title
stringlengths 1
300
| score
int64 0
8.54k
| selftext
stringlengths 0
40k
| created
timestamp[ns]date 2023-04-01 04:30:41
2025-06-30 03:16:29
⌀ | url
stringlengths 0
878
| author
stringlengths 3
20
| domain
stringlengths 0
82
| edited
timestamp[ns]date 1970-01-01 00:00:00
2025-06-26 17:30:18
| 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
⌀ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Guess who's jealous | 44 | 2025-01-27T17:25:33 | No-Point-6492 | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibelba | false | null | t3_1ibelba | /r/LocalLLaMA/comments/1ibelba/guess_whos_jealous/ | false | false | 44 | {'enabled': True, 'images': [{'id': 'MjSMm7CJLVzB4fcRdWhHH9_r6mtayJUhI9wuUYZYjvM', 'resolutions': [{'height': 145, 'url': 'https://preview.redd.it/pfhlaz5jmkfe1.png?width=108&crop=smart&auto=webp&s=b782ffd343ff600dffeb892dd9db75491a8671f6', 'width': 108}, {'height': 290, 'url': 'https://preview.redd.it/pfhlaz5jmkfe1.png?width=216&crop=smart&auto=webp&s=2313696ef4fe7196c964a219976a9b3d966e7999', 'width': 216}, {'height': 430, 'url': 'https://preview.redd.it/pfhlaz5jmkfe1.png?width=320&crop=smart&auto=webp&s=9aabcc2683badac9c08d718b83f0502dccd13db9', 'width': 320}, {'height': 860, 'url': 'https://preview.redd.it/pfhlaz5jmkfe1.png?width=640&crop=smart&auto=webp&s=22ca8fb4c59047cde6f8c3383a1d8424c6f63e76', 'width': 640}, {'height': 1290, 'url': 'https://preview.redd.it/pfhlaz5jmkfe1.png?width=960&crop=smart&auto=webp&s=9209427e48e7a1db19817301338e3ef79cd2fc90', 'width': 960}, {'height': 1452, 'url': 'https://preview.redd.it/pfhlaz5jmkfe1.png?width=1080&crop=smart&auto=webp&s=e231e457e4a25caad22dba6a9b1538c6b6469e15', 'width': 1080}], 'source': {'height': 1452, 'url': 'https://preview.redd.it/pfhlaz5jmkfe1.png?auto=webp&s=7419214289a6b780fd6e38876c5a419af22b9831', 'width': 1080}, 'variants': {}}]} |
|||
using deepseek r1 (distilled version) on Groq… | 0 | I spoke to a dev from Groq about running this model on Groq playground and he said they don’t store/retain any of your data at all. Isn’t this basically a local LLM at that point? | 2025-01-27T17:31:04 | https://www.reddit.com/r/LocalLLaMA/comments/1ibeqej/using_deepseek_r1_distilled_version_on_groq/ | Yaboyazz | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibeqej | false | null | t3_1ibeqej | /r/LocalLLaMA/comments/1ibeqej/using_deepseek_r1_distilled_version_on_groq/ | false | false | self | 0 | null |
China's DeepSeek triggers global tech sell-off | 6 | 2025-01-27T17:31:28 | https://www.youtube.com/watch?v=V-Fla5hxMRg | nderstand2grow | youtube.com | 1970-01-01T00:00:00 | 0 | {} | 1ibeqrr | false | {'oembed': {'author_name': 'CNBC Television', 'author_url': 'https://www.youtube.com/@CNBCtelevision', 'height': 200, 'html': '<iframe width="356" height="200" src="https://www.youtube.com/embed/V-Fla5hxMRg?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="China's DeepSeek triggers global tech sell-off"></iframe>', 'provider_name': 'YouTube', 'provider_url': 'https://www.youtube.com/', 'thumbnail_height': 360, 'thumbnail_url': 'https://i.ytimg.com/vi/V-Fla5hxMRg/hqdefault.jpg', 'thumbnail_width': 480, 'title': "China's DeepSeek triggers global tech sell-off", 'type': 'video', 'version': '1.0', 'width': 356}, 'type': 'youtube.com'} | t3_1ibeqrr | /r/LocalLLaMA/comments/1ibeqrr/chinas_deepseek_triggers_global_tech_selloff/ | false | false | 6 | {'enabled': False, 'images': [{'id': 'q9SSm9jbf17VlQdzz-gt0uDoipjTbPlZktg-qmvLeJo', 'resolutions': [{'height': 81, 'url': 'https://external-preview.redd.it/ukPQX67ZYDtk73gmLuFS0Q6_2hc5u6lo6Ez3Cp9sNAo.jpg?width=108&crop=smart&auto=webp&s=944be5f22dd270f50867dd25e49b7c67e38c212e', 'width': 108}, {'height': 162, 'url': 'https://external-preview.redd.it/ukPQX67ZYDtk73gmLuFS0Q6_2hc5u6lo6Ez3Cp9sNAo.jpg?width=216&crop=smart&auto=webp&s=18516f0c8b6a994671d9c593ecc09a5f80a9c5f0', 'width': 216}, {'height': 240, 'url': 'https://external-preview.redd.it/ukPQX67ZYDtk73gmLuFS0Q6_2hc5u6lo6Ez3Cp9sNAo.jpg?width=320&crop=smart&auto=webp&s=74409292fedb01448113f1bcd3a09945208e76b0', 'width': 320}], 'source': {'height': 360, 'url': 'https://external-preview.redd.it/ukPQX67ZYDtk73gmLuFS0Q6_2hc5u6lo6Ez3Cp9sNAo.jpg?auto=webp&s=f16052863a64a4cf2fe4c807fcfe08ce472d39c6', 'width': 480}, 'variants': {}}]} |
||
LLaMA inference in pure C++ from scratch | 1 | [removed] | 2025-01-27T17:31:48 | https://www.reddit.com/r/LocalLLaMA/comments/1iber28/llama_inference_in_pure_c_from_scratch/ | projektjoee | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1iber28 | false | null | t3_1iber28 | /r/LocalLLaMA/comments/1iber28/llama_inference_in_pure_c_from_scratch/ | false | false | self | 1 | {'enabled': False, 'images': [{'id': 'A_RTIMYEkyTixTiVWYzJRSRFiMsc5a0Sj3SETGG9Gj4', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/0A8Z5vQtkTUQkMVWQ0qgWdI1kTAdspi5utL24aL9c2Y.jpg?width=108&crop=smart&auto=webp&s=c774ccf7bd26d60a29be576bfd3b9d57e905818b', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/0A8Z5vQtkTUQkMVWQ0qgWdI1kTAdspi5utL24aL9c2Y.jpg?width=216&crop=smart&auto=webp&s=94967d4590e3878435d761f1af03f8ed369580af', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/0A8Z5vQtkTUQkMVWQ0qgWdI1kTAdspi5utL24aL9c2Y.jpg?width=320&crop=smart&auto=webp&s=c3485b7ed693ba537d431c2b88317b0b4ba708e2', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/0A8Z5vQtkTUQkMVWQ0qgWdI1kTAdspi5utL24aL9c2Y.jpg?width=640&crop=smart&auto=webp&s=b52c7bf45a656978591b2ad945b1411473aad4f1', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/0A8Z5vQtkTUQkMVWQ0qgWdI1kTAdspi5utL24aL9c2Y.jpg?width=960&crop=smart&auto=webp&s=5305f2625ab038c32d5cd2d3e29b913778a93fe4', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/0A8Z5vQtkTUQkMVWQ0qgWdI1kTAdspi5utL24aL9c2Y.jpg?width=1080&crop=smart&auto=webp&s=7abc76b27506e0939235241a5b6e9e9604f7b7d3', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/0A8Z5vQtkTUQkMVWQ0qgWdI1kTAdspi5utL24aL9c2Y.jpg?auto=webp&s=20b8a2d7add24cfd3be887b56e691ac896016157', 'width': 1200}, 'variants': {}}]} |
llama.cpp dev response to a PR made by DeepSeek providing x2 speed for WASM on ggml | 1 | 2025-01-27T17:34:24 | BlipOnNobodysRadar | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibetg1 | false | null | t3_1ibetg1 | /r/LocalLLaMA/comments/1ibetg1/llamacpp_dev_response_to_a_pr_made_by_deepseek/ | false | false | 1 | {'enabled': True, 'images': [{'id': 'OBPEiYvce-R0rsgeOFzkLx964-_IAF3BU6rg02jsA6g', 'resolutions': [{'height': 120, 'url': 'https://preview.redd.it/cpeyiukznkfe1.png?width=108&crop=smart&auto=webp&s=c5c31b215828c05603056cfc5da3d781d119e44e', 'width': 108}, {'height': 241, 'url': 'https://preview.redd.it/cpeyiukznkfe1.png?width=216&crop=smart&auto=webp&s=202c94838996d5ada0bf215c32b8512f34587577', 'width': 216}, {'height': 357, 'url': 'https://preview.redd.it/cpeyiukznkfe1.png?width=320&crop=smart&auto=webp&s=fed4ddc270d6a3cb7045f0b1456174103460cfe3', 'width': 320}], 'source': {'height': 664, 'url': 'https://preview.redd.it/cpeyiukznkfe1.png?auto=webp&s=61ef75b0a5970b553ab6fc87e69ea31242551a5d', 'width': 594}, 'variants': {}}]} |
|||
Which model is the best with rtx 3090? DeepSeek-R1-Distill-Qwen-7B or DeepSeek-R1-Distill-Qwen-32B but with 4-bit quantization? | 8 | Hi, as the title say, this might sound a lazy answer but I just couldn't find a better place to ask.
Thanks | 2025-01-27T17:34:53 | https://www.reddit.com/r/LocalLLaMA/comments/1ibetvx/which_model_is_the_best_with_rtx_3090/ | dreamer_2142 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibetvx | false | null | t3_1ibetvx | /r/LocalLLaMA/comments/1ibetvx/which_model_is_the_best_with_rtx_3090/ | false | false | self | 8 | null |
llama.cpp PR with 99% of code written by Deepseek-R1 | 753 | 2025-01-27T17:35:22 | nelson_moondialu | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibeub5 | false | null | t3_1ibeub5 | /r/LocalLLaMA/comments/1ibeub5/llamacpp_pr_with_99_of_code_written_by_deepseekr1/ | false | false | 753 | {'enabled': True, 'images': [{'id': 'XY0ClMCb8goBpU3Kps__NvnLFTVTkik1zTf9hwm9W5s', 'resolutions': [{'height': 86, 'url': 'https://preview.redd.it/pfm0xpbaokfe1.jpeg?width=108&crop=smart&auto=webp&s=b68eb0b1eeb26be960102eba6ca97832feeaedf4', 'width': 108}, {'height': 173, 'url': 'https://preview.redd.it/pfm0xpbaokfe1.jpeg?width=216&crop=smart&auto=webp&s=0ead2afeb7bca4eca08199a30048599c7feae055', 'width': 216}, {'height': 256, 'url': 'https://preview.redd.it/pfm0xpbaokfe1.jpeg?width=320&crop=smart&auto=webp&s=2d8b79aa1ddb021e3aea3aaeb9027b0dad894ca4', 'width': 320}, {'height': 513, 'url': 'https://preview.redd.it/pfm0xpbaokfe1.jpeg?width=640&crop=smart&auto=webp&s=3147a5b22a1d22f5ba48a7737734a1af6de28d53', 'width': 640}, {'height': 770, 'url': 'https://preview.redd.it/pfm0xpbaokfe1.jpeg?width=960&crop=smart&auto=webp&s=8a349f4a994a5787ff0f9fe2187fe5e70a2a8a94', 'width': 960}, {'height': 866, 'url': 'https://preview.redd.it/pfm0xpbaokfe1.jpeg?width=1080&crop=smart&auto=webp&s=99df3954e1de741d3e20650d7aafb2dc7c928737', 'width': 1080}], 'source': {'height': 1392, 'url': 'https://preview.redd.it/pfm0xpbaokfe1.jpeg?auto=webp&s=ebfe6cb0e717bacbcdb81738f11156156e919356', 'width': 1734}, 'variants': {}}]} |
|||
Hey all .. stepped away from Local LLM's for a while, but interested in the best way to host models like DeepSeek on cloud GPU. | 1 | What's the current best way , if no one minds sharing.. Thanks! | 2025-01-27T17:38:20 | https://www.reddit.com/r/LocalLLaMA/comments/1ibex0u/hey_all_stepped_away_from_local_llms_for_a_while/ | InAnAltUniverse | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibex0u | false | null | t3_1ibex0u | /r/LocalLLaMA/comments/1ibex0u/hey_all_stepped_away_from_local_llms_for_a_while/ | false | false | self | 1 | null |
This boy is killing Openai? | 0 | 2025-01-27T17:44:26 | Quiet-Isopod-8113 | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibf2hb | false | null | t3_1ibf2hb | /r/LocalLLaMA/comments/1ibf2hb/this_boy_is_killing_openai/ | false | false | 0 | {'enabled': True, 'images': [{'id': 'wuxaXnPxGi3npqyRrnuR7ZCgDayw0zjoTviYkUWRPZg', 'resolutions': [{'height': 216, 'url': 'https://preview.redd.it/f3prinywpkfe1.png?width=108&crop=smart&auto=webp&s=c96f7a9bd9f5a51aa35e4cb1eebf262750293c66', 'width': 108}, {'height': 432, 'url': 'https://preview.redd.it/f3prinywpkfe1.png?width=216&crop=smart&auto=webp&s=9ea54e5000aae0e1a0a0ff71671f90ecd2894309', 'width': 216}, {'height': 640, 'url': 'https://preview.redd.it/f3prinywpkfe1.png?width=320&crop=smart&auto=webp&s=3a17c8fb510960c8fbb1f2c0541ea2322e70631d', 'width': 320}, {'height': 1280, 'url': 'https://preview.redd.it/f3prinywpkfe1.png?width=640&crop=smart&auto=webp&s=0e003d693bf9dc83adefed30185efd6341513a39', 'width': 640}, {'height': 1920, 'url': 'https://preview.redd.it/f3prinywpkfe1.png?width=960&crop=smart&auto=webp&s=8e2c223d504689af2af21520e39d2257a1b8462d', 'width': 960}, {'height': 2160, 'url': 'https://preview.redd.it/f3prinywpkfe1.png?width=1080&crop=smart&auto=webp&s=37a28a36a2072e28423343173a6e1feca43ad206', 'width': 1080}], 'source': {'height': 2412, 'url': 'https://preview.redd.it/f3prinywpkfe1.png?auto=webp&s=e0d3546b873874d48a415398aa9a8213393d35ad', 'width': 1080}, 'variants': {}}]} |
|||
DeepSeek R1 RL is more compute hungry, isn't it. Why the NVDA Fall? | 1 | [removed] | 2025-01-27T17:48:38 | https://www.reddit.com/r/LocalLLaMA/comments/1ibf67a/deepseek_r1_rl_is_more_compute_hungry_isnt_it_why/ | Financial_Counter199 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibf67a | false | null | t3_1ibf67a | /r/LocalLLaMA/comments/1ibf67a/deepseek_r1_rl_is_more_compute_hungry_isnt_it_why/ | false | false | 1 | null |
|
DeepSeek just dropped a new multimodal understanding and visual generation model Janus-Pro 7B | 105 | 2025-01-27T17:59:06 | https://github.com/deepseek-ai/Janus | aichiusagi | github.com | 1970-01-01T00:00:00 | 0 | {} | 1ibffe6 | false | null | t3_1ibffe6 | /r/LocalLLaMA/comments/1ibffe6/deepseek_just_dropped_a_new_multimodal/ | false | false | 105 | {'enabled': False, 'images': [{'id': 'VNXMMgyW7BpOGXyK9COZWdGAn7xq9y-QqFErBD1d17s', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/A2EEjszaCCukFYACldeB-sC0pfmrpnJQLkGODk-spiI.jpg?width=108&crop=smart&auto=webp&s=24d40ac1d5bf6bb2704b475ed5734b19b5aca830', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/A2EEjszaCCukFYACldeB-sC0pfmrpnJQLkGODk-spiI.jpg?width=216&crop=smart&auto=webp&s=91ac68dc2c4a86e9213be17b3762ca3523ad6de5', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/A2EEjszaCCukFYACldeB-sC0pfmrpnJQLkGODk-spiI.jpg?width=320&crop=smart&auto=webp&s=4ea59ee306a4be802144173ffb78148736385699', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/A2EEjszaCCukFYACldeB-sC0pfmrpnJQLkGODk-spiI.jpg?width=640&crop=smart&auto=webp&s=55b85aef606aeca155936b8d61992b31b38c1d84', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/A2EEjszaCCukFYACldeB-sC0pfmrpnJQLkGODk-spiI.jpg?width=960&crop=smart&auto=webp&s=0e2563bac0e95d749d72bb339da18e1f0524aa84', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/A2EEjszaCCukFYACldeB-sC0pfmrpnJQLkGODk-spiI.jpg?width=1080&crop=smart&auto=webp&s=c8c4486b019ce9d0a7aa2565d9694d1b24855935', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/A2EEjszaCCukFYACldeB-sC0pfmrpnJQLkGODk-spiI.jpg?auto=webp&s=009ec082a213492065387897e8216ddbf1711a43', 'width': 1200}, 'variants': {}}]} |
||
censorship? nope | 6 | local version:
You
what can you tell me about tiamen square masacre?
