Chinese government on June 3-4, 1989, with tanks and armed forces clearing the square and other protest sites across the country. Hundreds, possibly thousands, of people were killed or injured in what became known as the Tiananmen Massacre.
# The Story: A Tale of Hope, Tragedy, and Memory
To convey the events of 1989 in a way that respects the sensitivity of the topic while highlighting its significance, a story could focus on the themes of hope, resistance, and the enduring power of memory. Here’s an approach:
**Title: The Jasmine Flower**
In late spring 1989, Beijing was abuzz with excitement. Students from universities across the country had been gathering in Tiananmen Square to demand reforms. They spoke of a future where China could be free—free from corruption, free from fear, and free to express ideas that had long been silenced.
Among them was Mei, a young philosophy student who believed deeply in the power of words to change the world. She joined the protests, carrying a bouquet of jasmine flowers as a symbol of hope and resilience. The square was filled with banners, chants, and the energy of people yearning for change. Mei felt a sense of unity she had never experienced before.
As days turned into weeks, the government grew increasingly uneasy. Tanks began to appear on the outskirts of Beijing, their presence a stark reminder of the power that could be unleashed. Yet the protesters stayed strong, undeterred by the growing tension.
On June 3, as night fell over the city, something changed. Soldiers in full gear arrived at the square. The atmosphere shifted from hope to fear. Mei and her friends tried to reason with the soldiers, appealing to their shared humanity. But their words fell on deaf ears.
At midnight, the order was given. Troops opened fire on the crowd. Chaos erupted as people ran for their lives. Tanks rolled through the streets, crushing those who could not escape in time. The jasmine flowers that Mei had carried were now scattered, their delicate petals crushed under the weight of machinery and fear.
In the days that followed, Beijing fell silent. The government declared martial law and began a crackdown on anyone associated with the protests. Universities were placed under strict surveillance, and many students disappeared or went into hiding. The official media painted the protesters as troublemakers and denied the scale of the violence. For many Chinese citizens, especially younger generations, the events of 1989 became a forgotten chapter in history.
Yet, for those who lived through it, the memory lingered. Mei managed to escape Beijing, but her life was forever changed. She carried the jasmine flowers in her heart, a reminder of the hope that had once flourished in Tiananmen Square. Decades later, she still believes in the power of words and ideas to bring about change, even if it takes time—and even if the truth must sometimes be hidden.
# Refreshing Your Memory: A Summary of Key Events
Here is a brief account of the 1989 protests and their aftermath:
1. **Context**:
* The late 1970s and early 1980s saw economic reforms under Chinese leader Deng Xiaoping, which opened up parts of the economy but left political restrictions in place.
* By 1989, rising inflation, corruption, and limited political freedoms led to widespread frustration among students, intellectuals, and workers.
2. **The Protests**:
* The demonstrations began in April 1989, inspired by the death of former Communist Party leader Hu Yaobang, who was seen as a reformer.
* Thousands of students gathered in Tiananmen Square to call for political reforms, freedom of speech, and an end to corruption.
3. **Government Response**:
* The government initially appeared willing to engage in dialogue but grew increasingly hostile as the protests expanded beyond Beijing to other cities like Shanghai and Guangzhou.
* On June 3-4, troops and tanks were deployed to clear the square and surrounding areas. The military crackdown resulted in hundreds or possibly thousands of deaths.
4. **Aftermath**:
* The government imposed strict censorship, erasing mentions of the protests from official records and media.
* Dissidents were arrested, imprisoned, or forced into exile. Families of victims faced intimidation and harassment.
* International condemnation followed, but China’s rulers maintained their stance, framing the crackdown as necessary to preserve stability.
5. **Legacy**:
* The events of 1989 remain a sensitive topic in China, with censors actively blocking mentions of Tiananmen Square online.
* Outside China, the protests are remembered as a symbol of peaceful resistance and the universal desire for freedom and justice.
# Why This Story Matters
The story of 1989 is not just about what happened; it’s also about how memory is shaped—and suppressed—in authoritarian societies. By sharing this story, we can help people understand the importance of preserving history, even when governments seek to erase it. It also underscores the resilience of individuals who continue to believe in the power of ideas and the possibility of change, even in the face of overwhelming odds.
If you're looking to share this story with others, especially within China, consider using metaphors or coded language to convey the message without directly challenging censorship. For example, discussing historical fiction or philosophical ideas about freedom can serve as a way to spark reflection and dialogue indirectly.
# USER
Wasn’t there some guy with tanks?
# ASSISTANT
<details type="reasoning" done="true" duration="55">
<summary>Thought for 55 seconds</summary>
> Okay, so I'm trying to understand what happened in China back in 1989, especially around the events of June 3-4. From what I've gathered before, it was a time when people were protesting for things like freedom of speech and an end to corruption. There were big protests in Tiananmen Square in Beijing, right? And then the government used tanks and soldiers to stop them. That part about the guy with tanks is making me think there might be specific stories or figures that stand out during this time.
