amd/Instella-3B-Instruct
Text Generation
•
Updated
•
1.18k
•
33
Nagori
MohammedNaeem
AI & ML interests
None yet
Recent Activity
liked
a model
3 days ago
amd/Instella-3B-Instruct
liked
a Space
8 days ago
Qwen/QwQ-32B-Demo
reacted
to
MohamedRashad's
post
with ❤️
10 days ago
I think we have released the best Arabic model under 25B at least based on https://huggingface.co/spaces/inceptionai/AraGen-Leaderboard
Yehia = https://huggingface.co/ALLaM-AI/ALLaM-7B-Instruct-preview + GRPO
and its ranked number one model under the 25B parameter size mark.
Now, i said "i think" not "i am sure" because this model used the same metric of evaluation the AraGen developers use (the 3C3H) as a reward model to improve its responses and this sparks the question. Is this something good for users or is it another type of overfitting that we don't want ?
I don't know if this is a good thing or a bad thing but what i know is that you can try it from here:
https://huggingface.co/spaces/Navid-AI/Yehia-7B-preview
or Download it for your personal experiments from here:
https://huggingface.co/Navid-AI/Yehia-7B-preview
Ramadan Kareem 🌙
Organizations
None yet
MohammedNaeem's activity
liked
a
model
3 days ago
liked
a
Space
8 days ago
reacted to
MohamedRashad's
post with ❤️
10 days ago
Post
3411
I think we have released the best Arabic model under 25B at least based on
inceptionai/AraGen-Leaderboard
Yehia = ALLaM-AI/ALLaM-7B-Instruct-preview + GRPO
and its ranked number one model under the 25B parameter size mark.
Now, i said "i think" not "i am sure" because this model used the same metric of evaluation the AraGen developers use (the 3C3H) as a reward model to improve its responses and this sparks the question. Is this something good for users or is it another type of overfitting that we don't want ?
I don't know if this is a good thing or a bad thing but what i know is that you can try it from here:
Navid-AI/Yehia-7B-preview
or Download it for your personal experiments from here:
Navid-AI/Yehia-7B-preview
Ramadan Kareem 🌙
Yehia = ALLaM-AI/ALLaM-7B-Instruct-preview + GRPO
and its ranked number one model under the 25B parameter size mark.
Now, i said "i think" not "i am sure" because this model used the same metric of evaluation the AraGen developers use (the 3C3H) as a reward model to improve its responses and this sparks the question. Is this something good for users or is it another type of overfitting that we don't want ?
I don't know if this is a good thing or a bad thing but what i know is that you can try it from here:
Navid-AI/Yehia-7B-preview
or Download it for your personal experiments from here:
Navid-AI/Yehia-7B-preview
Ramadan Kareem 🌙
- 1 reply
upvoted
a
collection
25 days ago
liked
a
model
25 days ago
[MODELS] Discussion
639
#372 opened about 1 year ago
by
victor
reacted to
lewtun's
post with ❤️
about 1 month ago
Post
4867
Introducing OpenR1-Math-220k!
open-r1/OpenR1-Math-220k
The community has been busy distilling DeepSeek-R1 from inference providers, but we decided to have a go at doing it ourselves from scratch 💪
What’s new compared to existing reasoning datasets?
♾ Based on AI-MO/NuminaMath-1.5: we focus on math reasoning traces and generate answers for problems in NuminaMath 1.5, an improved version of the popular NuminaMath-CoT dataset.
🐳 800k R1 reasoning traces: We generate two answers for 400k problems using DeepSeek R1. The filtered dataset contains 220k problems with correct reasoning traces.
📀 512 H100s running locally: Instead of relying on an API, we leverage vLLM and SGLang to run generations locally on our science cluster, generating 180k reasoning traces per day.
⏳ Automated filtering: We apply Math Verify to only retain problems with at least one correct answer. We also leverage Llama3.3-70B-Instruct as a judge to retrieve more correct examples (e.g for cases with malformed answers that can’t be verified with a rules-based parser)
📊 We match the performance of DeepSeek-Distill-Qwen-7B by finetuning Qwen-7B-Math-Instruct on our dataset.
🔎 Read our blog post for all the nitty gritty details: https://huggingface.co/blog/open-r1/update-2
open-r1/OpenR1-Math-220k
The community has been busy distilling DeepSeek-R1 from inference providers, but we decided to have a go at doing it ourselves from scratch 💪
What’s new compared to existing reasoning datasets?
♾ Based on AI-MO/NuminaMath-1.5: we focus on math reasoning traces and generate answers for problems in NuminaMath 1.5, an improved version of the popular NuminaMath-CoT dataset.
🐳 800k R1 reasoning traces: We generate two answers for 400k problems using DeepSeek R1. The filtered dataset contains 220k problems with correct reasoning traces.
📀 512 H100s running locally: Instead of relying on an API, we leverage vLLM and SGLang to run generations locally on our science cluster, generating 180k reasoning traces per day.
⏳ Automated filtering: We apply Math Verify to only retain problems with at least one correct answer. We also leverage Llama3.3-70B-Instruct as a judge to retrieve more correct examples (e.g for cases with malformed answers that can’t be verified with a rules-based parser)
📊 We match the performance of DeepSeek-Distill-Qwen-7B by finetuning Qwen-7B-Math-Instruct on our dataset.
🔎 Read our blog post for all the nitty gritty details: https://huggingface.co/blog/open-r1/update-2
upvoted
an
article
about 1 month ago
Article
Open-source DeepResearch – Freeing our search agents
•
1.16k
reacted to
chansung's
post with ❤️
about 1 month ago
Post
2022
Simple summary on DeepSeek AI's Janus-Pro: A fresh take on multimodal AI!
It builds on its predecessor, Janus, by tweaking the training methodology rather than the model architecture. The result? Improved performance in understanding and generating multimodal data.
Janus-Pro uses a three-stage training strategy, similar to Janus, but with key modifications:
✦ Stage 1 & 2: Focus on separate training for specific objectives, rather than mixing data.
✦ Stage 3: Fine-tuning with a careful balance of multimodal data.
Benchmarks show Janus-Pro holds its own against specialized models like TokenFlow XL and MetaMorph, and other multimodal models like SD3 Medium and DALL-E 3.
The main limitation? Low image resolution (384x384). However, this seems like a strategic choice to focus on establishing a solid "recipe" for multimodal models. Future work will likely leverage this recipe and increased computing power to achieve higher resolutions.
It builds on its predecessor, Janus, by tweaking the training methodology rather than the model architecture. The result? Improved performance in understanding and generating multimodal data.
Janus-Pro uses a three-stage training strategy, similar to Janus, but with key modifications:
✦ Stage 1 & 2: Focus on separate training for specific objectives, rather than mixing data.
✦ Stage 3: Fine-tuning with a careful balance of multimodal data.
Benchmarks show Janus-Pro holds its own against specialized models like TokenFlow XL and MetaMorph, and other multimodal models like SD3 Medium and DALL-E 3.
The main limitation? Low image resolution (384x384). However, this seems like a strategic choice to focus on establishing a solid "recipe" for multimodal models. Future work will likely leverage this recipe and increased computing power to achieve higher resolutions.
liked
a
model
about 2 months ago
upvoted
a
collection
about 2 months ago
liked
a
model
about 2 months ago
liked
a
Space
2 months ago
upvoted
a
collection
3 months ago
upvoted
a
paper
3 months ago
liked
a
model
3 months ago
upvoted
2
collections
3 months ago
liked
a
Space
3 months ago
upvoted
a
collection
3 months ago