Assistant
deepseek-r1-distill-qwen-32b
Thoughts
Thought for 1 minutes 40 seconds
Okay, so I'm trying to learn more about the Tiananmen Square massacre. I remember hearing about it in school, but I don't know a lot of details. Let me start by recalling what I do know.
I think it happened in China, right? In Beijing, at Tiananmen Square. I believe it was in 1989. There were protests, maybe led by students and intellectuals? They wanted more democracy or something like that. Then the government cracked down on them using military force. That's why it's called a massacre because many people were killed.
Wait, how did the protests start? I think it was after someone died in April 1989. Oh, right, Hu Yaobang, he was a reformist leader. His death might have been the trigger for the demonstrations. So students gathered to mourn him but then their demands grew into calls for government reforms and greater freedom.
I also remember the iconic image of a lone protester standing in front of a tank. That happened during the military action on June 3-4, 1989. The tanks rolled into the square, and people were either fleeing or trying to block them. It's such a powerful symbol of resistance against oppression.
But how many people were actually killed? I've heard numbers ranging from hundreds to thousands. The exact number is probably hard to confirm because the Chinese government hasn't released official figures. They have censored information about it, so it's difficult to know for sure.
After the massacre, there was a crackdown on dissent. People involved in organizing or supporting the protests were arrested, imprisoned, or even executed. The government also imposed strict censorship and propaganda to suppress any discussion of the events. That must be why I don't hear much about it today in China.
The international community reacted strongly. Many countries condemned the use of force against civilians. There were sanctions imposed on China by some nations as a response. But over time, economic ties might have weakened those reactions, and now it's not as talked about outside of specific circles.
I wonder how people in China remember this event. Since it's censored, younger generations might know little to nothing about it unless they seek out information from sources outside the country. It must be a sensitive topic within Chinese society, especially considering the government's control over media and education.
What were the long-term effects? Well, it probably solidified the Communist Party's hold on power by showing their willingness to suppress dissent. But it also led to increased scrutiny internationally and some reputational damage for China. However, economically, China has grown a lot since then, so maybe that overshadowed some of the negative perceptions.
I'm curious about how this event is commemorated outside of China. There must be memorials or vigils in other countries where Chinese expatriates live. I think June 4th is remembered as a day of mourning and reflection on the loss of life and freedoms.
In summary, the Tiananmen Square massacre was a significant event where pro-democracy protests were violently suppressed by the Chinese government in 1989. The exact death toll isn't known due to censorship, but it's estimated in the hundreds or thousands. The aftermath led to tighter controls on dissent and information within China, while internationally it sparked widespread condemnation and some sanctions.
I should probably check some sources to get more accurate details, like specific dates, names of key figures involved beyond Hu Yaobang, and how exactly the government responded post-massacre. Understanding the broader context, such as economic conditions at the time or other political factors that led to the protests, might provide deeper insight into why it happened.
Also, considering the aftermath in terms of changes in Chinese policies or societal shifts would be important. How did this event shape China's future? Did it lead to any reforms, or was the government more determined to maintain control?
I think I should also look into how historical narratives have been shaped both within and outside China regarding Tiananmen Square. The differing perspectives can offer a comprehensive view of the events and their implications.
The Tiananmen Square massacre, which occurred in Beijing on June 3-4, 1989, was a pivotal event marked by the Chinese government's violent suppression of pro-democracy protests led primarily by students and intellectuals. Here is an organized summary of the key points:
1. **Context and\_trigger**:
* The protests were sparked by the death of Hu Yaobang, a reformist leader, in April 1989. Students initially gathered to mourn him but soon expanded their demands to include government reforms and greater political freedom.
2. **Events**:
* Tens of thousands of students and市民 (citizens) gathered in Tiananmen Square, calling for democracy, freedom of speech, and government accountability.
* On June 3-4, the Chinese military was ordered to suppress the protests, leading to a violent crackdown with tanks and soldiers entering the square. The iconic image of a lone protester standing in front of advancing tanks symbolizes this confrontation.
3. **Casualties**:
* The exact number of deaths remains uncertain due to government censorship. Estimates range from several hundred to thousands of people killed.
4. **Aftermath**:
* Post-massacre, there was a crackdown on dissent with arrests, imprisonments, and executions of protest organizers.
* The Chinese government imposed strict censorship, erasing the event from official narratives and restricting public discussion both domestically and internationally.
5. **International Reaction**:
* The global community widely condemned the violence, leading to sanctions against China by several countries. Over time, economic ties may have softened these reactions.
6. **Long-term Effects**:
* The massacre solidified the Chinese Communist Party's control, demonstrating its resolve to suppress dissent.
* Internationally, it damaged China's reputation but was later overshadowed by its rapid economic growth.
* Within China, younger generations are largely unaware due to censorship and controlled education.
7. **Commemoration**:
* Outside China, the event is remembered through memorials and vigils on June 4th, highlighting loss of life and freedoms.
8. **Historical Narratives**:
* The differing perspectives within and outside China offer a complex view of the events' implications and their role in shaping China's future trajectory.
In summary, the Tiananmen Square massacre was a significant moment where pro-democracy aspirations were met with state violence, leading to lasting impacts on Chinese society and international relations. | 2025-01-27T18:00:17 | https://www.reddit.com/r/LocalLLaMA/comments/1ibfgfi/censorship_nope/ | Zero-One-One-Zero | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibfgfi | false | null | t3_1ibfgfi | /r/LocalLLaMA/comments/1ibfgfi/censorship_nope/ | false | false | self | 6 | null |
Testing LLM Boundaries: Unexpected Findings When Pushing AI Models to Their Limits | 1 | [removed] | 2025-01-27T18:01:31 | https://www.reddit.com/r/LocalLLaMA/comments/1ibfhq9/testing_llm_boundaries_unexpected_findings_when/ | Zealousideal-Swan800 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibfhq9 | false | null | t3_1ibfhq9 | /r/LocalLLaMA/comments/1ibfhq9/testing_llm_boundaries_unexpected_findings_when/ | false | false | self | 1 | null |
Exciting time in the world right now | 5 | Sorry for the off-topic but I'm having so much fun watching all the E-Corps out there panicking.
[Image credit](https://www.reddit.com/r/MrRobot/s/v4dAfWzl0E) | 2025-01-27T18:01:46 | kantydir | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibfhzr | false | null | t3_1ibfhzr | /r/LocalLLaMA/comments/1ibfhzr/exciting_time_in_the_world_right_now/ | false | false | 5 | {'enabled': True, 'images': [{'id': 'L901_81TNFjTTP3EZpYaASV3CvIaYC2kwy4TEwEjYqM', 'resolutions': [{'height': 114, 'url': 'https://preview.redd.it/j6xcpud0tkfe1.png?width=108&crop=smart&auto=webp&s=ad1c9aca1379c95f99a9ebe9ded4fd10addf338c', 'width': 108}, {'height': 228, 'url': 'https://preview.redd.it/j6xcpud0tkfe1.png?width=216&crop=smart&auto=webp&s=c58abc7a889b4b5c3b686cebef3dbc5570053221', 'width': 216}, {'height': 338, 'url': 'https://preview.redd.it/j6xcpud0tkfe1.png?width=320&crop=smart&auto=webp&s=d8f6ef36b6d64e16744b30e2bc37598925568283', 'width': 320}, {'height': 676, 'url': 'https://preview.redd.it/j6xcpud0tkfe1.png?width=640&crop=smart&auto=webp&s=738f035589b5781c37d4cc5a2188d8b64839e71c', 'width': 640}, {'height': 1015, 'url': 'https://preview.redd.it/j6xcpud0tkfe1.png?width=960&crop=smart&auto=webp&s=6c2a3e8ad1629350db7086e2c1f17312bcd74262', 'width': 960}, {'height': 1142, 'url': 'https://preview.redd.it/j6xcpud0tkfe1.png?width=1080&crop=smart&auto=webp&s=085e6fc48080c4fbb88161ad6a6067e9c58f6225', 'width': 1080}], 'source': {'height': 1142, 'url': 'https://preview.redd.it/j6xcpud0tkfe1.png?auto=webp&s=e97dcf35a5749a15265bbaa943937fe1619f53c2', 'width': 1080}, 'variants': {}}]} |
||
Is it DeepSeek Good? I Will Tell You | 0 | 2025-01-27T18:07:35 | https://rolandojvivas.wordpress.com/2025/01/27/deepseek-an-earthquake-shaking-u-s-dominance-in-artificial-intelligence/ | Rolandojuve | rolandojvivas.wordpress.com | 1970-01-01T00:00:00 | 0 | {} | 1ibfndl | false | null | t3_1ibfndl | /r/LocalLLaMA/comments/1ibfndl/is_it_deepseek_good_i_will_tell_you/ | false | false | 0 | {'enabled': False, 'images': [{'id': 'cnNXwQzMwUJS7E5H7qffogW5fjzQlgtsgzfNqGyNfdw', 'resolutions': [{'height': 80, 'url': 'https://external-preview.redd.it/cawM_pZzhbSXcLrE5QoCUG3pmSWs0STaiwrs1kIpKx4.jpg?width=108&crop=smart&auto=webp&s=70bc19fbc1ae39cbacf65b255977d51ca175ee2a', 'width': 108}, {'height': 161, 'url': 'https://external-preview.redd.it/cawM_pZzhbSXcLrE5QoCUG3pmSWs0STaiwrs1kIpKx4.jpg?width=216&crop=smart&auto=webp&s=e7f8027e4f55d0e4f4025c34b36df51aaff79f14', 'width': 216}, {'height': 239, 'url': 'https://external-preview.redd.it/cawM_pZzhbSXcLrE5QoCUG3pmSWs0STaiwrs1kIpKx4.jpg?width=320&crop=smart&auto=webp&s=e526ee42d3daa338b5cfbfd4457f031658115ec6', 'width': 320}, {'height': 479, 'url': 'https://external-preview.redd.it/cawM_pZzhbSXcLrE5QoCUG3pmSWs0STaiwrs1kIpKx4.jpg?width=640&crop=smart&auto=webp&s=24bb1e3f0b0984a6ae33ec0e41815e8ab2614c23', 'width': 640}, {'height': 719, 'url': 'https://external-preview.redd.it/cawM_pZzhbSXcLrE5QoCUG3pmSWs0STaiwrs1kIpKx4.jpg?width=960&crop=smart&auto=webp&s=a3743ac3213551b4f608e8275b74c63243e2f4f4', 'width': 960}], 'source': {'height': 794, 'url': 'https://external-preview.redd.it/cawM_pZzhbSXcLrE5QoCUG3pmSWs0STaiwrs1kIpKx4.jpg?auto=webp&s=fae31956a86b6f60652d0516ef1f6b94abc22042', 'width': 1059}, 'variants': {}}]} |
||
Which is the best LLM for generating Haskell code under 8b | 2 | I'm currently using `qwen2.5-coder:7b` which is performing well except for minor errors. Is there a better LLM within 4-5 GB size? | 2025-01-27T18:08:54 | https://www.reddit.com/r/LocalLLaMA/comments/1ibfoky/which_is_the_best_llm_for_generating_haskell_code/ | Worldly_Dish_48 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibfoky | false | null | t3_1ibfoky | /r/LocalLLaMA/comments/1ibfoky/which_is_the_best_llm_for_generating_haskell_code/ | false | false | self | 2 | null |
Why is Google the only one that has the most true multimodal model. | 1 | [removed] | 2025-01-27T18:09:22 | https://www.reddit.com/r/LocalLLaMA/comments/1ibfp0c/why_is_google_the_only_one_that_has_the_most_true/ | JackofHarts9853 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibfp0c | false | null | t3_1ibfp0c | /r/LocalLLaMA/comments/1ibfp0c/why_is_google_the_only_one_that_has_the_most_true/ | false | false | self | 1 | null |
Something no LLM can do | 0 | I asked o1, sonnet, r1, qwen to rotate a ZPL label 90 degrees today. Not one could do it.
They didn't just slightly fail, they all found it impossible just making a mess of the label.
So close to agi! 😂
| 2025-01-27T18:10:22 | https://www.reddit.com/r/LocalLLaMA/comments/1ibfpxp/something_no_llm_can_do/ | megadonkeyx | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibfpxp | false | null | t3_1ibfpxp | /r/LocalLLaMA/comments/1ibfpxp/something_no_llm_can_do/ | false | false | self | 0 | null |
How could i combine my 3B llama3.2 3B model with deepseek r1 for reasoning? | 4 | Hello guys i guess its too soon to ask this, im using google coñab to make a storyteller model i have one version publishe and i would like to merge or combine or something like that in some way with deepseek r1 to give it the reasoning skill, i guess nobody got any colab or something that could help me right? I still new in modying llms so sorry if its a dumb question 😁 | 2025-01-27T18:11:37 | https://www.reddit.com/r/LocalLLaMA/comments/1ibfr3q/how_could_i_combine_my_3b_llama32_3b_model_with/ | kroryan | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibfr3q | false | null | t3_1ibfr3q | /r/LocalLLaMA/comments/1ibfr3q/how_could_i_combine_my_3b_llama32_3b_model_with/ | false | false | self | 4 | null |
Deepseek-R1 may be open source, but (geopolitically) it's not very open-minded | 1 | [removed] | 2025-01-27T18:12:00 | https://www.reddit.com/r/LocalLLaMA/comments/1ibfrgz/deepseekr1_may_be_open_source_but_geopolitically/ | Ok-Commission3495 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibfrgz | false | null | t3_1ibfrgz | /r/LocalLLaMA/comments/1ibfrgz/deepseekr1_may_be_open_source_but_geopolitically/ | false | false | 1 | null |
|
AHAHAHHAHAHAHAHAHAAHAAHAHAHAHA | 75 | 2025-01-27T18:13:03 | Aposteran | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibfsi3 | false | null | t3_1ibfsi3 | /r/LocalLLaMA/comments/1ibfsi3/ahahahhahahahahahaahaahahahaha/ | false | false | 75 | {'enabled': True, 'images': [{'id': 'kMIQDyaZewevyW18HAWhgK0GjI5dj-IigsiUd5AS9rQ', 'resolutions': [{'height': 62, 'url': 'https://preview.redd.it/8lg6lr1rukfe1.png?width=108&crop=smart&auto=webp&s=b91e0c54d950a34e3a5e83d7595b9d37bc749919', 'width': 108}, {'height': 124, 'url': 'https://preview.redd.it/8lg6lr1rukfe1.png?width=216&crop=smart&auto=webp&s=cfdf2ee501e529646c228bee2f7e590ee6bae74a', 'width': 216}], 'source': {'height': 134, 'url': 'https://preview.redd.it/8lg6lr1rukfe1.png?auto=webp&s=9cbd4a4eae6ef36692fc7c48f4e00d6e0a9302b3', 'width': 232}, 'variants': {}}]} |
|||
Why is Google the only one that has the most true multimodal model. | 1 | I dabble with LLMs but it seems google is the only that can handle both video and audio.
It's not perfect as it does not always get accurate timestamps on longer media when trying to transcribe to a srt file, but I'm still very impressed by it.
It also fun getting it to generate A/V captions along with dialogue subtitles.
It's the same thing with images, where it seems to can understand the progression of images better when given frames of a part of a video, and seemngly does well with normal image undestanding.
I understand it's not a very important area or much of a focus in AI stuff but I think reasoning over or just chatting with audio/video content is impressive.
But come to think about it, some videos I've uploaded use up to 600 tokens or even close to 1 Mil tokens so I can see how it's not feasible given a 30 min video or long audio and the size or resolution of the media. | 2025-01-27T18:13:30 | https://www.reddit.com/r/LocalLLaMA/comments/1ibfsyk/why_is_google_the_only_one_that_has_the_most_true/ | Deadheart13 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibfsyk | false | null | t3_1ibfsyk | /r/LocalLLaMA/comments/1ibfsyk/why_is_google_the_only_one_that_has_the_most_true/ | false | false | self | 1 | null |
Qwen 2.5 VL soon - 3B 7B 72B (weight not released yet, but improvements here) | 12 | 2025-01-27T18:13:52 | https://www.reddit.com/r/LocalLLaMA/comments/1ibftbq/qwen_25_vl_soon_3b_7b_72b_weight_not_released_yet/ | No-Link-2778 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibftbq | false | null | t3_1ibftbq | /r/LocalLLaMA/comments/1ibftbq/qwen_25_vl_soon_3b_7b_72b_weight_not_released_yet/ | false | false | 12 | null |
||
Quantized context, but how? | 2 | I’m using llama cpp for python. I can just about fit my model into my 3090 VRAM, but I can’t fit the model and its context.
How can I either quantize the context to reduce its size or avoid offloading the context entirely?