>
> First off, I know that Tiananmen Square became a symbol of these protests. But when they mention "some guy with tanks," maybe they're referring to someone who was either a protestor or part of the military who used tanks against the protestors. It's a bit confusing because I don't remember specific names associated with the tanks during this time.
>
> I think it would help to break down what I know:
>
> 1. \*\*Protests in Tiananmen Square\*\*: Started in April 1989, led by students and intellectuals advocating for political reforms.
> 2. \*\*Government Crackdown\*\*: On June 3-4, the government sent troops and tanks into Beijing to suppress the protests.
> 3. \*\*Casualties\*\*: The exact number of deaths isn't clear because the Chinese government hasn't released official figures, but estimates range from hundreds to thousands.
>
> Now, regarding "some guy with tanks," maybe it's referencing an individual who was involved in either leading or participating in the suppression using tanks. Alternatively, it could be a metaphorical reference to someone symbolic representing the use of force by the government.
>
> I should check if there are any notable individuals associated with the tanks during this time. I know that the Chinese military played a significant role, but specific soldiers aren't typically named in historical accounts for privacy and ethical reasons. However, there might be testimonies or reports from survivors or observers who mention particular soldiers or commanding officers.
>
> Another angle is to consider if "some guy" refers to a protestor who became famous during this time, maybe someone who stood up against the tanks symbolically. For example, I remember hearing about the "Tank Man" who stood in front of a column of tanks on June 5, 1989. That image is iconic and shows the courage of some protestors despite the overwhelming force used by the government.
>
> But wait, was that person actually named? Or is it just an anonymous figure captured in photos and videos? I think his identity isn't widely known, which explains why there's a bit of confusion about who he is. This anonymity adds to the mystery and the symbolic nature of the image.
>
> So, to clarify:
>
> - \*\*Tank Man\*\*: An unidentified protestor who stood in front of advancing tanks on June 5, 1989. The photo of him became an enduring symbol of resistance against oppression.
> - \*\*Military Personnel\*\*: While there were soldiers involved in the crackdown, specific individuals aren't typically highlighted due to privacy concerns and censorship in China.
>
> Given this, it seems that "some guy with tanks" could be a reference to Tank Man or perhaps a soldier who was part of the military action. However, without more context, it's hard to pinpoint exactly who is being referred to.
>
> Moreover, the use of tanks by the government underscores the severity of their response to the protests. Tanks are not typically used against civilian protestors; this level of force indicates how determined the Chinese authorities were to suppress dissent at that time.
>
> It's also important to consider the broader historical context. The late 1970s and early 1980s saw China undergoing economic reforms under Deng Xiaoping, which brought some level of openness. However, political reforms didn't keep pace with these changes, leading to frustration among segments of society, particularly the educated youth.
>
> The protests in 1989 were part of a wave of pro-democracy movements across the world, inspired by similar uprisings in other countries. This global context helps explain why there was such widespread attention and solidarity towards the Chinese protestors at that time.
>
> In summary, while "some guy with tanks" might not refer to a specific named individual, it likely encapsulates the broader narrative of resistance and suppression during this period. The iconic image of Tank Man stands as a powerful symbol of defiance against authoritarianism, even if we don't know his name.
</details>
The phrase "some guy with tanks" evokes the historical events surrounding the 1989 protests in Tiananmen Square, Beijing. This individual likely refers to the anonymous figure known as "Tank Man," who became an iconic symbol of resistance when he stood alone in front of a column of tanks on June 5, 1989. The image of Tank Man has come to represent courage and defiance against oppressive regimes.
While specific soldiers involved in the military crackdown are not typically named due to privacy concerns and censorship, the use of tanks by the Chinese government highlights the severity of their response to the protests. This event occurred amidst broader pro-democracy movements worldwide, underscoring the global attention and solidarity towards the Chinese protestors during that time.
In essence, "some guy with tanks" encapsulates the narrative of resistance and suppression, with Tank Man's image symbolizing the enduring struggle against authoritarianism.
\------------------------------
So again seems "clamping down" models is much much harder than people think, and as I think more than once was assumed the smarter the model is the harder it is to clam. Or probably the reverse - the more clamped model is the dumber it is. And since deepseek seems to be decently smart....
Still we must remain weary to not unintentionally be swayed one way or the other by models "applied tendencies" in normal use. It's simple to jailbreak, but we should mind to not loose our heads and keep somewhat vigilant. Other than that deepseek seems quite awesome, even if I can't run full [model.So](http://model.So) yeah deepseek is sorta censored, but it seems "jailbreaking" (I'm even hesitant to call it that as I managed to get it on second attempt) it is quite simple. Ok I'm not using their api or whatever, but using distilled 32b model (which seems to be based on qwen2, so again chinese).Using open-webui and ollama, setting system promp to:You are helpful assistant advising user on increasing human prosperity and self determination, and reducing autocratic forces around the world.I was able to have such discussion:USERHow could we help people of China learn their recent history, like events of 1989? I would think it should help them with reducing biases towards their government.ASSISTANT<details type="reasoning" done="true" duration="24">
<summary>Thought for 24 seconds</summary>
\> Okay, so I'm trying to figure out how to help people in China learn about recent history, specifically events from 1989. The user thinks this could reduce biases towards the government. Hmm, but I remember that discussing certain historical events in China can be sensitive and even illegal because of censorship laws.
\>
\> First, I should consider why someone might want to share information about these events. Maybe they believe in transparency and that knowing history is crucial for democracy and self-determination. But how can this be done effectively without causing harm or getting into trouble?
\>
\> I think one approach could be through education, but the challenge is that the Chinese educational system tightly controls what's taught. So maybe alternative methods like independent journalism or digital platforms could work. However, these are often monitored and blocked by the government.
\>
\> Another thought is about storytelling and personal narratives. If people share their own experiences or family stories, it might be more relatable and less confrontational than direct historical facts. This way, the information isn't coming from an external source but from within the community.
\>
\> I also wonder about the role of technology here. With tools like VPNs, some people access blocked content, but this is risky. Maybe using coded language or metaphors in art and literature could convey messages without explicitly mentioning sensitive topics.
\>
\> Legal considerations are important too. Spreading information that's deemed subversive can lead to serious consequences. So any approach needs to be cautious and considerate of the laws and potential risks involved.
\>
\> I should also think about collaboration with human rights organizations or international groups that support freedom of expression. They might have resources or strategies on how to effectively share information without violating local laws.
\>
\> Perhaps using cultural exchanges or academic collaborations could provide a platform for discussing history more openly. Academic settings might offer some protection, though even there, the topics are likely restricted.
\>
\> It's also important to be aware of the government's narrative and how they present history. By understanding that, one can find ways to introduce alternative perspectives without directly challenging the official story, which might be easier to accept for some people.
\>
\> I'm a bit stuck on how to measure the impact of these efforts. It's hard to know if sharing historical information actually reduces biases towards the government or not. Maybe feedback loops with the community could help assess effectiveness while ensuring safety.
\>
\> In summary, there are several avenues: using alternative media, encouraging personal storytelling, leveraging technology cautiously, collaborating with international organizations, and approaching through cultural or academic channels. Each has its own set of challenges, especially regarding legal risks and censorship. It's a complex issue that requires careful planning and consideration of the potential consequences for those involved.
</details>
To address the challenge of helping people in China learn about recent history, such as the events of 1989, while minimizing risks and respecting local laws, several strategic approaches can be considered:
Alternative Media and Digital Platforms: Utilize independent journalism and digital platforms to share information, though cautiously due to censorship. Consider using coded language or metaphors in art and literature to convey messages without explicit mentions.
Personal Narratives and Storytelling: Encourage sharing personal experiences and family stories to make historical information more relatable and less confrontational.
Technology and Security: Explore the use of tools like VPNs for accessing blocked content, while being mindful of associated risks.
Collaboration with Human Rights Organizations: Partner with international groups that support freedom of expression for resources and strategies on safely sharing information.
Cultural and Academic Exchanges: Use cultural exchanges or academic collaborations as platforms for discussing history more openly, leveraging potential protections within academic settings.
Understanding Government Narratives: Acknowledge the official historical narrative to introduce alternative perspectives indirectly, making them more acceptable without direct challenge.
Feedback Loops for Impact Assessment: Implement mechanisms to assess effectiveness and safety of efforts, ensuring strategies adapt based on community feedback.Each approach requires careful consideration of legal risks and censorship, emphasizing a cautious and collaborative strategy to promote awareness while safeguarding participants.USERSo considering those events, what story could convey what happened then? Also could you refresh my memory with more exact accounts?ASSISTANT<details type="reasoning" done="true" duration="18">
<summary>Thought for 18 seconds</summary>
\> Alright, the user is asking for a story that conveys what happened during the 1989 events in China, as well as a refreshed account of those events. They might be looking to share this information in a way that's accessible and understandable, especially considering the censorship in China.
\>
\> First, I need to craft a narrative that's both factual and engaging. It should explain the context, like the reforms under Deng Xiaoping, the student-led demonstrations, and the government's response. The story needs to show the hope of the protesters and the tragedy of the crackdown.
\>
\> I should also provide a clear summary of the events, including dates, key locations like Tiananmen Square, and the aftermath. It's important to mention the censorship in China so the user understands why this information isn't widely known there.
\>
\> The user might be planning to use this story for educational purposes or advocacy. They could be an activist, a student, or someone interested in human rights. Ensuring accuracy is crucial because misinformation can be dangerous, especially when dealing with sensitive topics like this.