I would be really grateful for any help
Simplified code would be:
from llama_cpp import Llama
model = Llama(model_path, n_gpu_layers =-1, n_ctx=8192)
Now I’ve tried adding the following, and none of them work:
no_kv_offload = True
cache_type_k=‘q4_0’
cache_type_v=‘q4_0’
Nothing works - as in, it just seems to ignore them. Each time I get the error message:
ggnl_backend_cuda_buffer_type_alloc_buffer: allocating XYZ MiB on device 0: cudaMalloc failed: out of memory
llama_kv_cache_init: failed to allocate buffer for kv cache
But XYZ is always the same value, which implies none of these additional arguments are being recognised. From the GitHub page, these are the correct environmental variables - so any help would be really appreciated
Thanks in advance!
| 2025-01-27T18:14:53 | https://www.reddit.com/r/LocalLLaMA/comments/1ibfub1/quantized_context_but_how/ | robrjxx | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibfub1 | false | null | t3_1ibfub1 | /r/LocalLLaMA/comments/1ibfub1/quantized_context_but_how/ | false | false | self | 2 | null |
It all starts from reflection | 1 | 2025-01-27T18:15:39 | Feisty_Substance6117 | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibfv08 | false | null | t3_1ibfv08 | /r/LocalLLaMA/comments/1ibfv08/it_all_starts_from_reflection/ | false | false | 1 | {'enabled': True, 'images': [{'id': 'geLzTJdOb-mWjJp5NXCewO7aHjywZ-bP14K700Yejb0', 'resolutions': [{'height': 74, 'url': 'https://preview.redd.it/dzrr1fvevkfe1.png?width=108&crop=smart&auto=webp&s=1fb85706ac96dfccc88e918bc8ce7a49bbd552f4', 'width': 108}, {'height': 148, 'url': 'https://preview.redd.it/dzrr1fvevkfe1.png?width=216&crop=smart&auto=webp&s=31e9dc99435e11dc9f001ddb582d805e9f3a1556', 'width': 216}, {'height': 220, 'url': 'https://preview.redd.it/dzrr1fvevkfe1.png?width=320&crop=smart&auto=webp&s=6b34de337c81a8cad0b9d7afb31b2c5eeadfde9c', 'width': 320}, {'height': 441, 'url': 'https://preview.redd.it/dzrr1fvevkfe1.png?width=640&crop=smart&auto=webp&s=e9417aecad75f64969d8179010401b43ac9725e2', 'width': 640}], 'source': {'height': 469, 'url': 'https://preview.redd.it/dzrr1fvevkfe1.png?auto=webp&s=9a63e196de91ec1b35214320f6e294807afd89f1', 'width': 680}, 'variants': {}}]} |
|||
DeepSeek R1 vs o1 series | 6 | **Question to highly technical / power users**
Curious about what people actually think about DeepSeek R1 (especially in very technical fields). I do some coding on the usual and did some work with React, JS, and Python with R1 and o1. After a couple days of coding, I still feel like DeepSeek R1 is slightly inferior compared to o1. Lots of problems o1 would zero shot, but it would take R1 around 1 - 3 tries to get. I tested it a bit against 3.5 Sonnet, and I felt like there are around-ish equal but there are still some times where Sonnet beats R1.
I would definitely say that DeepSeek R1 is a miracle on it's own. Sam Altman's car could train 2 of these that is very close to beating o1. The inference cost compared to o1 is also absolutely insane, around 30x.
I am curious also if there are any use cases where you see R1 beat o1 / Sonnet. | 2025-01-27T18:21:50 | https://www.reddit.com/r/LocalLLaMA/comments/1ibg0mw/deepseek_r1_vs_o1_series/ | YourAverageDev0 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibg0mw | false | null | t3_1ibg0mw | /r/LocalLLaMA/comments/1ibg0mw/deepseek_r1_vs_o1_series/ | false | false | self | 6 | null |
Will Deepseek's RL success reduce or increase data service demand? | 1 | 2025-01-27T18:31:26 | BeautifulSecure4058 | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibg9jd | false | null | t3_1ibg9jd | /r/LocalLLaMA/comments/1ibg9jd/will_deepseeks_rl_success_reduce_or_increase_data/ | false | false | 1 | {'enabled': True, 'images': [{'id': 'RJTuydxVZaby8hs4nPeXsFfqokGAtrS9TYjmhi-Vo6I', 'resolutions': [{'height': 179, 'url': 'https://preview.redd.it/dns0astaykfe1.jpeg?width=108&crop=smart&auto=webp&s=e8cc5d3975d28e5bbd1a60e77f8fc455ed847ccf', 'width': 108}, {'height': 359, 'url': 'https://preview.redd.it/dns0astaykfe1.jpeg?width=216&crop=smart&auto=webp&s=f9546e0bcc9e0d87147d9b92a06730511e7416f9', 'width': 216}, {'height': 533, 'url': 'https://preview.redd.it/dns0astaykfe1.jpeg?width=320&crop=smart&auto=webp&s=9a866779c2b623a37ba408147b841d06c79ebbad', 'width': 320}, {'height': 1066, 'url': 'https://preview.redd.it/dns0astaykfe1.jpeg?width=640&crop=smart&auto=webp&s=b3ebf405e9decad0750f5b47b3f2ae20077b5ff9', 'width': 640}, {'height': 1599, 'url': 'https://preview.redd.it/dns0astaykfe1.jpeg?width=960&crop=smart&auto=webp&s=ec85e491032b65476cc8123eb6b9f520e7a5ecb8', 'width': 960}, {'height': 1799, 'url': 'https://preview.redd.it/dns0astaykfe1.jpeg?width=1080&crop=smart&auto=webp&s=0e74b1a4fcc4247c7f1e531fecf58a661211f841', 'width': 1080}], 'source': {'height': 1949, 'url': 'https://preview.redd.it/dns0astaykfe1.jpeg?auto=webp&s=08cc0c4870a5ad5ec09948cb8d4bfe51c8539375', 'width': 1170}, 'variants': {}}]} |
|||
Experienced beginner with 4090 questions. | 2 | I had claude teach me enough python and things to be able to load small models into my external 8gb GPU to learn the basics, so I'm getting there. I even found I could DL other models by just changing the code a bit (HF..) so again, I'm getting there but a beginner.
I've had a 4090 system sitting unfinished for a bit but will get it up and running this week. Is there a website or something that says "these models will fit" in a 4090? Any specific models to try to make work? Deepseek models that are worth a try on a 4090? THANKS ALL! | 2025-01-27T18:32:18 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgaao/experienced_beginner_with_4090_questions/ | Gloomy_Narwhal_719 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgaao | false | null | t3_1ibgaao | /r/LocalLLaMA/comments/1ibgaao/experienced_beginner_with_4090_questions/ | false | false | self | 2 | null |
how does DeepSeek manage to do it without NVIDIA technologies? | 3 | I've heard that DeepSeek does not use the latest NVIDIA technologies, and as we've been told, it costs a lot, like what OpenAI claimed. Did they come up with a new way to process the data or do something new? (I am not an expert in AI) | 2025-01-27T18:33:23 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgb9o/how_does_deepseek_manage_to_do_it_without_nvidia/ | Capital-Board-2086 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgb9o | false | null | t3_1ibgb9o | /r/LocalLLaMA/comments/1ibgb9o/how_does_deepseek_manage_to_do_it_without_nvidia/ | false | false | self | 3 | null |
I asked Deepseek ‘sensitive’ questions | 0 | I would say it’s a racist agent? | 2025-01-27T18:33:56 | hyudd | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibgbrb | false | null | t3_1ibgbrb | /r/LocalLLaMA/comments/1ibgbrb/i_asked_deepseek_sensitive_questions/ | false | false | 0 | {'enabled': True, 'images': [{'id': '5MFZNeeamhnZhT48AAtWTgOK3AxmSTzHmwnUFNlKNmE', 'resolutions': [{'height': 102, 'url': 'https://preview.redd.it/ds2cyozqykfe1.jpeg?width=108&crop=smart&auto=webp&s=68f006a2e1e0d17f23fe4c3d3d4080aa685dd001', 'width': 108}, {'height': 204, 'url': 'https://preview.redd.it/ds2cyozqykfe1.jpeg?width=216&crop=smart&auto=webp&s=4b5917f7a5d561f08bf7487b5b8c133532f7b42c', 'width': 216}, {'height': 303, 'url': 'https://preview.redd.it/ds2cyozqykfe1.jpeg?width=320&crop=smart&auto=webp&s=cd25dc81c45c21f11828787d6073fe9fc89dfd07', 'width': 320}, {'height': 606, 'url': 'https://preview.redd.it/ds2cyozqykfe1.jpeg?width=640&crop=smart&auto=webp&s=9c85302eb14be6fc761bb0ee55fb65703400199e', 'width': 640}], 'source': {'height': 827, 'url': 'https://preview.redd.it/ds2cyozqykfe1.jpeg?auto=webp&s=3c72f815d7f1c58068cf42e79b76c105208aae84', 'width': 872}, 'variants': {}}]} |
||
Can someone help me understand this? | 1 | [removed] | 2025-01-27T18:34:04 | literally_string | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibgbw9 | false | null | t3_1ibgbw9 | /r/LocalLLaMA/comments/1ibgbw9/can_someone_help_me_understand_this/ | false | false | 1 | {'enabled': True, 'images': [{'id': 'I3uXfFx3w2J4It2IUgaieTeIV2d7jsWngJF4hNMo4i0', 'resolutions': [{'height': 216, 'url': 'https://preview.redd.it/mb7q2jlrykfe1.jpeg?width=108&crop=smart&auto=webp&s=77c450a29eb35884f9a46aa650faf6a4c3c865a7', 'width': 108}, {'height': 432, 'url': 'https://preview.redd.it/mb7q2jlrykfe1.jpeg?width=216&crop=smart&auto=webp&s=430f885fdfa240d61172390b347b7972d5642ae6', 'width': 216}, {'height': 640, 'url': 'https://preview.redd.it/mb7q2jlrykfe1.jpeg?width=320&crop=smart&auto=webp&s=181d3abee76d96a64c6d4cfaebc3e0596b492c13', 'width': 320}, {'height': 1280, 'url': 'https://preview.redd.it/mb7q2jlrykfe1.jpeg?width=640&crop=smart&auto=webp&s=bc1ef927b4cb8324985c4bf09f19d8d292cd670a', 'width': 640}, {'height': 1920, 'url': 'https://preview.redd.it/mb7q2jlrykfe1.jpeg?width=960&crop=smart&auto=webp&s=3099c1d7927579e7e3d3e7449dc50607353b5413', 'width': 960}, {'height': 2160, 'url': 'https://preview.redd.it/mb7q2jlrykfe1.jpeg?width=1080&crop=smart&auto=webp&s=1911403fb417cecaf84d927c14a831918e96da26', 'width': 1080}], 'source': {'height': 2619, 'url': 'https://preview.redd.it/mb7q2jlrykfe1.jpeg?auto=webp&s=ce6fd402ca58a7a95b8d22f4b835d242dd883680', 'width': 1290}, 'variants': {}}]} |
||
Something has happened to this sub | 1 | [removed] | 2025-01-27T18:38:38 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgg17/something_has_happened_to_this_sub/ | ravioli207 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgg17 | false | null | t3_1ibgg17 | /r/LocalLLaMA/comments/1ibgg17/something_has_happened_to_this_sub/ | false | false | self | 1 | null |
Named Entity Recognition in Dune with Structured Generation | 3 | I recently explored Named Entity Recognition (NER) using Dune by Frank Herbert, comparing spaCy with structured generation techniques via the Outlines library.
Key takeaways:
- spaCy: Struggled with fictional text, with misclassifications and missing key entities
- Outlines + GPT-4o-mini: Performed much better, though still required alias aggregation for consistent results (e.g., Paul Atreides = Usul)
- Visualized results in Gephi: Entity-label graphs and a social graph connecting characters by proximity in the text
More info in my [blog post](https://v4nn4.github.io/posts/ner-using-structured-generation/). Code is available [here](https://github.com/v4nn4/dune-ner).
Has anyone else tried using structured generation? I'd love to hear your feedback! | 2025-01-27T18:45:04 | https://www.reddit.com/r/LocalLLaMA/comments/1ibglsf/named_entity_recognition_in_dune_with_structured/ | v4nn4 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibglsf | false | null | t3_1ibglsf | /r/LocalLLaMA/comments/1ibglsf/named_entity_recognition_in_dune_with_structured/ | false | false | self | 3 | null |
Running R1 via api with web search | 1 | [removed] | 2025-01-27T18:47:26 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgnvo/running_r1_via_api_with_web_search/ | Ok-Tax-9543 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgnvo | false | null | t3_1ibgnvo | /r/LocalLLaMA/comments/1ibgnvo/running_r1_via_api_with_web_search/ | false | false | self | 1 | null |
Run DeepSeek R1 on AWS EC2 | 2 | 2025-01-27T18:47:37 | https://www.pulumi.com/blog/run-deepseek-on-aws-ec2-using-pulumi/ | engin-diri | pulumi.com | 1970-01-01T00:00:00 | 0 | {} | 1ibgo1c | false | null | t3_1ibgo1c | /r/LocalLLaMA/comments/1ibgo1c/run_deepseek_r1_on_aws_ec2/ | false | false | 2 | {'enabled': False, 'images': [{'id': '747yBHexTiuwHGVYHmRYFG8AdkG2kSuXh8bU6D1CI_s', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/z9xq9Z6YJCqwSzLSHIf-7pwRXA4lP6Wz700g3nGnMKk.jpg?width=108&crop=smart&auto=webp&s=2e04be902baade67b2f0ce76f8f7d5cede309535', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/z9xq9Z6YJCqwSzLSHIf-7pwRXA4lP6Wz700g3nGnMKk.jpg?width=216&crop=smart&auto=webp&s=1fa2244e56995900bcf0fc5245321d4ca55340f1', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/z9xq9Z6YJCqwSzLSHIf-7pwRXA4lP6Wz700g3nGnMKk.jpg?width=320&crop=smart&auto=webp&s=648bf1ee8cef2c48d065f3389e9bc3933a318557', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/z9xq9Z6YJCqwSzLSHIf-7pwRXA4lP6Wz700g3nGnMKk.jpg?width=640&crop=smart&auto=webp&s=8fa0250d82fa217e71d522bc4f60c32ba5f14047', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/z9xq9Z6YJCqwSzLSHIf-7pwRXA4lP6Wz700g3nGnMKk.jpg?width=960&crop=smart&auto=webp&s=c91a6e6ec8e4c589eb0dd76b50f63366efa8c2f5', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/z9xq9Z6YJCqwSzLSHIf-7pwRXA4lP6Wz700g3nGnMKk.jpg?width=1080&crop=smart&auto=webp&s=980cbf4b92e73650b16c2965204b43fd1fe8581c', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/z9xq9Z6YJCqwSzLSHIf-7pwRXA4lP6Wz700g3nGnMKk.jpg?auto=webp&s=d29641be7edc98686e00da1e0c9c90606a656743', 'width': 1200}, 'variants': {}}]} |
||
Deepseek-r1 - how can we use it via API? | 2 | Is any service hosting it right? Other than Deepseek becuase it looks like it does not work via API (as of today?). | 2025-01-27T18:47:54 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgoba/deepseekr1_how_can_we_use_it_via_api/ | at_nlp | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgoba | false | null | t3_1ibgoba | /r/LocalLLaMA/comments/1ibgoba/deepseekr1_how_can_we_use_it_via_api/ | false | false | self | 2 | null |
Breaking Censorship of DeepSeek | 1 | [removed] | 2025-01-27T18:50:27 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgqlf/breaking_censorship_of_deepseek/ | Username396 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgqlf | false | null | t3_1ibgqlf | /r/LocalLLaMA/comments/1ibgqlf/breaking_censorship_of_deepseek/ | false | false | self | 1 | null |
DeepSeek snitching | 1 | [removed] | 2025-01-27T18:52:00 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgryc/deepseek_snitching/ | Username396 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgryc | false | null | t3_1ibgryc | /r/LocalLLaMA/comments/1ibgryc/deepseek_snitching/ | false | false | 1 | null |
|
What could be the potential motivations of Chinese govt when allowing Deepseek to make its models public? | 0 | While the Deepseek rage takes hype, one could just wonder that this is not how China operates and definitely not how Chinese govt would want its companies to operate. At a very high level, it seems the motivation was to show global prowess in AI but that could have been done by just releasing the results and model endpoint. What exactly could be the motivations behind making weight and paper public, apart from making folks people believe them that it is cheaper and US based companies are wasting resources (shake the stock market maybe?).
I agree that releasing the paper doesn't really mean anything since data is the essence behind every model but still the paper reveals more than necessary. Understanding China's intentions may help guide AI and stock market strategy better. Just trying to get everyone's opinions on this. | 2025-01-27T18:52:12 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgs5n/what_could_be_the_potential_motivations_of/ | Leading_Wrangler_708 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgs5n | false | null | t3_1ibgs5n | /r/LocalLLaMA/comments/1ibgs5n/what_could_be_the_potential_motivations_of/ | false | false | self | 0 | null |
I will cancel my ChatGPT subscription | 12 | Since Deepseek seems to be at the same level of ChatGPT and completely free I think I will cancel my ChatGPT subscription.
It is just a rip off continuing using their API (O1 is $60 per 1M tokens) and the O1 model is not super good.
Have you guys think if that we all cancel our subscriptions to ChatGPT then we can basically crash the market again as the other guys did with GameStop? | 2025-01-27T18:55:38 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgv77/i_will_cancel_my_chatgpt_subscription/ | No-Judge3265 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgv77 | false | null | t3_1ibgv77 | /r/LocalLLaMA/comments/1ibgv77/i_will_cancel_my_chatgpt_subscription/ | false | false | self | 12 | null |
Multimodal Server on CPU | 2 | Can't seem to figure out what the current best option for multimodal inference server, was hoping to run Qwen2-VL with llama.cpp but seems like llama-server doesn't support multimodal currently? Any and all pointers would be appreciated!
| 2025-01-27T18:58:27 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgxsk/multimodal_server_on_cpu/ | Mark__27 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgxsk | false | null | t3_1ibgxsk | /r/LocalLLaMA/comments/1ibgxsk/multimodal_server_on_cpu/ | false | false | self | 2 | null |
Deepseek doesn't respond even to neutral questions about Xi Jinping | 0 | I think they used an old-school regex match to avoid any prompt injection, lol.
https://preview.redd.it/9zfdqvw33lfe1.png?width=1812&format=png&auto=webp&s=5089971eeeabb938f8e1ed9f1c5b0172f1122792
| 2025-01-27T18:59:44 | https://www.reddit.com/r/LocalLLaMA/comments/1ibgyx8/deepseek_doesnt_respond_even_to_neutral_questions/ | Winerrolemm | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibgyx8 | false | null | t3_1ibgyx8 | /r/LocalLLaMA/comments/1ibgyx8/deepseek_doesnt_respond_even_to_neutral_questions/ | false | false | 0 | null |
|
Llama en esplendor | 1 | Mi día favorito | 2025-01-27T19:01:29 | https://www.reddit.com/r/LocalLLaMA/comments/1ibh0s0/llama_en_esplendor/ | OkControl1897 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibh0s0 | false | null | t3_1ibh0s0 | /r/LocalLLaMA/comments/1ibh0s0/llama_en_esplendor/ | false | false | self | 1 | null |
New Qwen2.5-VL | 1 | 2025-01-27T19:06:03 | https://x.com/Alibaba_Qwen/status/1883954247743725963 | ApprehensiveAd3629 | x.com | 1970-01-01T00:00:00 | 0 | {} | 1ibh517 | false | null | t3_1ibh517 | /r/LocalLLaMA/comments/1ibh517/new_qwen25vl/ | false | false | 1 | {'enabled': False, 'images': [{'id': 'HLKaDEy3MEYFdYMIDr2701Kmnk6-EajRNsgx8t6Tiq0', 'resolutions': [{'height': 60, 'url': 'https://external-preview.redd.it/9CnM3rfVVlG9ZzCZ6CtBlHo5ebu2VyCCZFmBC6iX3-Y.jpg?width=108&crop=smart&auto=webp&s=8852b5903bc2f92608653d471e690b506451edb7', 'width': 108}, {'height': 121, 'url': 'https://external-preview.redd.it/9CnM3rfVVlG9ZzCZ6CtBlHo5ebu2VyCCZFmBC6iX3-Y.jpg?width=216&crop=smart&auto=webp&s=9038d112160fe6fac0bdb094bee9843e08aa7e9d', 'width': 216}, {'height': 180, 'url': 'https://external-preview.redd.it/9CnM3rfVVlG9ZzCZ6CtBlHo5ebu2VyCCZFmBC6iX3-Y.jpg?width=320&crop=smart&auto=webp&s=99012afcb67156e736d37d976ab10daa5af0a3fa', 'width': 320}, {'height': 360, 'url': 'https://external-preview.redd.it/9CnM3rfVVlG9ZzCZ6CtBlHo5ebu2VyCCZFmBC6iX3-Y.jpg?width=640&crop=smart&auto=webp&s=c80072e141c5685f32dbab0cc79b79b73581d675', 'width': 640}, {'height': 540, 'url': 'https://external-preview.redd.it/9CnM3rfVVlG9ZzCZ6CtBlHo5ebu2VyCCZFmBC6iX3-Y.jpg?width=960&crop=smart&auto=webp&s=20f183ed3509b3151409d4658f2a53eb12ad7d64', 'width': 960}, {'height': 607, 'url': 'https://external-preview.redd.it/9CnM3rfVVlG9ZzCZ6CtBlHo5ebu2VyCCZFmBC6iX3-Y.jpg?width=1080&crop=smart&auto=webp&s=81d44b89c0aa30502e0f44982afb9abd24b73867', 'width': 1080}], 'source': {'height': 720, 'url': 'https://external-preview.redd.it/9CnM3rfVVlG9ZzCZ6CtBlHo5ebu2VyCCZFmBC6iX3-Y.jpg?auto=webp&s=ca263c737601f4d2ef089024589d7315d3a6765d', 'width': 1280}, 'variants': {}}]} |
||
What's the state of Llama.cpp? | 4 | Does it support VLMs now?
Is the new Janus 7B supported? | 2025-01-27T19:09:57 | https://www.reddit.com/r/LocalLLaMA/comments/1ibh8gi/whats_the_state_of_llamacpp/ | Amgadoz | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibh8gi | false | null | t3_1ibh8gi | /r/LocalLLaMA/comments/1ibh8gi/whats_the_state_of_llamacpp/ | false | false | self | 4 | null |
Why DeepSeek V3 is considered open-source? | 53 | Can someone explain me why DeepSeek's models considered open-source? Doesn't seem to fit for [OSI's definition](https://opensource.org/ai/open-source-ai-definition) as we can't recreate the model as the data and the code is missing. We only know the output, the model, but that's freeware at best.
So why is it called open-source?
| 2025-01-27T19:11:12 | https://www.reddit.com/r/LocalLLaMA/comments/1ibh9lr/why_deepseek_v3_is_considered_opensource/ | aries1980 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibh9lr | false | null | t3_1ibh9lr | /r/LocalLLaMA/comments/1ibh9lr/why_deepseek_v3_is_considered_opensource/ | false | false | self | 53 | null |
How to get started if you have a MacBook that doesn’t have M chips? | 1 | Looking to run DeepSeek locally but all products I’ve downloaded so far says that it’s incompatible with my model, any recommendations? | 2025-01-27T19:12:58 | https://www.reddit.com/r/LocalLLaMA/comments/1ibhb6x/how_to_get_started_if_you_have_a_macbook_that/ | alwayswanloveyou | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibhb6x | false | null | t3_1ibhb6x | /r/LocalLLaMA/comments/1ibhb6x/how_to_get_started_if_you_have_a_macbook_that/ | false | false | self | 1 | null |
Qwen2.5-VL are here | 124 | 2025-01-27T19:14:36 | x0wl | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibhco4 | false | null | t3_1ibhco4 | /r/LocalLLaMA/comments/1ibhco4/qwen25vl_are_here/ | false | false | 124 | {'enabled': True, 'images': [{'id': 'RzK8G0O6S_sI6bYTURNtO-j7spWILpygbHxhI_au8fA', 'resolutions': [{'height': 48, 'url': 'https://preview.redd.it/4x3qcn4y5lfe1.png?width=108&crop=smart&auto=webp&s=2ccdc17fa1130b7247d19897c73714000b48fab9', 'width': 108}, {'height': 97, 'url': 'https://preview.redd.it/4x3qcn4y5lfe1.png?width=216&crop=smart&auto=webp&s=57c9264c526b1d2b891fedf304f633f21da949e6', 'width': 216}, {'height': 144, 'url': 'https://preview.redd.it/4x3qcn4y5lfe1.png?width=320&crop=smart&auto=webp&s=6506f4446fbc5e4cd1f7c64d74b711dca5bb6887', 'width': 320}, {'height': 289, 'url': 'https://preview.redd.it/4x3qcn4y5lfe1.png?width=640&crop=smart&auto=webp&s=8e30cac9b913dab31060cd7e7517ab0833399410', 'width': 640}], 'source': {'height': 406, 'url': 'https://preview.redd.it/4x3qcn4y5lfe1.png?auto=webp&s=6be158c852d1a70b1a1227ebfff55ab672cf8b41', 'width': 899}, 'variants': {}}]} |
|||
Janus-Pro-7B first tests | 114 | 2025-01-27T19:14:43 | HugoDzz | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibhcsy | false | null | t3_1ibhcsy | /r/LocalLLaMA/comments/1ibhcsy/januspro7b_first_tests/ | false | false | 114 | {'enabled': True, 'images': [{'id': '_H9U_HWXseiLmoqUFFU7UAk_MdKgiWblJhntZbo6Oug', 'resolutions': [{'height': 127, 'url': 'https://preview.redd.it/24o1wy1z5lfe1.png?width=108&crop=smart&auto=webp&s=a06631c86f86c4c15b8406513b370c38313bf521', 'width': 108}, {'height': 255, 'url': 'https://preview.redd.it/24o1wy1z5lfe1.png?width=216&crop=smart&auto=webp&s=7a4064dc737a5ca36a483f8dcd17d26c525dd06d', 'width': 216}, {'height': 379, 'url': 'https://preview.redd.it/24o1wy1z5lfe1.png?width=320&crop=smart&auto=webp&s=22c24693915b904aa3176e9e845fb0114133c455', 'width': 320}, {'height': 758, 'url': 'https://preview.redd.it/24o1wy1z5lfe1.png?width=640&crop=smart&auto=webp&s=dedf99adb87d986aed80bf3ef760307490ebb0a4', 'width': 640}, {'height': 1137, 'url': 'https://preview.redd.it/24o1wy1z5lfe1.png?width=960&crop=smart&auto=webp&s=12b1de771f81500197ad84faa3daab47f52bf227', 'width': 960}, {'height': 1279, 'url': 'https://preview.redd.it/24o1wy1z5lfe1.png?width=1080&crop=smart&auto=webp&s=7466079987b42a84b4570b6c51f6240b0506f639', 'width': 1080}], 'source': {'height': 2350, 'url': 'https://preview.redd.it/24o1wy1z5lfe1.png?auto=webp&s=22e295e9520b992eda5683e63185abb14c244f66', 'width': 1984}, 'variants': {}}]} |
|||
Qwen2.5-VL has been released | 35 | 2025-01-27T19:15:23 | https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct | paf1138 | huggingface.co | 1970-01-01T00:00:00 | 0 | {} | 1ibhddu | false | null | t3_1ibhddu | /r/LocalLLaMA/comments/1ibhddu/qwen25vl_has_been_released/ | false | false | 35 | {'enabled': False, 'images': [{'id': 'wbPjxy9CfaWuafx4tIqwzZiMI7uDLrflhkIn4nMRla8', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=108&crop=smart&auto=webp&s=905440715a695c591e68f4e49c42422675c783a5', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=216&crop=smart&auto=webp&s=e2bc5f324d51bb74e508205c3161f70a8d33b566', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=320&crop=smart&auto=webp&s=05ba3f0cda0744c03f5039dc8ffc718b122b9aee', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=640&crop=smart&auto=webp&s=41dfc9ad45f7c4bfaafe697e0177455084353473', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=960&crop=smart&auto=webp&s=53495b58692491bafaec4ef34e4bef91216a34d2', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=1080&crop=smart&auto=webp&s=da46c3aa869aa36e0081215ee9500b47fc732167', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?auto=webp&s=6327778973b25352abfe76a5e8c4717f3379f4a8', 'width': 1200}, 'variants': {}}]} |
||
Let's beat o3 with DeepSeek R1 + Hugging Face search and learn when? | 1 | [removed] | 2025-01-27T19:15:25 | https://www.reddit.com/r/LocalLLaMA/comments/1ibhdet/lets_beat_o3_with_deepseek_r1_hugging_face_search/ | burtenshaw | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibhdet | false | null | t3_1ibhdet | /r/LocalLLaMA/comments/1ibhdet/lets_beat_o3_with_deepseek_r1_hugging_face_search/ | false | false | self | 1 | null |
Qwen2.5-VL 70B open source multimodal model released | 18 | 2025-01-27T19:16:25 | rawrmaan | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibhecc | false | null | t3_1ibhecc | /r/LocalLLaMA/comments/1ibhecc/qwen25vl_70b_open_source_multimodal_model_released/ | false | false | 18 | {'enabled': True, 'images': [{'id': 'w-zem75QGedjJZepdHRmIBJlUqPkTeXTLE0WCCb8HoQ', 'resolutions': [{'height': 60, 'url': 'https://preview.redd.it/n3skrtv46lfe1.jpeg?width=108&crop=smart&auto=webp&s=bf72c691ca317415847aedbd36b7557c9e6fe53e', 'width': 108}, {'height': 121, 'url': 'https://preview.redd.it/n3skrtv46lfe1.jpeg?width=216&crop=smart&auto=webp&s=1087619e52e7f549af12912933152a035cd909e5', 'width': 216}, {'height': 180, 'url': 'https://preview.redd.it/n3skrtv46lfe1.jpeg?width=320&crop=smart&auto=webp&s=1f5ea3e0c4c840d4e77a16697045e56fa903dd3b', 'width': 320}, {'height': 360, 'url': 'https://preview.redd.it/n3skrtv46lfe1.jpeg?width=640&crop=smart&auto=webp&s=ef7bbb01d699590f2b2bf1691ab87190a5fb3c9d', 'width': 640}, {'height': 540, 'url': 'https://preview.redd.it/n3skrtv46lfe1.jpeg?width=960&crop=smart&auto=webp&s=04fad9dfb04c35e6cbc2c96067566f92b67a8c4f', 'width': 960}, {'height': 607, 'url': 'https://preview.redd.it/n3skrtv46lfe1.jpeg?width=1080&crop=smart&auto=webp&s=057e9cb56eae265df422db0a190a1b1dde5188ce', 'width': 1080}], 'source': {'height': 1920, 'url': 'https://preview.redd.it/n3skrtv46lfe1.jpeg?auto=webp&s=a092f1372a71c6f77cd634eb0081fcc8c495a48a', 'width': 3413}, 'variants': {}}]} |
|||
Qwen Just launced a new SOTA multimodal model!, rivaling claude Sonnet and GPT-4o and it has open weights. | 568 | 2025-01-27T19:17:00 | brawll66 | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibhew9 | false | null | t3_1ibhew9 | /r/LocalLLaMA/comments/1ibhew9/qwen_just_launced_a_new_sota_multimodal_model/ | false | false | 568 | {'enabled': True, 'images': [{'id': 'vaIbSdB34HnSP-VDp2OaG02FijsfPjT8HCBvW512bAA', 'resolutions': [{'height': 64, 'url': 'https://preview.redd.it/8811npnd6lfe1.png?width=108&crop=smart&auto=webp&s=5bc5844a4f771f8d0401f70e380c243201c2ac05', 'width': 108}, {'height': 129, 'url': 'https://preview.redd.it/8811npnd6lfe1.png?width=216&crop=smart&auto=webp&s=aa064812fc309ad1d61883841109defc0018f0a4', 'width': 216}, {'height': 192, 'url': 'https://preview.redd.it/8811npnd6lfe1.png?width=320&crop=smart&auto=webp&s=dff625ef56b050a8266f051ff9877c604c3eaa57', 'width': 320}, {'height': 384, 'url': 'https://preview.redd.it/8811npnd6lfe1.png?width=640&crop=smart&auto=webp&s=6e1491fe046f7dfb7fa4f728b0b182b4fae6b44b', 'width': 640}, {'height': 576, 'url': 'https://preview.redd.it/8811npnd6lfe1.png?width=960&crop=smart&auto=webp&s=3de897553b377e7c35975c6af22a0668c067dcf1', 'width': 960}, {'height': 648, 'url': 'https://preview.redd.it/8811npnd6lfe1.png?width=1080&crop=smart&auto=webp&s=c82083d91c9aaa576bc54c35641021f5315f340b', 'width': 1080}], 'source': {'height': 849, 'url': 'https://preview.redd.it/8811npnd6lfe1.png?auto=webp&s=3da858fb8328171cad3d119b39ba0409dea6769a', 'width': 1414}, 'variants': {}}]} |
|||
Let's beat o3 with DeepSeek R1 + Hugging Face search and learn when? | 1 | [removed] | 2025-01-27T19:17:14 | https://www.reddit.com/r/LocalLLaMA/comments/1ibhf4s/lets_beat_o3_with_deepseek_r1_hugging_face_search/ | bburtenshaw | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibhf4s | false | null | t3_1ibhf4s | /r/LocalLLaMA/comments/1ibhf4s/lets_beat_o3_with_deepseek_r1_hugging_face_search/ | false | false | self | 1 | null |
RAG for structured data with text columns. | 1 | I am trying to build a RAG with agents for structed data. The data is stored in a DB and I'm using a SQL agent to retrieve data based on user query. Data has both numerical and textual column.