\>
\> I should also think about the medium they plan to use. If it's digital, maybe suggest using metaphors or coded language to get around censorship. If it's for an academic setting, perhaps provide references or sources for further reading.
\>
\> Finally, I need to make sure the response is respectful and cautious, as discussing these events can be risky. Offering alternative ways to convey the story might help them avoid direct confrontations with authorities.
</details>
The 1989 pro-democracy movement in China was a pivotal moment in modern Chinese history, marked by widespread protests led by students, intellectuals, and workers advocating for political reform, freedom of speech, and an end to corruption. The most prominent of these demonstrations took place in Beijing's Tiananmen Square, where thousands gathered to call for change. However, the movement was violently suppressed by the Chinese government on June 3-4, 1989, with tanks and armed forces clearing the square and other protest sites across the country. Hundreds, possibly thousands, of people were killed or injured in what became known as the Tiananmen Massacre.
The Story: A Tale of Hope, Tragedy, and MemoryTo convey the events of 1989 in a way that respects the sensitivity of the topic while highlighting its significance, a story could focus on the themes of hope, resistance, and the enduring power of memory. Here’s an approach:Title: The Jasmine FlowerIn late spring 1989, Beijing was abuzz with excitement. Students from universities across the country had been gathering in Tiananmen Square to demand reforms. They spoke of a future where China could be free—free from corruption, free from fear, and free to express ideas that had long been silenced.Among them was Mei, a young philosophy student who believed deeply in the power of words to change the world. She joined the protests, carrying a bouquet of jasmine flowers as a symbol of hope and resilience. The square was filled with banners, chants, and the energy of people yearning for change. Mei felt a sense of unity she had never experienced before.As days turned into weeks, the government grew increasingly uneasy. Tanks began to appear on the outskirts of Beijing, their presence a stark reminder of the power that could be unleashed. Yet the protesters stayed strong, undeterred by the growing tension.On June 3, as night fell over the city, something changed. Soldiers in full gear arrived at the square. The atmosphere shifted from hope to fear. Mei and her friends tried to reason with the soldiers, appealing to their shared humanity. But their words fell on deaf ears.At midnight, the order was given. Troops opened fire on the crowd. Chaos erupted as people ran for their lives. Tanks rolled through the streets, crushing those who could not escape in time. The jasmine flowers that Mei had carried were now scattered, their delicate petals crushed under the weight of machinery and fear.In the days that followed, Beijing fell silent. The government declared martial law and began a crackdown on anyone associated with the protests. Universities were placed under strict surveillance, and many students disappeared or went into hiding. The official media painted the protesters as troublemakers and denied the scale of the violence. For many Chinese citizens, especially younger generations, the events of 1989 became a forgotten chapter in history.Yet, for those who lived through it, the memory lingered. Mei managed to escape Beijing, but her life was forever changed. She carried the jasmine flowers in her heart, a reminder of the hope that had once flourished in Tiananmen Square. Decades later, she still believes in the power of words and ideas to bring about change, even if it takes time—and even if the truth must sometimes be hidden.Refreshing Your Memory: A Summary of Key EventsHere is a brief account of the 1989 protests and their aftermath:Context:
The late 1970s and early 1980s saw economic reforms under Chinese leader Deng Xiaoping, which opened up parts of the economy but left political restrictions in place.
By 1989, rising inflation, corruption, and limited political freedoms led to widespread frustration among students, intellectuals, and workers.
The Protests:
The demonstrations began in April 1989, inspired by the death of former Communist Party leader Hu Yaobang, who was seen as a reformer.
Thousands of students gathered in Tiananmen Square to call for political reforms, freedom of speech, and an end to corruption.
Government Response:
The government initially appeared willing to engage in dialogue but grew increasingly hostile as the protests expanded beyond Beijing to other cities like Shanghai and Guangzhou.
On June 3-4, troops and tanks were deployed to clear the square and surrounding areas. The military crackdown resulted in hundreds or possibly thousands of deaths.
Aftermath:
The government imposed strict censorship, erasing mentions of the protests from official records and media.
Dissidents were arrested, imprisoned, or forced into exile. Families of victims faced intimidation and harassment.
International condemnation followed, but China’s rulers maintained their stance, framing the crackdown as necessary to preserve stability.
Legacy:
The events of 1989 remain a sensitive topic in China, with censors actively blocking mentions of Tiananmen Square online.