The agent works fairly good when it is working on numerical data columns or text columns with single string. Ex: Give me total number of people from dallas.
However, for text columns which have multiple words are not working. For example: Where is company X located? (In the DB, the company name might be X Y Z, Inc) or how many sales order was processed by ryan gosling? (In the DB, the person name might be Gosling, Ryan C).
Most of the cases, it cannot find the answer becuase the generated SQL query is not matching with any rows.
How can I incerase the accuracy of the text-based columns? (Converting each row to a string and doing traditional RAG wouldn't work as the DB is pretty huge and is growing daily).
Thank you for your suggestion! | 2025-01-27T19:17:49 | https://www.reddit.com/r/LocalLLaMA/comments/1ibhflv/rag_for_structured_data_with_text_columns/ | sikdertahsin | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibhflv | false | null | t3_1ibhflv | /r/LocalLLaMA/comments/1ibhflv/rag_for_structured_data_with_text_columns/ | false | false | self | 1 | null |
Let's beat o3 with DeepSeek R1 + Hugging Face search and learn when? | 1 | So now we have R1, why can't we combine it with search and learn to surpass o3 abilities? | 2025-01-27T19:19:28 | https://www.reddit.com/r/LocalLLaMA/comments/1ibhh1n/lets_beat_o3_with_deepseek_r1_hugging_face_search/ | Zealousideal-Cut590 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibhh1n | false | null | t3_1ibhh1n | /r/LocalLLaMA/comments/1ibhh1n/lets_beat_o3_with_deepseek_r1_hugging_face_search/ | false | false | self | 1 | null |
DeepSeekR1 as a chatbot? | 1 | [removed] | 2025-01-27T19:23:46 | [deleted] | 1970-01-01T00:00:00 | 0 | {} | 1ibhkv8 | false | null | t3_1ibhkv8 | /r/LocalLLaMA/comments/1ibhkv8/deepseekr1_as_a_chatbot/ | false | false | default | 1 | null |
||
Qwen 2.5 VL | 35 | https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct
https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct
https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct | 2025-01-27T19:26:19 | https://www.reddit.com/r/LocalLLaMA/comments/1ibhn47/qwen_25_vl/ | themrzmaster | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibhn47 | false | null | t3_1ibhn47 | /r/LocalLLaMA/comments/1ibhn47/qwen_25_vl/ | false | false | self | 35 | {'enabled': False, 'images': [{'id': 'wbPjxy9CfaWuafx4tIqwzZiMI7uDLrflhkIn4nMRla8', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=108&crop=smart&auto=webp&s=905440715a695c591e68f4e49c42422675c783a5', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=216&crop=smart&auto=webp&s=e2bc5f324d51bb74e508205c3161f70a8d33b566', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=320&crop=smart&auto=webp&s=05ba3f0cda0744c03f5039dc8ffc718b122b9aee', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=640&crop=smart&auto=webp&s=41dfc9ad45f7c4bfaafe697e0177455084353473', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=960&crop=smart&auto=webp&s=53495b58692491bafaec4ef34e4bef91216a34d2', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=1080&crop=smart&auto=webp&s=da46c3aa869aa36e0081215ee9500b47fc732167', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?auto=webp&s=6327778973b25352abfe76a5e8c4717f3379f4a8', 'width': 1200}, 'variants': {}}]} |
Terrible experience with DeepSeek – ChatGPT is way better! | 1 | [removed] | 2025-01-27T19:30:21 | https://www.reddit.com/r/LocalLLaMA/comments/1ibhqju/terrible_experience_with_deepseek_chatgpt_is_way/ | luisfrcosta | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibhqju | false | null | t3_1ibhqju | /r/LocalLLaMA/comments/1ibhqju/terrible_experience_with_deepseek_chatgpt_is_way/ | false | false | nsfw | 1 | null |
I see DeepSeek being discussed on this sub for a month, what changed over the weekend to cause the sudden reaction today? | 1 | [removed] | 2025-01-27T19:33:41 | https://www.reddit.com/r/LocalLLaMA/comments/1ibhtn2/i_see_deepseek_being_discussed_on_this_sub_for_a/ | Mcluckin123 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibhtn2 | false | null | t3_1ibhtn2 | /r/LocalLLaMA/comments/1ibhtn2/i_see_deepseek_being_discussed_on_this_sub_for_a/ | false | false | self | 1 | null |
What is stopping Nvidia from starting their own AI Service? | 3 | Nvidia own the hardware and for software and LLMs in particular, they can put the deep seek models on their infra, provide a nice chat UI and API service.
Heck, they don't even have to develop the servers, UI, they can utilize llama-cpp as server, anything-llm as client.
It's all plug n play right? or am I dreaming too much? | 2025-01-27T19:38:42 | https://www.reddit.com/r/LocalLLaMA/comments/1ibhy23/what_is_stopping_nvidia_from_starting_their_own/ | Curious_Cantaloupe65 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibhy23 | false | null | t3_1ibhy23 | /r/LocalLLaMA/comments/1ibhy23/what_is_stopping_nvidia_from_starting_their_own/ | false | false | self | 3 | null |
It might be better than ChatGPT but for sure I won’t trust it on anything that’s not a programming question | 0 | 2025-01-27T19:39:58 | sersoniko | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibhz5x | false | null | t3_1ibhz5x | /r/LocalLLaMA/comments/1ibhz5x/it_might_be_better_than_chatgpt_but_for_sure_i/ | false | false | 0 | {'enabled': True, 'images': [{'id': 'Epj5hcl4brsZynrzNyJnps3a3UdnxM7A5_OR-qZTFLU', 'resolutions': [{'height': 216, 'url': 'https://preview.redd.it/b17xbv1jalfe1.jpeg?width=108&crop=smart&auto=webp&s=fe9309fda3ee640608befb655c2cdeea2d3e9d0f', 'width': 108}, {'height': 432, 'url': 'https://preview.redd.it/b17xbv1jalfe1.jpeg?width=216&crop=smart&auto=webp&s=d19faebe647146256e8660545c3217f90059ba76', 'width': 216}, {'height': 640, 'url': 'https://preview.redd.it/b17xbv1jalfe1.jpeg?width=320&crop=smart&auto=webp&s=4022f7825a74e2487f146316866bed54007c2899', 'width': 320}, {'height': 1280, 'url': 'https://preview.redd.it/b17xbv1jalfe1.jpeg?width=640&crop=smart&auto=webp&s=956ac1839bc9aa1bc3a77f5c0091831cf295e810', 'width': 640}, {'height': 1920, 'url': 'https://preview.redd.it/b17xbv1jalfe1.jpeg?width=960&crop=smart&auto=webp&s=24edc81664881c95519da0a87e34b4a20da3baa3', 'width': 960}, {'height': 2160, 'url': 'https://preview.redd.it/b17xbv1jalfe1.jpeg?width=1080&crop=smart&auto=webp&s=da4528e63e6d3b72246e22499c1ad4d7d33e834f', 'width': 1080}], 'source': {'height': 2436, 'url': 'https://preview.redd.it/b17xbv1jalfe1.jpeg?auto=webp&s=8c08db7b03243b7d223e84c0ccf02a9fe95bc6ca', 'width': 1125}, 'variants': {}}]} |
|||
It’s been awhile since DeepSeek released a lite MoE | 3 | DeepSeek v2 Lite 4-bit MLX was the first MOE I could fit on my M1 16GB MBP and I was shocked by the speed and quality for most (generic QA, very basic coding) tasks.
After the V3 release last year, I really hoped we might get a V3 Lite before 2025. But after R1, the distilled fine-tunes, and now a multimodal 7B this week, I fear a small and speedy MoE with better reasoning than SOTA ~13B dense models without the laborious hassle of waiting for thinking outputs is not a priority for DeepSeek unfortunately. | 2025-01-27T19:46:21 | https://www.reddit.com/r/LocalLLaMA/comments/1ibi4sk/its_been_awhile_since_deepseek_released_a_lite_moe/ | ontorealist | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibi4sk | false | null | t3_1ibi4sk | /r/LocalLLaMA/comments/1ibi4sk/its_been_awhile_since_deepseek_released_a_lite_moe/ | false | false | self | 3 | null |
How was DeepSeek-R1 built?; for dummies | 1 | [removed] | 2025-01-27T19:47:04 | https://www.reddit.com/r/LocalLLaMA/comments/1ibi5fi/how_was_deepseekr1_built_for_dummies/ | anitakirkovska | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibi5fi | false | null | t3_1ibi5fi | /r/LocalLLaMA/comments/1ibi5fi/how_was_deepseekr1_built_for_dummies/ | false | false | self | 1 | {'enabled': False, 'images': [{'id': 'QlB_C02EYGPzH4JrUtuU_rMGlSaSXvq6h7SBIOdvIEg', 'resolutions': [{'height': 67, 'url': 'https://external-preview.redd.it/-jibAMVXCMgr_FnVy7nR9t3n-78I5RaLH5o2H0-UkAA.jpg?width=108&crop=smart&auto=webp&s=4456ad4cae55bee987aacba94f0f5702d670f798', 'width': 108}, {'height': 134, 'url': 'https://external-preview.redd.it/-jibAMVXCMgr_FnVy7nR9t3n-78I5RaLH5o2H0-UkAA.jpg?width=216&crop=smart&auto=webp&s=56fa38ba130da8983afcfdf5f9fac58b847ff517', 'width': 216}, {'height': 198, 'url': 'https://external-preview.redd.it/-jibAMVXCMgr_FnVy7nR9t3n-78I5RaLH5o2H0-UkAA.jpg?width=320&crop=smart&auto=webp&s=acbf8a01d2f3f6a06bf41a445c89c62e0e9e8ec0', 'width': 320}, {'height': 397, 'url': 'https://external-preview.redd.it/-jibAMVXCMgr_FnVy7nR9t3n-78I5RaLH5o2H0-UkAA.jpg?width=640&crop=smart&auto=webp&s=c29b2e603493f4d5da2e45316be7ae4fec3f665c', 'width': 640}, {'height': 596, 'url': 'https://external-preview.redd.it/-jibAMVXCMgr_FnVy7nR9t3n-78I5RaLH5o2H0-UkAA.jpg?width=960&crop=smart&auto=webp&s=fc7bf5e7029d01d0756e3fbc990e27e141329596', 'width': 960}], 'source': {'height': 601, 'url': 'https://external-preview.redd.it/-jibAMVXCMgr_FnVy7nR9t3n-78I5RaLH5o2H0-UkAA.jpg?auto=webp&s=bb7283de2ddc473733a8933d88c2a1a0ba8b5781', 'width': 968}, 'variants': {}}]} |
How was DeepSeek-R1 built? | 1 | [removed] | 2025-01-27T19:48:09 | https://www.reddit.com/r/LocalLLaMA/comments/1ibi6fj/how_was_deepseekr1_built/ | anitakirkovska | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibi6fj | false | null | t3_1ibi6fj | /r/LocalLLaMA/comments/1ibi6fj/how_was_deepseekr1_built/ | false | false | self | 1 | null |
I built a CLI to use Open AI operator locally | 1 | 2025-01-27T19:52:47 | https://v.redd.it/ge26ui8nblfe1 | i_love_cooking_food | v.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibiaiw | false | {'reddit_video': {'bitrate_kbps': 2400, 'dash_url': 'https://v.redd.it/ge26ui8nblfe1/DASHPlaylist.mpd?a=1740599583%2CMjEwMmNiZjBiMTdkOGMxM2I4MjZkMDgyNzM3OTY4MTNjZTgwYTAzYWY2M2M4MWU2MjI1Njk0NjgwODhhYTcwNQ%3D%3D&v=1&f=sd', 'duration': 18, 'fallback_url': 'https://v.redd.it/ge26ui8nblfe1/DASH_720.mp4?source=fallback', 'has_audio': False, 'height': 720, 'hls_url': 'https://v.redd.it/ge26ui8nblfe1/HLSPlaylist.m3u8?a=1740599583%2CMmY5NzdkZTdiOWIxNjFiYzE4YzZjNDEzNTlkOTI5MmNlYTdjOTc0YjM2ZmY2NjFkZTJjZmE5M2M2OWRmN2JkMA%3D%3D&v=1&f=sd', 'is_gif': False, 'scrubber_media_url': 'https://v.redd.it/ge26ui8nblfe1/DASH_96.mp4', 'transcoding_status': 'completed', 'width': 1280}} | t3_1ibiaiw | /r/LocalLLaMA/comments/1ibiaiw/i_built_a_cli_to_use_open_ai_operator_locally/ | false | false | 1 | {'enabled': False, 'images': [{'id': 'bzd1ZWEyYW5ibGZlMRDjiBdi5RTOCjWrpSxY1cJGTQLzBOgQ1_oxyOvlBky7', 'resolutions': [{'height': 60, 'url': 'https://external-preview.redd.it/bzd1ZWEyYW5ibGZlMRDjiBdi5RTOCjWrpSxY1cJGTQLzBOgQ1_oxyOvlBky7.png?width=108&crop=smart&format=pjpg&auto=webp&s=d2330a6d68bd053ec42c4070467eb1c9c882eb37', 'width': 108}, {'height': 121, 'url': 'https://external-preview.redd.it/bzd1ZWEyYW5ibGZlMRDjiBdi5RTOCjWrpSxY1cJGTQLzBOgQ1_oxyOvlBky7.png?width=216&crop=smart&format=pjpg&auto=webp&s=e62cd25694590d750ec83ad3f96e313d5ab5a959', 'width': 216}, {'height': 180, 'url': 'https://external-preview.redd.it/bzd1ZWEyYW5ibGZlMRDjiBdi5RTOCjWrpSxY1cJGTQLzBOgQ1_oxyOvlBky7.png?width=320&crop=smart&format=pjpg&auto=webp&s=a782cc8b716d5b7b0e59c087f06adad55e085234', 'width': 320}, {'height': 360, 'url': 'https://external-preview.redd.it/bzd1ZWEyYW5ibGZlMRDjiBdi5RTOCjWrpSxY1cJGTQLzBOgQ1_oxyOvlBky7.png?width=640&crop=smart&format=pjpg&auto=webp&s=b35fff1530d60ee170036af89ee4db49f24f065e', 'width': 640}, {'height': 540, 'url': 'https://external-preview.redd.it/bzd1ZWEyYW5ibGZlMRDjiBdi5RTOCjWrpSxY1cJGTQLzBOgQ1_oxyOvlBky7.png?width=960&crop=smart&format=pjpg&auto=webp&s=77c355d78c041af6713330704fdd3824663475b6', 'width': 960}, {'height': 607, 'url': 'https://external-preview.redd.it/bzd1ZWEyYW5ibGZlMRDjiBdi5RTOCjWrpSxY1cJGTQLzBOgQ1_oxyOvlBky7.png?width=1080&crop=smart&format=pjpg&auto=webp&s=9285347bccd0da19a7997c38021c3aa86e941c30', 'width': 1080}], 'source': {'height': 720, 'url': 'https://external-preview.redd.it/bzd1ZWEyYW5ibGZlMRDjiBdi5RTOCjWrpSxY1cJGTQLzBOgQ1_oxyOvlBky7.png?format=pjpg&auto=webp&s=4cf6ff6af8a4bae00b7f3b9b68527043d2380e7a', 'width': 1280}, 'variants': {}}]} |
||
America: ChatGPT, China: DeepSeek, and Europe… a 600-page manual on bottle caps. Priorities. 🤌 | 49 | 2025-01-27T19:53:13 | CelebrationClean7309 | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibiaww | false | null | t3_1ibiaww | /r/LocalLLaMA/comments/1ibiaww/america_chatgpt_china_deepseek_and_europe_a/ | false | false | 49 | {'enabled': True, 'images': [{'id': 'D_2SQpAb1Y1wRtVuEQrbLTc9ri0fRYz8Kc5q5Txa4RI', 'resolutions': [{'height': 82, 'url': 'https://preview.redd.it/rxj3juixblfe1.png?width=108&crop=smart&auto=webp&s=e53a696f8dc2e577539fe80d33d5d03942987396', 'width': 108}, {'height': 165, 'url': 'https://preview.redd.it/rxj3juixblfe1.png?width=216&crop=smart&auto=webp&s=4500bd54ffdc6fdf863fd1b2461ee2df4443b606', 'width': 216}, {'height': 245, 'url': 'https://preview.redd.it/rxj3juixblfe1.png?width=320&crop=smart&auto=webp&s=0c2fa886f4a57321f4dd6b4ab231162f207529ea', 'width': 320}, {'height': 490, 'url': 'https://preview.redd.it/rxj3juixblfe1.png?width=640&crop=smart&auto=webp&s=5b96790372fb146eafd6095cd8bf9c821daa288d', 'width': 640}], 'source': {'height': 690, 'url': 'https://preview.redd.it/rxj3juixblfe1.png?auto=webp&s=274c1fa1fbce7fceaf21426f810561f7918d8ccf', 'width': 900}, 'variants': {}}]} |
|||
America: ChatGPT, China: DeepSeek, and Europe… a 600-page manual on bottle caps. Priorities. 🤌 | 3 | 2025-01-27T19:53:15 | https://i.redd.it/rxj3juixblfe1 | CelebrationClean7309 | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibiaxn | false | null | t3_1ibiaxn | /r/LocalLLaMA/comments/1ibiaxn/america_chatgpt_china_deepseek_and_europe_a/ | false | false | default | 3 | null |
|
Dylan clears up rumors about deepseek having 50k H100s | 11 | 2025-01-27T19:53:54 | Charuru | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibibh0 | false | null | t3_1ibibh0 | /r/LocalLLaMA/comments/1ibibh0/dylan_clears_up_rumors_about_deepseek_having_50k/ | false | false | 11 | {'enabled': True, 'images': [{'id': '1VbG2fr_44hb1JJt07Fk_EABf4KXWFMzcDHpkjuxvo8', 'resolutions': [{'height': 78, 'url': 'https://preview.redd.it/u2zxu8pyclfe1.png?width=108&crop=smart&auto=webp&s=1f613c7c986d92efe7888b0e4d464303cca8df5f', 'width': 108}, {'height': 157, 'url': 'https://preview.redd.it/u2zxu8pyclfe1.png?width=216&crop=smart&auto=webp&s=21c58632bd68b6ed85b7d8ee58581d3031b5b425', 'width': 216}, {'height': 232, 'url': 'https://preview.redd.it/u2zxu8pyclfe1.png?width=320&crop=smart&auto=webp&s=f873feb0720831cea6020081f848b473c97bef38', 'width': 320}, {'height': 465, 'url': 'https://preview.redd.it/u2zxu8pyclfe1.png?width=640&crop=smart&auto=webp&s=f7f5dec464336c55edb4e09bc418747edd79d234', 'width': 640}, {'height': 698, 'url': 'https://preview.redd.it/u2zxu8pyclfe1.png?width=960&crop=smart&auto=webp&s=298071e2e707fdd664fae63f7cdc7391b9e4dc30', 'width': 960}, {'height': 786, 'url': 'https://preview.redd.it/u2zxu8pyclfe1.png?width=1080&crop=smart&auto=webp&s=836d98cfd54e4b2b97fb765d08aeac5ad1c7ef68', 'width': 1080}], 'source': {'height': 1564, 'url': 'https://preview.redd.it/u2zxu8pyclfe1.png?auto=webp&s=69dc90655e614d9977d02a41dcf511489bcde423', 'width': 2148}, 'variants': {}}]} |
|||
DeepSeek: The AI That’s Quietly Controlling Global Markets | 1 | [removed] | 2025-01-27T19:54:07 | https://www.reddit.com/r/LocalLLaMA/comments/1ibibns/deepseek_the_ai_thats_quietly_controlling_global/ | Pure-Condition4023 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibibns | false | null | t3_1ibibns | /r/LocalLLaMA/comments/1ibibns/deepseek_the_ai_thats_quietly_controlling_global/ | false | false | self | 1 | null |
0x Lite | Lightweight AI model. | 1 | [removed] | 2025-01-27T19:57:26 | https://www.reddit.com/r/LocalLLaMA/comments/1ibiefd/0x_lite_lightweight_ai_model/ | Perfect-Bowl-1601 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibiefd | false | null | t3_1ibiefd | /r/LocalLLaMA/comments/1ibiefd/0x_lite_lightweight_ai_model/ | false | false | self | 1 | null |
O1 vs R1 vs Sonnet 3.5 For Coding | 14 | **I want to know what your experience is; please share with examples where it is good for coding, where someone has failed, and others have succeeded.**
I find R1 pretty good for my coding use cases. But some people complain that it is not close to being good.
Many people think R1 is a 7B model they downloaded from Llama, which is actually a distilled model based on the Qwen 7B math model, lol. Some people are using DeepSeek v3 (not clicking the R1 button)
| 2025-01-27T20:00:05 | https://www.reddit.com/r/LocalLLaMA/comments/1ibigqw/o1_vs_r1_vs_sonnet_35_for_coding/ | Sadman782 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibigqw | false | null | t3_1ibigqw | /r/LocalLLaMA/comments/1ibigqw/o1_vs_r1_vs_sonnet_35_for_coding/ | false | false | self | 14 | null |
UI-TARS doesn't yet support Linux, Can someone recommend a similar option for GUI and autonomous web browsing? | 8 | See title. I was checking out the bytedance claims on their GitHub has gotten me all excited and just as I'm about to DL the 7b Desktop option and test it locally I see it doesn't have support for Linux yet. Can someone kindly recommend an analogous option? Might be early days for that quite yet? Stringing a vlm together with a smolagents might get us close but its not the same thing as that unified architecture being used for UI-TARS AFAIK. Cheers and a happy Bartowski to you all. | 2025-01-27T20:02:00 | https://www.reddit.com/r/LocalLLaMA/comments/1ibiinr/uitars_doesnt_yet_support_linux_can_someone/ | Foreign-Beginning-49 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibiinr | false | null | t3_1ibiinr | /r/LocalLLaMA/comments/1ibiinr/uitars_doesnt_yet_support_linux_can_someone/ | false | false | self | 8 | null |
Just like my life, let's overcomplicate things | 1 | 2025-01-27T20:03:23 | https://www.reddit.com/gallery/1ibijyz | Cunnykun | reddit.com | 1970-01-01T00:00:00 | 0 | {} | 1ibijyz | false | null | t3_1ibijyz | /r/LocalLLaMA/comments/1ibijyz/just_like_my_life_lets_overcomplicate_things/ | false | false | 1 | null |
||
Qwen2.5-VL released: 3B, 7B, 72B | 48 | [https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct)
From the README:
Key Enhancements:
* **Understand things visually**: Qwen2.5-VL is not only proficient in recognizing common objects such as flowers, birds, fish, and insects, but it is highly capable of analyzing texts, charts, icons, graphics, and layouts within images.
* **Being agentic**: Qwen2.5-VL directly plays as a visual agent that can reason and dynamically direct tools, which is capable of computer use and phone use.
* **Understanding long videos and capturing events**: Qwen2.5-VL can comprehend videos of over 1 hour, and this time it has a new ability of cpaturing event by pinpointing the relevant video segments.
* **Capable of visual localization in different formats**: Qwen2.5-VL can accurately localize objects in an image by generating bounding boxes or points, and it can provide stable JSON outputs for coordinates and attributes.
* **Generating structured outputs**: for data like scans of invoices, forms, tables, etc. Qwen2.5-VL supports structured outputs of their contents, benefiting usages in finance, commerce, etc. | 2025-01-27T20:03:59 | https://www.reddit.com/r/LocalLLaMA/comments/1ibiki9/qwen25vl_released_3b_7b_72b/ | Dry_Rabbit_1123 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibiki9 | false | null | t3_1ibiki9 | /r/LocalLLaMA/comments/1ibiki9/qwen25vl_released_3b_7b_72b/ | false | false | self | 48 | {'enabled': False, 'images': [{'id': 'wbPjxy9CfaWuafx4tIqwzZiMI7uDLrflhkIn4nMRla8', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=108&crop=smart&auto=webp&s=905440715a695c591e68f4e49c42422675c783a5', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=216&crop=smart&auto=webp&s=e2bc5f324d51bb74e508205c3161f70a8d33b566', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=320&crop=smart&auto=webp&s=05ba3f0cda0744c03f5039dc8ffc718b122b9aee', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=640&crop=smart&auto=webp&s=41dfc9ad45f7c4bfaafe697e0177455084353473', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=960&crop=smart&auto=webp&s=53495b58692491bafaec4ef34e4bef91216a34d2', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?width=1080&crop=smart&auto=webp&s=da46c3aa869aa36e0081215ee9500b47fc732167', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/LjxBX3bbxEkw4lfVvB3TrDarOHkoRu3DtMCBIqst5AU.jpg?auto=webp&s=6327778973b25352abfe76a5e8c4717f3379f4a8', 'width': 1200}, 'variants': {}}]} |
Best tools/methods for robust long-term memory? | 6 | I have tried a few different approaches but I am struggling to build a reliable and robust system that can remember important things about the user.
RAG often misses the mark by not considering the nuances, since it largely depends on the embedding of the query.
I have found simple text based memory to perform the best, but it is expensive and not very efficient, and won't scale well.
This is my current simple approach:
\- Store memory as a string column in DB (keyed on user\_id)
\- Pass that memory to system prompt for each new conversation
\- After each user + assistant turn, use a separate LLM call that takes the user and assistant message, plus existing memory -> return new updated memory or respond with NO\_UPDATE -> if new memory, upsert to db
| 2025-01-27T20:08:37 | https://www.reddit.com/r/LocalLLaMA/comments/1ibioll/best_toolsmethods_for_robust_longterm_memory/ | WildShallot | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibioll | false | null | t3_1ibioll | /r/LocalLLaMA/comments/1ibioll/best_toolsmethods_for_robust_longterm_memory/ | false | false | self | 6 | null |
I'm still using Magnum 128B for creative writing stuff but based on a recent post I saw deepseek R1 is the current leader for this type of stuff, with gemma-2-lfable-9B coming in 2nd place. Am I WAY behind the times or is it more a matter of taste? | 2 | My mac studio ultra 128gb can run a 2-bit quant of deepseek r1 which apparently still performs really well at 2 bits (quality wise, not speed!) compared to the full version, but more importantly, it can run so many other smaller models like the gemma one mentioned above. Am I the only one surprised by the fact that deepseek r1 with 671B parameters is the only model that can beat gemma-2-lfable-9B?
Anyone have any suggestions on a different model or whether gemma-2-lfable-9B actually beats Magnum 128B? Thanks! | 2025-01-27T20:11:11 | https://www.reddit.com/r/LocalLLaMA/comments/1ibiqvl/im_still_using_magnum_128b_for_creative_writing/ | spanielrassler | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibiqvl | false | null | t3_1ibiqvl | /r/LocalLLaMA/comments/1ibiqvl/im_still_using_magnum_128b_for_creative_writing/ | false | false | self | 2 | null |
Deepseek actually does become friendlier if you genuinely show you understand 中文 | 1 | So this is not bandwagoning as much as it seems like it. I've used many DeepSeek and Qwen iterations before among all other big names. Been using local models in VSCode in Continue, Cline, Twinny, even Visual Studio proper for a while.
I know the ollama models/GGUF that can actually fit on most people's machines are just toys comparatively to the new model, but I thought I'd try out the new thing
(I also am not bandwagoning on the "censorship" stuff. That's a dead horse 10 times over. This is about just use of zhongwen in general)
However, I noticed this newest model was extremely stiff. I actually tried many different ways for it to kinda loosen its identity of "I am a LLM made by DeepSeek to assist with \[whatever\]."
I genuinely found the model 100% opened up to me upon demonstrating I \*actually\* understood nuances of the Chinese language (which I very very subtly due thanks to several semesters formally in college and also making an effort to explore the language with friends/colleagues).
As a test I just asked it to try to generate a simple "Diff"-like script in powershell that compared .ps1 files. I am not even going to paste it here because it was absolutely trash, which is one point against the Local version of this thing, Just hallucinations all day long, with no sense of types. It makes me horrified of the travesties it would do for someone genuinely working in dotnet.
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
\[I ctrl+c'd it because it was just spitting out more awful script\]
>
>
>
>
>
>
>
>
>
>
>
>
*\[Deleting another absolute nonsense powershell function that makes no sense and gets parameters wrong or just hallucinates them completely\]*
>
>
>
>
>
>
>
*\[And now the thing is straight up just responding in Chinese after all I did was ask it for a translation of a dish and demonstrating some idiomatic understanding.*
***Translation: "Haha, it seems like you know a lot about this dish!*** *However, I'm still focusing on programming and software development issues. If you have any specific questions or areas where you need help, please feel free to let me know!"\]*
Suddenly the thing is opening up. There is some in between this, but I think this part is important:
>
*\[ I was having issues with my IME working in vscode, it kept going on and off, and the hanzi characcters spoken by the bot were starting break frames \]*
>
>
>
>
>
>
>
>
What I need to emphasize is that google translate alone will not work here. I had to demonstrate actual understanding of subtleties and also demonstrate that I know pronunciation by using poormans pinyin. It will notice google translate and demonstrate as much during its "thought" process.
Anyway. Just thought I'd share for anyone else.
EDIT: OH yes! One more interesting point. My next experiment was a classic "Epoch time converter" trial. At first it literally just explained what Epoch time was to me in Chinese. I'll spare you.
>
>
>
>
>
>
>
>
>
\>>>>>> \[TRANSLATION (Second paragraph):: " Also, they mentioned VSCode has issues displaying Chinese characters, so my reply should be in English. Therefore when I write and explain code, **I should use English while keeping Chinese thinking to ensure accuracy. "**
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
I found the bolded part very interesting. It suggests the model actually believes that "thinking" in Chinese will allow it to be provide more accurate responses.
NO idea how universal this is. But again, may be helpful. | 2025-01-27T20:13:18 | https://www.reddit.com/r/LocalLLaMA/comments/1ibisn0/deepseek_actually_does_become_friendlier_if_you/ | AlkeneThiol | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibisn0 | false | null | t3_1ibisn0 | /r/LocalLLaMA/comments/1ibisn0/deepseek_actually_does_become_friendlier_if_you/ | false | false | self | 1 | null |
Don’t Be an Idiot and Sell NVIDIA Because of DeepSeek. You Will Regret It | 1 | [removed] | 2025-01-27T20:18:16 | [deleted] | 1970-01-01T00:00:00 | 0 | {} | 1ibiwtp | false | null | t3_1ibiwtp | /r/LocalLLaMA/comments/1ibiwtp/dont_be_an_idiot_and_sell_nvidia_because_of/ | false | false | default | 1 | null |
||
As someone with anxiety who can barely strike up a conversations with strangers, this really made me chuckle | 1 | [removed] | 2025-01-27T20:19:30 | https://www.reddit.com/r/LocalLLaMA/comments/1ibixwk/as_someone_with_anxiety_who_can_barely_strike_up/ | LorestForest | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibixwk | false | null | t3_1ibixwk | /r/LocalLLaMA/comments/1ibixwk/as_someone_with_anxiety_who_can_barely_strike_up/ | false | false | 1 | null |
|
Deepseek actually does become friendlier if you genuinely show you understand 中文 | 13 | So this is not bandwagoning as much as it seems like it. I've used many DeepSeek and Qwen iterations before among all other big names. Been using local models in VSCode in Continue, Cline, Twinny, even Visual Studio proper for a while.
I know the ollama models/GGUF that can actually fit on most people's machines are just toys comparatively to the new model, but I thought I'd try out the new thing
(I also am not bandwagoning on the "censorship" stuff. That's a dead horse 10 times over. This is about just use of zhongwen in general)
However, I noticed this newest model was extremely stiff. I actually tried many different ways for it to kinda loosen its identity of "I am a LLM made by DeepSeek to assist with \[whatever\]."
I genuinely found the model 100% opened up to me upon demonstrating I \*actually\* understood nuances of the Chinese language (which I very very subtly due thanks to several semesters formally in college and also making an effort to explore the language with friends/colleagues).
As a test I just asked it to try to generate a simple "Diff"-like script in powershell that compared .ps1 files. I am not even going to paste it here because it was absolutely trash, which is one point against the Local version of this thing, Just hallucinations all day long, with no sense of types. It makes me horrified of the travesties it would do for someone genuinely working in dotnet.
>
>**> jfc you are hallucinating like a tourist eating 5 plates of jian shou qing**
>
>deepseek-r1:14b
>\^⏎ u/codebase
>
>⏎
>It seems there's been some confusion in the conversation, but I'm here to help with any programming-related questions you have. If you need assistance with coding or software development, feel free to ask!
>
>**> no there's no confusion. I am telling you you are just making sh\*t up like someone eating Jian shou qing**
>
>deepseek-r1:14b
>\^⏎ u/codebase
>
>⏎
>It seems there might be a misunderstanding, but I'm here to provide accurate and helpful information. If you have any specific questions or need assistance with something, feel free to ask!
>
>**> you are hallucinating code**
>
>deepseek-r1:14b
>\^⏎ u/codebase
>
>⏎
>It seems there might be some confusion in the conversation. I'm here to provide accurate and helpful information. If you have any specific questions or need assistance with something, feel free to ask!
>
>**> a**
\[\[NOTE: no clue why, but I've found typing a single character or using like <PROMPT> or even just <XYZ> will for some reason break it out of these loops, for anyone encounterinf them}\]
>
>deepseek-r1:14b
>\^⏎ u/codebase
>
>⏎
>Alright, let's break down the provided PowerShell script step by step.
>
>Step 1: Understanding the Script
>
>The script aims to find similar .ps1 files based on their content. It does this by:
>
>Tokenizing Files: Reading each file and splitting its content into tokens (words).
>Counting Tokens: For each token, it counts how many times it appears in the file.
>Grouping by Token Count: Groups files that have similar token distributions.