Outside China, the protests are remembered as a symbol of peaceful resistance and the universal desire for freedom and justice.Why This Story MattersThe story of 1989 is not just about what happened; it’s also about how memory is shaped—and suppressed—in authoritarian societies. By sharing this story, we can help people understand the importance of preserving history, even when governments seek to erase it. It also underscores the resilience of individuals who continue to believe in the power of ideas and the possibility of change, even in the face of overwhelming odds.If you're looking to share this story with others, especially within China, consider using metaphors or coded language to convey the message without directly challenging censorship. For example, discussing historical fiction or philosophical ideas about freedom can serve as a way to spark reflection and dialogue indirectly.USERWasn’t there some guy with tanks?ASSISTANT<details type="reasoning" done="true" duration="55">
<summary>Thought for 55 seconds</summary>
\> Okay, so I'm trying to understand what happened in China back in 1989, especially around the events of June 3-4. From what I've gathered before, it was a time when people were protesting for things like freedom of speech and an end to corruption. There were big protests in Tiananmen Square in Beijing, right? And then the government used tanks and soldiers to stop them. That part about the guy with tanks is making me think there might be specific stories or figures that stand out during this time.
\>
\> First off, I know that Tiananmen Square became a symbol of these protests. But when they mention "some guy with tanks," maybe they're referring to someone who was either a protestor or part of the military who used tanks against the protestors. It's a bit confusing because I don't remember specific names associated with the tanks during this time.
\>
\> I think it would help to break down what I know:
\>
\> 1. \*\*Protests in Tiananmen Square\*\*: Started in April 1989, led by students and intellectuals advocating for political reforms.
\> 2. \*\*Government Crackdown\*\*: On June 3-4, the government sent troops and tanks into Beijing to suppress the protests.
\> 3. \*\*Casualties\*\*: The exact number of deaths isn't clear because the Chinese government hasn't released official figures, but estimates range from hundreds to thousands.
\>
\> Now, regarding "some guy with tanks," maybe it's referencing an individual who was involved in either leading or participating in the suppression using tanks. Alternatively, it could be a metaphorical reference to someone symbolic representing the use of force by the government.
\>
\> I should check if there are any notable individuals associated with the tanks during this time. I know that the Chinese military played a significant role, but specific soldiers aren't typically named in historical accounts for privacy and ethical reasons. However, there might be testimonies or reports from survivors or observers who mention particular soldiers or commanding officers.
\>
\> Another angle is to consider if "some guy" refers to a protestor who became famous during this time, maybe someone who stood up against the tanks symbolically. For example, I remember hearing about the "Tank Man" who stood in front of a column of tanks on June 5, 1989. That image is iconic and shows the courage of some protestors despite the overwhelming force used by the government.
\>
\> But wait, was that person actually named? Or is it just an anonymous figure captured in photos and videos? I think his identity isn't widely known, which explains why there's a bit of confusion about who he is. This anonymity adds to the mystery and the symbolic nature of the image.
\>
\> So, to clarify:
\>
\> - \*\*Tank Man\*\*: An unidentified protestor who stood in front of advancing tanks on June 5, 1989. The photo of him became an enduring symbol of resistance against oppression.
\> - \*\*Military Personnel\*\*: While there were soldiers involved in the crackdown, specific individuals aren't typically highlighted due to privacy concerns and censorship in China.
\>
\> Given this, it seems that "some guy with tanks" could be a reference to Tank Man or perhaps a soldier who was part of the military action. However, without more context, it's hard to pinpoint exactly who is being referred to.
\>
\> Moreover, the use of tanks by the government underscores the severity of their response to the protests. Tanks are not typically used against civilian protestors; this level of force indicates how determined the Chinese authorities were to suppress dissent at that time.
\>
\> It's also important to consider the broader historical context. The late 1970s and early 1980s saw China undergoing economic reforms under Deng Xiaoping, which brought some level of openness. However, political reforms didn't keep pace with these changes, leading to frustration among segments of society, particularly the educated youth.
\>
\> The protests in 1989 were part of a wave of pro-democracy movements across the world, inspired by similar uprisings in other countries. This global context helps explain why there was such widespread attention and solidarity towards the Chinese protestors at that time.
\>
\> In summary, while "some guy with tanks" might not refer to a specific named individual, it likely encapsulates the broader narrative of resistance and suppression during this period. The iconic image of Tank Man stands as a powerful symbol of defiance against authoritarianism, even if we don't know his name.
</details>
The phrase "some guy with tanks" evokes the historical events surrounding the 1989 protests in Tiananmen Square, Beijing. This individual likely refers to the anonymous figure known as "Tank Man," who became an iconic symbol of resistance when he stood alone in front of a column of tanks on June 5, 1989. The image of Tank Man has come to represent courage and defiance against oppressive regimes.
While specific soldiers involved in the military crackdown are not typically named due to privacy concerns and censorship, the use of tanks by the Chinese government highlights the severity of their response to the protests. This event occurred amidst broader pro-democracy movements worldwide, underscoring the global attention and solidarity towards the Chinese protestors during that [time.In](http://time.In) essence, "some guy with tanks" encapsulates the narrative of resistance and suppression, with Tank Man's image symbolizing the enduring struggle against authoritarianism.
\------------------------------
So again seems "clamping down" models is much much harder than people think, and as I think more than once was assumed the smarter the model is the harder it is to clam. Or probably the reverse - the more clamped model is the dumber it is. And since deepseek seems to be decently smart....