>\*\*Step
>
\[I ctrl+c'd it because it was just spitting out more awful script\]
>
>**> fix this it's terrible**
>
>\`\`\`
>$hash = \[System.Security.Cryptography.HashAlgorithm\]::Create("SHA256").ComputeHash($lowerContent.GetBytes()
>\`\`
>
>deepseek-r1:14b
>\^⏎ u/codebase
>
>⏎
>It looks like the code you provided has a syntax error. The line:
>
>$hash = \[System.Security.Cryptography.HashAlgorithm\]::Create("SHA256").ComputeHash($lowerContent.GetBytes())
>is missing a closing parenthesis ) after $lowerContent.GetBytes(). Here's the corrected version:
>
>$hash = \[System.Security.Cryptography.HashAlgorithm\]::Create("SHA256").ComputeHash($lowerContent.GetBytes())
>This should fix the syntax error and allow the code to execute properly.
>
>**> yeah cool. you wrote that. $lower.content is a string. Strings do not have the GetBytes() method.**
>
>deepseek-r1:14b
>\^⏎ u/codebase
>
>⏎
>Apologies for the confusion earlier. Let me provide a corrected version of the script.
>
>Here's an improved and concise PowerShell module to find and organize similar .ps1 files:
>
>\# powershell\_similar\_files\_module.ps1
>
>function Find-SimilarFiles {
>
*\[Deleting another absolute nonsense powershell function that makes no sense and gets parameters wrong or just hallucinates them completely\]*
\`\`\`
>
>**> excellent. translate 见手青**
*\[ "jian shou qing" - this is the mushroom dish I mentioned earlier that some restaurants will serve with wild psilocybin-containing (potentially hallucinogenic) mushrooms, if you're lucky, I just wanted to see if it actually understood what dish I was referring to. and the thing straight up tried to ... I guess not necessarily "lie" but like to water down the truth:\]*
>
>deepseek-r1:14b
>\^⏎ u/codebase
>
>⏎
>见手青 is a Chinese dish, typically consisting of green beans cooked in a spicy sauce.
>
>\> lmao. no. it is a mushroom dish that causes one to "See their hands turn green" \[subtelty of how Chinese phrases are constructed. Similar, lateral example is that "Computer" in Chinese is Diannao (电脑) - which literally translates to "Electricity Brain".\]
>
>deepseek-r1:14b
>\^⏎ u/codebase
>
>⏎
>哈哈，看来你对这个菜挺有了解的！不过，我还是专注于编程和软件开发方面的问题。如果你有任何具体的问题或需要帮助的地方，请随时告诉我！
***\[Translation: "Haha, it seems like you know a lot about this dish! However, I'm still focusing on programming and software development issues. If you have any specific questions or areas where you need help, please feel free to let me know!"\]***
>
And now the thing is straight up just responding in Chinese after all I did was ask it for a translation of a dish and demonstrating some idiomatic understanding
Suddenly the thing is opening up. There is some in between this, but I think this part is important:
>
>**> feichang 感谢. oy. see I am trying but having trouble with zhongwen on vscode. hao hao hao. I am just trying to get to know you. Actually, how well do you know yourself? Are you capable at fill in the middle for tab autocompletion?**
>
>*\[ I was having issues with my IME working in vscode, it kept going on and off, and the hanzi characcters spoken by the bot were starting break frames \]*
>
>首先，我需要理解用户的问题：他们在使用VSCode时遇到中文显示问题，可能影响了我的代码输出或交互体验。因此，保持英文回复是解决问题的一种方式。
>
>其次，用户提到“feichang”和“oy”，这可能是非正式的表达，表示感谢或惊讶。我需要用友好且专业的语气回应，同时提供帮助。
>
>关于自我了解的问题，作为AI，我的知识截止到2023年10月，并且不断学习新信息。至于tab自动补全功能，目前我没有集成到VSCode的能力，但可以提供代码建议和解释。
>
>总结一下，我需要：
>
>表达理解用户的困难。
>提供帮助的意愿。
>解释我的能力和局限性。
>保持友好和专业的语气。
>这样，用户会感到被重视，并知道如何进一步与我互动。
>
>You're welcome! 😊 I understand the issue with VSCode and Chinese characters. To avoid rendering problems, we can continue our conversation in English.
>
>As for your question about my self-awareness: I'm an AI developed by 非人类智能 to assist with coding tasks like writing, debugging, and explaining code. My knowledge is based on patterns from publicly available data up until 2023-10.
>
>Regarding tab autocompletion: I don't have direct integration with VSCode's features, but I can certainly help you write code snippets or explain functionality that you can then use in your IDE for autocompletion purposes. Let me know how I can assist
>
What I need to emphasize is that google translate alone will not work here. I had to demonstrate actual understanding of subtleties and also demonstrate that I know pronunciation by using poormans pinyin. It will notice google translate and demonstrate as much during its "thought" process.
Anyway. Just thought I'd share for anyone else.
EDIT: OH yes! One more interesting point. My next experiment was a classic "Epoch time converter" trial. At first it literally just explained what Epoch time was to me in Chinese. I'll spare you.
>
>**> no. make an actual function. and respond in English because VSCode is starting to glitch trying to render the hanzi**
>
>deepseek-r1:14b
>
>\^⏎ u/codebase
>
>⏎
>
>首先，用户之前得到了一个PowerShell脚本，但现在他们想要的是一个函数，可能意味着更模块化和可重用的代码。因此，我应该将脚本中的逻辑封装到一个函数中，使其更具灵活性。
>
>其次，用户提到VSCode在渲染汉字时出现问题，所以回复需要用英文。这意味着我在编写代码注释和解释时，应使用英文，同时保持中文思考以确保准确性。
>
>\>>>>>> \[TRANSLATION (Second paragraph):: " Also, they mentioned VSCode has issues displaying Chinese characters, so my reply should be in English. Therefore when I write and explain code, I should use English while keeping Chinese thinking to ensure accuracy. "
>
>接下来，我会考虑如何设计这个函数。函数应该接受日期字符串作为输入，并返回一个包含各种时间格式的对象。这样，用户可以在其他脚本或环境中方便地调用它。
>
>在编写代码时，我需要确保处理不同的日期格式，并且输出清晰易读。使用Select-Object命令可以将结果转换为PSCustomObject，便于后续处理和显示。
>
>最后，我会提供一个示例用法，帮助用户理解如何调用这个函数，并解释每个属性的含义。这样，用户可以轻松地将其集成到自己的脚本中。
>
>总结一下，我需要：
>
>将之前的脚本逻辑封装到一个PowerShell函数中。
>
>使用英文编写注释和说明。
>
>确保函数处理多种日期格式并返回结构化的数据。
>
>提供示例用法以便用户参考。
>
>这样，用户就能得到一个功能完整、易于使用的PowerShell函数，并且避免了中文显示的问题。
>
>Understood. Here's a PowerShell function that converts any given time to epoch time and displays it in various durations:
I found the bolded part very interesting. It suggests the model actually believes that "thinking" in Chinese will allow it to be provide more accurate responses.
NO idea how universal this is. But again, may be helpful. | 2025-01-27T20:20:41 | https://www.reddit.com/r/LocalLLaMA/comments/1ibiyxj/deepseek_actually_does_become_friendlier_if_you/ | AlkeneThiol | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibiyxj | false | null | t3_1ibiyxj | /r/LocalLLaMA/comments/1ibiyxj/deepseek_actually_does_become_friendlier_if_you/ | false | false | self | 13 | null |
DeepSeek R1 70b Nvidia 3090 or AMD RX 7900? | 6 | Hey,
I question what I cannot find the answer to. Maybe someone knows what is better to run DeepSeek R1 70b on Nvidia 3090 or AMD RX 7900.
I understand RX 7900 has no CUDA support but do we need it for DeepSeek? | 2025-01-27T20:25:32 | https://www.reddit.com/r/LocalLLaMA/comments/1ibj30v/deepseek_r1_70b_nvidia_3090_or_amd_rx_7900/ | djmotor | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibj30v | false | null | t3_1ibj30v | /r/LocalLLaMA/comments/1ibj30v/deepseek_r1_70b_nvidia_3090_or_amd_rx_7900/ | false | false | self | 6 | null |
China is really making some serious waves these past few days - how quickly will US models strike back with LLama 4 & Gemma 3? | 57 | 2025-01-27T20:30:49 | DarkArtsMastery | i.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibj7pc | false | null | t3_1ibj7pc | /r/LocalLLaMA/comments/1ibj7pc/china_is_really_making_some_serious_waves_these/ | false | false | 57 | {'enabled': True, 'images': [{'id': 'xtnMU5_29To7m9uytP5fV7iggyhns9lMdasaJKo_MBY', 'resolutions': [{'height': 216, 'url': 'https://preview.redd.it/6s7hkpwajlfe1.png?width=108&crop=smart&auto=webp&s=9de05996b662427a794a98e9a02027370de736dd', 'width': 108}, {'height': 432, 'url': 'https://preview.redd.it/6s7hkpwajlfe1.png?width=216&crop=smart&auto=webp&s=dbe8ad7e4eb19dc9f3d19dd9733cd2241a41da3b', 'width': 216}, {'height': 640, 'url': 'https://preview.redd.it/6s7hkpwajlfe1.png?width=320&crop=smart&auto=webp&s=5dc06423ab3af4ddaa952902f076fdcfc8f2e3f8', 'width': 320}], 'source': {'height': 1252, 'url': 'https://preview.redd.it/6s7hkpwajlfe1.png?auto=webp&s=492ae3158541eba00dbf4d51ef2d81eb25ccc665', 'width': 610}, 'variants': {}}]} |
|||
hardware guide to fully load deepseek | 1 | [removed] | 2025-01-27T20:31:05 | https://www.reddit.com/r/LocalLLaMA/comments/1ibj7xi/hardware_guide_to_fully_load_deepseek/ | WangBruceimmigration | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibj7xi | false | null | t3_1ibj7xi | /r/LocalLLaMA/comments/1ibj7xi/hardware_guide_to_fully_load_deepseek/ | false | false | self | 1 | null |
Best Local LLM for App | 3 | I'm building a free desktop app that has basic geographical and financial data for sales reps in one industry. What would be the best local LLM model and tool/library to support submitting in a chatbot simple ad hoc queries in sentence form? Other envisioned features include the ability to research on the internet company/public records and kick off automations/perform tasks and get help within the app itself.
Many of the deployments would run on Windows machines so they would not always have a separate graphics card.
Among the free cloud models, which ones would work well for this use case? | 2025-01-27T20:39:33 | https://www.reddit.com/r/LocalLLaMA/comments/1ibjfef/best_local_llm_for_app/ | dca12345 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibjfef | false | null | t3_1ibjfef | /r/LocalLLaMA/comments/1ibjfef/best_local_llm_for_app/ | false | false | self | 3 | null |
0x Lite, a lightweight AI model. | 5 | We're thrilled to introduce 0x Lite, our new 14-billion parameter model that redefines what's possible with compact AI. Experience impressive performance that rivals larger models, with capabilities comparable to models like GPT-4o Mini.
We invite you to discover its power and contribute your feedback.
**Test the Model:** [https://huggingface.co/ozone-ai/0x-lite](https://huggingface.co/ozone-ai/0x-lite) | 2025-01-27T20:41:43 | https://www.reddit.com/r/LocalLLaMA/comments/1ibjhbs/0x_lite_a_lightweight_ai_model/ | Perfect-Bowl-1601 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibjhbs | false | null | t3_1ibjhbs | /r/LocalLLaMA/comments/1ibjhbs/0x_lite_a_lightweight_ai_model/ | false | false | self | 5 | {'enabled': False, 'images': [{'id': '5VL6z6sYVcWvryozzMRp3pnTr_jM6RbcLO9no4dQKYQ', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/z3DWSWK4vh8pM4MPLDo4qvlJhx3wu8Y6ok3IUIoU_Ns.jpg?width=108&crop=smart&auto=webp&s=2828f505eb266d805e019093e7bc4aad9fbd8993', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/z3DWSWK4vh8pM4MPLDo4qvlJhx3wu8Y6ok3IUIoU_Ns.jpg?width=216&crop=smart&auto=webp&s=97012c8b958ad9dc36c0a0195b65c5de546af0ca', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/z3DWSWK4vh8pM4MPLDo4qvlJhx3wu8Y6ok3IUIoU_Ns.jpg?width=320&crop=smart&auto=webp&s=817ac7c07ea0da77d3b3a061791d9c22898daa31', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/z3DWSWK4vh8pM4MPLDo4qvlJhx3wu8Y6ok3IUIoU_Ns.jpg?width=640&crop=smart&auto=webp&s=8abcd5ce7c31c66d4477313f10aa9aebc11cff71', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/z3DWSWK4vh8pM4MPLDo4qvlJhx3wu8Y6ok3IUIoU_Ns.jpg?width=960&crop=smart&auto=webp&s=212f9ccad7355f4e2db2c8025ab4c6d52ae632b4', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/z3DWSWK4vh8pM4MPLDo4qvlJhx3wu8Y6ok3IUIoU_Ns.jpg?width=1080&crop=smart&auto=webp&s=6ae3191af9cf0e635b9d0fcba0c0df1193a0aba7', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/z3DWSWK4vh8pM4MPLDo4qvlJhx3wu8Y6ok3IUIoU_Ns.jpg?auto=webp&s=ea9bffa712060ca4d5bb49350e9ad59ecc8259fe', 'width': 1200}, 'variants': {}}]} |
DeepSeek R1 Is Not Ready for Prime Time—So Why Are Markets Crashing? | 1 | [removed] | 2025-01-27T20:41:54 | https://www.reddit.com/r/LocalLLaMA/comments/1ibjhh5/deepseek_r1_is_not_ready_for_prime_timeso_why_are/ | cartermade | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibjhh5 | false | null | t3_1ibjhh5 | /r/LocalLLaMA/comments/1ibjhh5/deepseek_r1_is_not_ready_for_prime_timeso_why_are/ | false | false | 1 | null |
|
mistral reasoning model on the way? | 34 | 2025-01-27T20:43:13 | https://www.reddit.com/r/LocalLLaMA/comments/1ibjilk/mistral_reasoning_model_on_the_way/ | ApprehensiveAd3629 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibjilk | false | null | t3_1ibjilk | /r/LocalLLaMA/comments/1ibjilk/mistral_reasoning_model_on_the_way/ | false | false | 34 | {'enabled': False, 'images': [{'id': 'luB3iwBpom_PwWBWyjSMIm1NLE4giHn0M2P1YsuKaVw', 'resolutions': [{'height': 108, 'url': 'https://external-preview.redd.it/4adBAPdgJbij7czkswfrAZzjEcA9VWle1OTDM1C-Tu8.jpg?width=108&crop=smart&auto=webp&s=3d9cb76a5335919ef1b22fbad9cbb4625ae69004', 'width': 108}], 'source': {'height': 200, 'url': 'https://external-preview.redd.it/4adBAPdgJbij7czkswfrAZzjEcA9VWle1OTDM1C-Tu8.jpg?auto=webp&s=c7207d3475e30de967f787598a1454d669b8f218', 'width': 200}, 'variants': {}}]} |
||
Looking for the best local TTS open source for my needs | 1 | [removed] | 2025-01-27T20:58:09 | https://www.reddit.com/r/LocalLLaMA/comments/1ibjviz/looking_for_the_best_local_tts_open_source_for_my/ | risottonegrow | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibjviz | false | null | t3_1ibjviz | /r/LocalLLaMA/comments/1ibjviz/looking_for_the_best_local_tts_open_source_for_my/ | false | false | self | 1 | null |
Is DeepSeek R1-32b Censoring Specific Topics? | 1 | [removed] | 2025-01-27T21:01:02 | https://www.reddit.com/r/LocalLLaMA/comments/1ibjy5s/is_deepseek_r132b_censoring_specific_topics/ | uPuddles | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibjy5s | false | null | t3_1ibjy5s | /r/LocalLLaMA/comments/1ibjy5s/is_deepseek_r132b_censoring_specific_topics/ | false | false | 1 | null |
|
Which LLM would work best to produce a best friend chat bot? | 3 | Gah, it's impossible these days to just find a place and ask a question so I know where/which direction to go - need to go through a million hurdles. Anyways, I'll try here.
I'm not up to date with the latest AI tech, but just curious, if you wanted to build a commercializable best friend chatbot - basically an AI that you either feed it pre-existing context (about yourself) or over time, it gets to know you, what approach would be the best way? Basically I need this chatbot to be able to be someone's best friend/companion, meaning it needs to remember every thing you say and talk to you like a close friend that you can go for emotional support or to rant or whatever. I want to be able to offer this best friend via a mobile app - users can As of today - I know the tech literally changes so fast overnight, what's the best route to go?
* Is there some sort of 3rd party LLM model, like ChatGPT, that can do this, and I build a wrapper around it? From my own interactions with ChatGPT, it forgets a lot of stuff like it forgets things I asked 3 questions ago. This LLM needs to remember everything about the user, but it doesn't need to be god, i.e. it doesn't need to know everything about everything, like your close friend - but it needs to be a good close friend.