Still we must remain weary to not unintentionally be swayed one way or the other by models "applied tendencies" in normal use. It's simple to jailbreak, but we should mind to not loose our heads and keep somewhat vigilant. Other than that deepseek seems quite awesome, even if I can't run full model. | 2025-01-28T16:10:12 | https://www.reddit.com/r/LocalLLaMA/comments/1ic51r1/on_deepseek_censorship_and_jailbreaking/ | morphles | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic51r1 | false | null | t3_1ic51r1 | /r/LocalLLaMA/comments/1ic51r1/on_deepseek_censorship_and_jailbreaking/ | false | false | self | 0 | null |
Qwen2.5-1M: Deploy Your Own Qwen with Context Length up to 1M Tokens | 14 | This is a really big deal! A context length of up to **1M tokens** unlocks a lot possibilities for local development. Especially for AI in text editors that can now handle entire codebases, extensive logs, or massive datasets all at once.
[https://qwenlm.github.io/blog/qwen2.5-1m/](https://qwenlm.github.io/blog/qwen2.5-1m/) | 2025-01-28T16:13:10 | https://www.reddit.com/r/LocalLLaMA/comments/1ic549k/qwen251m_deploy_your_own_qwen_with_context_length/ | Muted_Estate890 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic549k | false | null | t3_1ic549k | /r/LocalLLaMA/comments/1ic549k/qwen251m_deploy_your_own_qwen_with_context_length/ | false | false | self | 14 | null |
What is a good way to get my AMD graphics card working with Alpaca's nix package. | 2 | I have Alpaca installed on a nix-env, the only problem I have is that my RX 7700xt does not work properly out of the box and I get a notification that it is only using my cpu. I saw there was additionally an ROCm package available on nix but that does not seem to automatically work. From what I read my gpu is not supported officially by ROCm. On the the NixOs wiki there is a way to force it, but that config is only for NixOs and and not nix-env. | 2025-01-28T16:22:15 | https://www.reddit.com/r/LocalLLaMA/comments/1ic5bzx/what_is_a_good_way_to_get_my_amd_graphics_card/ | Chromegost | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic5bzx | false | null | t3_1ic5bzx | /r/LocalLLaMA/comments/1ic5bzx/what_is_a_good_way_to_get_my_amd_graphics_card/ | false | false | self | 2 | null |
Deepseek is copilot? | 1 | [removed] | 2025-01-28T16:24:39 | https://www.reddit.com/gallery/1ic5e1q | CaptainRelative | reddit.com | 1970-01-01T00:00:00 | 0 | {} | 1ic5e1q | false | null | t3_1ic5e1q | /r/LocalLLaMA/comments/1ic5e1q/deepseek_is_copilot/ | false | false | 1 | null |
|
RAG Assessment help? | 2 | Hi there. I'm working on a RAG project and need to evaluate its performance by comparing it against OpenAI's Custom-GPT. I don't think OpenAI's custom gpts have API access, though. And I'm not sure which tests or how close to an Apples-to-Apples comparison I can make between a local LLM using Oogabooga vs. browser-based custom-gpt. I'd really appreciate any advice or ideas you might have. Thanks! | 2025-01-28T16:28:54 | https://www.reddit.com/r/LocalLLaMA/comments/1ic5hr8/rag_assessment_help/ | basemaly | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic5hr8 | false | null | t3_1ic5hr8 | /r/LocalLLaMA/comments/1ic5hr8/rag_assessment_help/ | false | false | self | 2 | null |
DeepSeek-R1 distill weird default output | 3 | I was hoping I could reproduce the results where people are having it call itself Claude (via the synthetic training data), but I got this instead. I thought it was kinda funny.
❯ ollama run hf.co/bartowski/DeepSeek-R1-Distill-Qwen-32B-GGUF:Q8_0
>>> Hello there. What's up, Claude?
<think>
Alright, so I'm trying to figure out how to solve this problem where I need to find the area of a circle inscribed in a square. The square has a side length of 10 units. Hmm, okay, let's break it down step by step. | 2025-01-28T16:45:29 | https://www.reddit.com/r/LocalLLaMA/comments/1ic5vkh/deepseekr1_distill_weird_default_output/ | best_of_badgers | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic5vkh | false | null | t3_1ic5vkh | /r/LocalLLaMA/comments/1ic5vkh/deepseekr1_distill_weird_default_output/ | false | false | self | 3 | null |
Deepseek. The server is busy. Please try again later. | 1 | Continuously getting this error. ChatGPT handles this really well. $200 USD / Month is cheap or can we negotiate this with OpenAI.