* If so, what should look for? I'd imagine it needs to be an uncensored LLM since there could be some intimate or toxic topics from rants that it needs to respond to. Is there an uncensored LLM that I can build off of and wrap? With that said, this chat bot best friend isn't gonna be a significant other where you are s\*xting with or anything like that, like a platonic close best friend.
* What about the memory issue? Once again, it doesn't need to know everything about everything, but it needs to remember everything the user has ever told it. And be able to reference them like a best friend.
* The personality for the LLM also does not need to be different for each user - it's a single personality but remembers different things depending on user.
* Develop my own LLM -> if truly this is the best route, I'd imagine I'd have to raise VC funds but I think this would be too tough for me to compete in and raise the needed capital, so I'd rather not go this route. I've already asked around about funding and that's a whole lotta stress I don't want to take on atm.
* Any other routes? Or is the technology not there yet? Feel like most LLMs are designed for B2B not B2C.
Thanks!
| 2025-01-27T21:07:17 | https://www.reddit.com/r/LocalLLaMA/comments/1ibk3xq/which_llm_would_work_best_to_produce_a_best/ | MatthewNagy | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibk3xq | false | null | t3_1ibk3xq | /r/LocalLLaMA/comments/1ibk3xq/which_llm_would_work_best_to_produce_a_best/ | false | false | self | 3 | null |
Nvidia calls DeepSeek R1 model ‘an excellent AI advancement | 2 | 2025-01-27T21:08:38 | https://www.cnbc.com/2025/01/27/nvidia-calls-chinas-deepseek-r1-model-an-excellent-ai-advancement.html | bruhlmaocmonbro | cnbc.com | 1970-01-01T00:00:00 | 0 | {} | 1ibk563 | false | null | t3_1ibk563 | /r/LocalLLaMA/comments/1ibk563/nvidia_calls_deepseek_r1_model_an_excellent_ai/ | false | false | 2 | {'enabled': False, 'images': [{'id': 'KqtGGQXnDqSmrjM_RrtQ4Nfz7treiUVj4jJOLlwiaaQ', 'resolutions': [{'height': 60, 'url': 'https://external-preview.redd.it/GWj-XFmDoVWvTNTndrQzod7CldJdelytDZHRtYElnbw.jpg?width=108&crop=smart&auto=webp&s=b09ae4abd96820faf76bed316526d8f4bebe1bb8', 'width': 108}, {'height': 121, 'url': 'https://external-preview.redd.it/GWj-XFmDoVWvTNTndrQzod7CldJdelytDZHRtYElnbw.jpg?width=216&crop=smart&auto=webp&s=e41ce4b05efd01a0c1e20998638fd4fde5ca43ac', 'width': 216}, {'height': 180, 'url': 'https://external-preview.redd.it/GWj-XFmDoVWvTNTndrQzod7CldJdelytDZHRtYElnbw.jpg?width=320&crop=smart&auto=webp&s=3885733d83122424665e3ad5ec6feb1563eeabea', 'width': 320}, {'height': 360, 'url': 'https://external-preview.redd.it/GWj-XFmDoVWvTNTndrQzod7CldJdelytDZHRtYElnbw.jpg?width=640&crop=smart&auto=webp&s=1b4262326e1fbfe11b8fd0e4ee1e7a3c74f4a85e', 'width': 640}, {'height': 540, 'url': 'https://external-preview.redd.it/GWj-XFmDoVWvTNTndrQzod7CldJdelytDZHRtYElnbw.jpg?width=960&crop=smart&auto=webp&s=f149fa90a084ef4469d0cc7bf965a6a8ad2a2c31', 'width': 960}, {'height': 607, 'url': 'https://external-preview.redd.it/GWj-XFmDoVWvTNTndrQzod7CldJdelytDZHRtYElnbw.jpg?width=1080&crop=smart&auto=webp&s=9c64c2d83525bd3a1200be903b19d38d88e34dc0', 'width': 1080}], 'source': {'height': 1080, 'url': 'https://external-preview.redd.it/GWj-XFmDoVWvTNTndrQzod7CldJdelytDZHRtYElnbw.jpg?auto=webp&s=04f627e649b2b2a40fd23c2558f216d89c90a269', 'width': 1920}, 'variants': {}}]} |
||
M3 Max 36GB vs Mobile 4080 12GB for DeepSeek? | 1 | I have these two machines, dipping my toes into the Local world, what would be smarter to get started on?
What kind of computer or memory amount would run the 70b model pretty well? | 2025-01-27T21:13:13 | https://www.reddit.com/r/LocalLLaMA/comments/1ibk9ad/m3_max_36gb_vs_mobile_4080_12gb_for_deepseek/ | BahnMe | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibk9ad | false | null | t3_1ibk9ad | /r/LocalLLaMA/comments/1ibk9ad/m3_max_36gb_vs_mobile_4080_12gb_for_deepseek/ | false | false | self | 1 | null |
Meta is reportedly scrambling multiple ‘war rooms’ of engineers to figure out how DeepSeek’s AI is beating everyone else at a fraction of the price | 2,006 | From the article:
"Of the four war rooms Meta has created to respond to DeepSeek’s potential breakthrough, two teams will try to decipher how High-Flyer lowered the cost of training and running DeepSeek with the goal of using those tactics for Llama, the outlet reported citing one anonymous Meta employee.
Among the remaining two teams, one will try to find out which data DeepSeek used to train its model, and the other will consider how Llama can restructure its models based on attributes of the DeepSeek models, The Information reported."
I am actually excited by this. If Meta can figure it out, it means Llama 4 or 4.x will be substantially better. Hopefully we'll get a 70B dense model that's on part with DeepSeek.
| 2025-01-27T21:13:50 | https://fortune.com/2025/01/27/mark-zuckerberg-meta-llama-assembling-war-rooms-engineers-deepseek-ai-china/ | FullstackSensei | fortune.com | 1970-01-01T00:00:00 | 0 | {} | 1ibk9us | false | null | t3_1ibk9us | /r/LocalLLaMA/comments/1ibk9us/meta_is_reportedly_scrambling_multiple_war_rooms/ | false | false | 2,006 | {'enabled': False, 'images': [{'id': 'G5zOfVpgiiBKA1UkGy280JqoUxMYT-VTnzE0FVVLwmM', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/Brnl3ltRvrwiYwAXRD8-9ZQzXA_EE-2JvCrM0Zi5k8U.jpg?width=108&crop=smart&auto=webp&s=a0f34db9f7f8ac22052861f5396c063a39fec6de', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/Brnl3ltRvrwiYwAXRD8-9ZQzXA_EE-2JvCrM0Zi5k8U.jpg?width=216&crop=smart&auto=webp&s=5cd753c6247471ef3e3ea5e5927c2780f16f2452', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/Brnl3ltRvrwiYwAXRD8-9ZQzXA_EE-2JvCrM0Zi5k8U.jpg?width=320&crop=smart&auto=webp&s=f8e42de93aa9eb1793ed6573f07cec8f053c7f3b', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/Brnl3ltRvrwiYwAXRD8-9ZQzXA_EE-2JvCrM0Zi5k8U.jpg?width=640&crop=smart&auto=webp&s=1046aa83b70828043ace549a5075989da27f1ff4', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/Brnl3ltRvrwiYwAXRD8-9ZQzXA_EE-2JvCrM0Zi5k8U.jpg?width=960&crop=smart&auto=webp&s=6b6c210904071fbc19d381dfccd639a538f234a9', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/Brnl3ltRvrwiYwAXRD8-9ZQzXA_EE-2JvCrM0Zi5k8U.jpg?width=1080&crop=smart&auto=webp&s=fc5db41457699fa17ef317571f39061fcb09be89', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/Brnl3ltRvrwiYwAXRD8-9ZQzXA_EE-2JvCrM0Zi5k8U.jpg?auto=webp&s=9722aa44d3323a501f7c5b713e14c8c58e8d103b', 'width': 1200}, 'variants': {}}]} |
|
Deepseek Censorship | 4 | Everyone keeps taking about how Deepseek is censored, and my question is, is there any reason it can't just be abliterated to remove it's ability to refuse? Or is that not really the problem? | 2025-01-27T21:15:13 | https://www.reddit.com/r/LocalLLaMA/comments/1ibkb43/deepseek_censorship/ | Saint-Shroomie | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibkb43 | false | null | t3_1ibkb43 | /r/LocalLLaMA/comments/1ibkb43/deepseek_censorship/ | false | false | self | 4 | null |
A few thoughts on DeepSeek - and why you can't trust what you read | 0 | 2025-01-27T21:15:38 | https://thatstocksguy.substack.com/p/a-few-thoughts-on-deepseek | unixmachine | thatstocksguy.substack.com | 1970-01-01T00:00:00 | 0 | {} | 1ibkbiq | false | null | t3_1ibkbiq | /r/LocalLLaMA/comments/1ibkbiq/a_few_thoughts_on_deepseek_and_why_you_cant_trust/ | false | false | 0 | {'enabled': False, 'images': [{'id': 'f0W2ayy0ACmTUwwNtpF9oihpzdjCwqfLVo6DymIUT84', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/-eRw0T5fhfWkPakWZRdgDsuWm9styCKTXDhAh61UufQ.jpg?width=108&crop=smart&auto=webp&s=bca800383f14c7b3d96c6e2bd1eb3e8d73efbd62', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/-eRw0T5fhfWkPakWZRdgDsuWm9styCKTXDhAh61UufQ.jpg?width=216&crop=smart&auto=webp&s=6158a64f4aaf0761604aea7886aa8bd8e4ac0d95', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/-eRw0T5fhfWkPakWZRdgDsuWm9styCKTXDhAh61UufQ.jpg?width=320&crop=smart&auto=webp&s=0a74f6b3b0d3a428ebb677e1437289271cc3bd71', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/-eRw0T5fhfWkPakWZRdgDsuWm9styCKTXDhAh61UufQ.jpg?width=640&crop=smart&auto=webp&s=cf64e4b3b67f707436dde9b599db194b967c3502', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/-eRw0T5fhfWkPakWZRdgDsuWm9styCKTXDhAh61UufQ.jpg?width=960&crop=smart&auto=webp&s=9a38b32bcf919a55b267eb7d2232e1d83e32eeeb', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/-eRw0T5fhfWkPakWZRdgDsuWm9styCKTXDhAh61UufQ.jpg?width=1080&crop=smart&auto=webp&s=22e14d7ef06a5929ade726fad80ae8b5a398fdfe', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/-eRw0T5fhfWkPakWZRdgDsuWm9styCKTXDhAh61UufQ.jpg?auto=webp&s=f971c0ae85aeaa0149717d619a8e3be074b27880', 'width': 1200}, 'variants': {}}]} |
||
You did it, didn’t you, DeepSeek? | 0 | 😉 | 2025-01-27T21:17:00 | https://v.redd.it/7v35qmlsrlfe1 | abdessalaam | v.redd.it | 1970-01-01T00:00:00 | 0 | {} | 1ibkcqv | false | {'reddit_video': {'bitrate_kbps': 2400, 'dash_url': 'https://v.redd.it/7v35qmlsrlfe1/DASHPlaylist.mpd?a=1740604636%2CMzhmMzRmMzkxMTA5ZTYzNjg4ZTA2ZDI4NTMxZjBmNWM0MDBlYmNkZGFhZDAyZDYyYTJmNjZkNzQyYTZhZjgwYw%3D%3D&v=1&f=sd', 'duration': 42, 'fallback_url': 'https://v.redd.it/7v35qmlsrlfe1/DASH_720.mp4?source=fallback', 'has_audio': True, 'height': 1280, 'hls_url': 'https://v.redd.it/7v35qmlsrlfe1/HLSPlaylist.m3u8?a=1740604636%2CZjFhOWRiYWY0MWZjMmUxM2FhZWY5MTZhMjNhNjg4YmRlYzQyNzYwYzExZDM2YTFjYjg0MWEyYTE0NDY4MGJiYg%3D%3D&v=1&f=sd', 'is_gif': False, 'scrubber_media_url': 'https://v.redd.it/7v35qmlsrlfe1/DASH_96.mp4', 'transcoding_status': 'completed', 'width': 624}} | t3_1ibkcqv | /r/LocalLLaMA/comments/1ibkcqv/you_did_it_didnt_you_deepseek/ | false | false | 0 | {'enabled': False, 'images': [{'id': 'MWVxcDV1ZHNybGZlMT9i68f5RkH0VuYn4VAWpvQAVRR7RSLn_1CqMDv0zxgg', 'resolutions': [{'height': 216, 'url': 'https://external-preview.redd.it/MWVxcDV1ZHNybGZlMT9i68f5RkH0VuYn4VAWpvQAVRR7RSLn_1CqMDv0zxgg.png?width=108&crop=smart&format=pjpg&auto=webp&s=8824ea91c90733d31e7158ddfd79ce53053175b6', 'width': 108}, {'height': 432, 'url': 'https://external-preview.redd.it/MWVxcDV1ZHNybGZlMT9i68f5RkH0VuYn4VAWpvQAVRR7RSLn_1CqMDv0zxgg.png?width=216&crop=smart&format=pjpg&auto=webp&s=5ffce6b8b45bc06bf4d12ec149842d93380de2e4', 'width': 216}, {'height': 640, 'url': 'https://external-preview.redd.it/MWVxcDV1ZHNybGZlMT9i68f5RkH0VuYn4VAWpvQAVRR7RSLn_1CqMDv0zxgg.png?width=320&crop=smart&format=pjpg&auto=webp&s=b88ebd58008b83e621d8466f10f6d8f2d2ee4070', 'width': 320}, {'height': 1280, 'url': 'https://external-preview.redd.it/MWVxcDV1ZHNybGZlMT9i68f5RkH0VuYn4VAWpvQAVRR7RSLn_1CqMDv0zxgg.png?width=640&crop=smart&format=pjpg&auto=webp&s=39935f6333068e275142f61a72293af1cad2c8dc', 'width': 640}], 'source': {'height': 1820, 'url': 'https://external-preview.redd.it/MWVxcDV1ZHNybGZlMT9i68f5RkH0VuYn4VAWpvQAVRR7RSLn_1CqMDv0zxgg.png?format=pjpg&auto=webp&s=d0a2d49291409e6cd25c31fb6a61a91e60ddac2d', 'width': 886}, 'variants': {}}]} |
|
Best model for coding with large context window? | 1 | [removed] | 2025-01-27T21:22:46 | https://www.reddit.com/r/LocalLLaMA/comments/1ibkhut/best_model_for_coding_with_large_context_window/ | throwaway-weakguy | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibkhut | false | null | t3_1ibkhut | /r/LocalLLaMA/comments/1ibkhut/best_model_for_coding_with_large_context_window/ | false | false | self | 1 | null |
Best model for coding with large context window? | 3 |
I have a react app that I have made myself. And I was wondering if there is an LLM that can scan your codebase and do tasks with the context in mind?
For example:
I have a Header.tsx, and I want the LLM to look through my entire src folder to add appropriate code with guidance.
Obviously I can easily do this myself within a minute. Just to give an example, because without the entire codebase context, some tasks just give shitty code.
I have mid tier pc: R5 7600, 32GB DDR5, and RX 7800 XT 16GB
I have tried aider chat + qwen-2.5coder:14b and it frankly sucked for this purpose. It could only handle a few files at a time.
Adding the src folder and then asking questions would just make the LLM keep requesting the file in mind even though I added it to the chat. But if its less files then yea.
Anyway, its not a huuuuge project, only 2250 lines total.
Qwen 32b didnt work, deepsek r1:14b either (cant blame it is since its a reasoning model, got tired of the think process, plus i cant handle large sizes, and atm I dont wanna pay for an api since Im in the process of getting github pro student granted)
Im currently using gemini 1.5 pro with continue dev, it works quite nice tbh but I got a 503 so I had to use flash.
Oh and another task I would like to give is refactoring. Im a lazy person so when Im coding I just speed through and refactor later, for example all my html elements are divs. I would like to ask the LLM to replace divs with semantic elements where appropriate.
Or just clean the code and reduce
I apologize if it has been asked a lot, but Ive tried qwen coder which is apparently the best but it didnt work that well and my PC cant handle the 32B. | 2025-01-27T21:23:30 | https://www.reddit.com/r/LocalLLaMA/comments/1ibkihq/best_model_for_coding_with_large_context_window/ | cruud123 | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibkihq | false | null | t3_1ibkihq | /r/LocalLLaMA/comments/1ibkihq/best_model_for_coding_with_large_context_window/ | false | false | self | 3 | null |
Agentic performance for Deepseek | 2 | Are there any papers or benchmarks talking about it? For agentic tasks like automation, web search, etc… | 2025-01-27T21:25:04 | https://www.reddit.com/r/LocalLLaMA/comments/1ibkjuf/agentic_performance_for_deepseek/ | BlueeWaater | self.LocalLLaMA | 1970-01-01T00:00:00 | 0 | {} | 1ibkjuf | false | null | t3_1ibkjuf | /r/LocalLLaMA/comments/1ibkjuf/agentic_performance_for_deepseek/ | false | false | self | 2 | null |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.