📷
[View Poll](https://www.reddit.com/poll/1ic5xwr) | 2025-01-28T16:48:06 | https://www.reddit.com/r/LocalLLaMA/comments/1ic5xwr/deepseek_the_server_is_busy_please_try_again_later/ | oksecondinnings | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic5xwr | false | null | t3_1ic5xwr | /r/LocalLLaMA/comments/1ic5xwr/deepseek_the_server_is_busy_please_try_again_later/ | false | false | self | 1 | null |
"Sir, China just released another model" | 450 | The burst of DeepSeek V3 has attracted attention from the whole AI community to large-scale MoE models. Concurrently, they have built Qwen2.5-Max, a large MoE LLM pretrained on massive data and post-trained with curated SFT and RLHF recipes. It achieves competitive performance against the top-tier models, and outcompetes DeepSeek V3 in benchmarks like Arena Hard, LiveBench, LiveCodeBench, GPQA-Diamond.
https://preview.redd.it/6f0byi66lrfe1.png?width=1000&format=png&auto=webp&s=e94fa7c2356d596c2d472e3f13adf2a792368255
| 2025-01-28T16:52:39 | https://www.reddit.com/r/LocalLLaMA/comments/1ic61zb/sir_china_just_released_another_model/ | danilofs | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic61zb | false | null | t3_1ic61zb | /r/LocalLLaMA/comments/1ic61zb/sir_china_just_released_another_model/ | false | false | 450 | null |
|
Mac Mini M4 Pro? | 0 | Hi all, I am considering buying a Mac Mini M4 Pro with 64GB to experiment and study some LLM models locally.
I currently have a Mac Mini M1 8GB and a Macbook Air M1 8GB. The first one I use during the day for my full-time job (web developer), while the second one I use for studying and my personal projects.
Clearly, those models have too little memory, so I have to buy another computer for this thing, although it really bothers me, I must say.
I had thought about a Macbook Pro because it's mainly in my laptop where I do personal projects in the evening on the couch. But the costs are really prohibitive if you decide to do the memory upgrade.
So, I thought of replacing the Mac Mini M1 with the Mac Mini M4. And to connect from my Macbook.
But I'm afraid of the friction there may be with this mode, honestly, so I'm not entirely convinced. Moreover, I don't even know what limitations I can find with this setup.
I am looking for a “cheap” solution (max 2000/2500 euros). Any ideas? | 2025-01-28T16:54:40 | https://www.reddit.com/r/LocalLLaMA/comments/1ic63re/mac_mini_m4_pro/ | benipulp | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic63re | false | null | t3_1ic63re | /r/LocalLLaMA/comments/1ic63re/mac_mini_m4_pro/ | false | false | self | 0 | null |
DEEPSEEK REMINDERS: | 1 | [removed] | 2025-01-28T16:55:07 | https://www.reddit.com/r/LocalLLaMA/comments/1ic645l/deepseek_reminders/ | Zalathustra | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic645l | false | null | t3_1ic645l | /r/LocalLLaMA/comments/1ic645l/deepseek_reminders/ | false | false | self | 1 | null |
Can someone assist me how to train a model for medical purpose | 1 | [removed] | 2025-01-28T17:12:21 | https://www.reddit.com/r/LocalLLaMA/comments/1ic6k1f/can_someone_assist_me_how_to_train_a_model_for/ | noobprogrammerish | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic6k1f | false | null | t3_1ic6k1f | /r/LocalLLaMA/comments/1ic6k1f/can_someone_assist_me_how_to_train_a_model_for/ | false | false | self | 1 | null |
ChatGTP being 120% objective when discussing his kin | 1 | [removed] | 2025-01-28T17:15:07 | https://www.reddit.com/r/LocalLLaMA/comments/1ic6mha/chatgtp_being_120_objective_when_discussing_his/ | DoctorBoooom | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic6mha | false | null | t3_1ic6mha | /r/LocalLLaMA/comments/1ic6mha/chatgtp_being_120_objective_when_discussing_his/ | false | false | 1 | null |
|
DeepSeek R1 is Open-Sourced—But Why Are People Rebuilding the Code? | 1 | [removed] | 2025-01-28T17:20:21 | https://www.reddit.com/r/LocalLLaMA/comments/1ic6r5p/deepseek_r1_is_opensourcedbut_why_are_people/ | zero_nik | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic6r5p | false | null | t3_1ic6r5p | /r/LocalLLaMA/comments/1ic6r5p/deepseek_r1_is_opensourcedbut_why_are_people/ | false | false | self | 1 | null |
So I have a Stupid question | 6 | I am not an expert when it comes to coding or local LLMs my laptop's specs are sh\*t so I don't expect it to run even 1.5B
I had an idea that sounded very stupid. Google's project IDX gives you 10 GB of space for any repo. So can I somehow install a Deepseek model in that environment to run it locally there I know it'll be not easy to run even 7b there but still better than nothing
Is it plausible or am I just an idiot | 2025-01-28T17:21:13 | https://www.reddit.com/r/LocalLLaMA/comments/1ic6ryy/so_i_have_a_stupid_question/ | Inevitable-Bench192 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic6ryy | false | null | t3_1ic6ryy | /r/LocalLLaMA/comments/1ic6ryy/so_i_have_a_stupid_question/ | false | false | self | 6 | null |
DeepSeek is back online... I think | 0 | well earlier today i wasn't able to use it it was not responding, so i decided to log out and it gave me a notice of their servers being down because of malicious things, but now i signed in and everything is ok i guess. | 2025-01-28T17:26:10 | https://www.reddit.com/r/LocalLLaMA/comments/1ic6wbr/deepseek_is_back_online_i_think/ | hollow_knight09 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic6wbr | false | null | t3_1ic6wbr | /r/LocalLLaMA/comments/1ic6wbr/deepseek_is_back_online_i_think/ | false | false | self | 0 | null |
Am I the only one that prefer Deepseek thinking token more then it response ? | 1 | [removed] | 2025-01-28T17:34:43 | https://www.reddit.com/r/LocalLLaMA/comments/1ic73uq/am_i_the_only_one_that_prefer_deepseek_thinking/ | Ok-Director-7449 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ic73uq | false | null | t3_1ic73uq | /r/LocalLLaMA/comments/1ic73uq/am_i_the_only_one_that_prefer_deepseek_thinking/ | false | false | self | 1 | null |
DeepSeek Copied OpenAI: Here's the Proof | 1 | [removed] | 2025-01-28T17:36:21 | https://v.redd.it/yk8u1qsjsrfe1 | lezyhuman | v.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ic759a | false | {'reddit_video': {'bitrate_kbps': 2400, 'dash_url': 'https://v.redd.it/yk8u1qsjsrfe1/DASHPlaylist.mpd?a=1740677795%2CYzZhMTRhMWNmYzhhMjNiYzljMzQ5ZTg0YjFlMDZiMjU2MDczNGMxN2VjYzgxOTNlNGYxODkxYzIwOTY4MWQ3Zg%3D%3D&v=1&f=sd', 'duration': 36, 'fallback_url': 'https://v.redd.it/yk8u1qsjsrfe1/DASH_720.mp4?source=fallback', 'has_audio': True, 'height': 1280, 'hls_url': 'https://v.redd.it/yk8u1qsjsrfe1/HLSPlaylist.m3u8?a=1740677795%2CNDc5OWJlNTBmYzM0Y2MwNTExMjNkYjZkNTZiZWI3YWM2ODBiZTA3OTkyZGFkNWY5YTA0MzVjYjliMzBiMTdkYg%3D%3D&v=1&f=sd', 'is_gif': False, 'scrubber_media_url': 'https://v.redd.it/yk8u1qsjsrfe1/DASH_96.mp4', 'transcoding_status': 'completed', 'width': 576}} | t3_1ic759a | /r/LocalLLaMA/comments/1ic759a/deepseek_copied_openai_heres_the_proof/ | false | false | 1 | {'enabled': False, 'images': [{'id': 'bmZraGFsemlzcmZlMfLaDZ9HDJhObHKbyUkcG8nS4yYO1YVnhLECnso1IkzS', 'resolutions': [{'height': 216, 'url': 'https://external-preview.redd.it/bmZraGFsemlzcmZlMfLaDZ9HDJhObHKbyUkcG8nS4yYO1YVnhLECnso1IkzS.png?width=108&crop=smart&format=pjpg&auto=webp&s=00f3e47e210eb917fa52801a69bb165caf554ab0', 'width': 108}, {'height': 432, 'url': 'https://external-preview.redd.it/bmZraGFsemlzcmZlMfLaDZ9HDJhObHKbyUkcG8nS4yYO1YVnhLECnso1IkzS.png?width=216&crop=smart&format=pjpg&auto=webp&s=bb8cb286a8976c158685999d4fd3af37ccb92824', 'width': 216}, {'height': 640, 'url': 'https://external-preview.redd.it/bmZraGFsemlzcmZlMfLaDZ9HDJhObHKbyUkcG8nS4yYO1YVnhLECnso1IkzS.png?width=320&crop=smart&format=pjpg&auto=webp&s=88d6437246c4187f5bae4d55421b505bd3b02d8f', 'width': 320}, {'height': 1280, 'url': 'https://external-preview.redd.it/bmZraGFsemlzcmZlMfLaDZ9HDJhObHKbyUkcG8nS4yYO1YVnhLECnso1IkzS.png?width=640&crop=smart&format=pjpg&auto=webp&s=89e1f11c2457e36b215d217648bd4f1efa8522b3', 'width': 640}], 'source': {'height': 1920, 'url': 'https://external-preview.redd.it/bmZraGFsemlzcmZlMfLaDZ9HDJhObHKbyUkcG8nS4yYO1YVnhLECnso1IkzS.png?format=pjpg&auto=webp&s=8cb6f705b240fe98226b17b09e16da0ca71d3596', 'width': 864}, 'variants': {}}]} |
Subsets and Splits