Hugging Face's logo

llmc
community
https://github.com/karpathy/llm.c
Activity Feed Request to join this org

AI & ML interests

LLM training in simple, pure C/CUDA

Recent Activity

eliebak  authored a paper 2 months ago
The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text
alielfilali01  authored a paper 2 months ago
Llama-3-Nanda-10B-Chat: An Open Generative Large Language Model for Hindi
1024m  authored a paper 2 months ago
Uncovering Cultural Representation Disparities in Vision-Language Models
View all activity

alielfilali01 
posted an update about 14 hours ago
eliebak 
posted an update 15 days ago
view post
Post
4508
Kimi K2 tech report is full of gems as always. Here are my notes on it:

> MuonClip: Pretty crazy how after 70k the training stabilizes and the QK-clip is basically inactive. There is also no loss in perf with QK-clip which is not trivial at all (at small scale but with aggressive threshold). Also a cool explanation of why muon makes the logit explode in appendix E (tl;dr is that muon makes the singular value of the update matrix higher)
> Sparsity scaling laws to justify their ratio, they have a very solid training infra that allows the model to be trained at this sparsity level, they could have increased even more but as sparsity increases the training becomes less efficient.
> They diminish the number of attention heads to make it more efficient for long context since attention heads are a big bottleneck for long context. They also remove 2 of the 3 "first dense" layers in the dsv3 arch.

With the sparsity and attention heads (divided by 2) they achieve 83% increased flops compared to deepseek v3 arch at 128k.

> Data: Rephrasing is KEY. They do a lot more synthetic data generation and rephrase their corpus to have different styles, for longer documents they do it by chunk. I'm (half) surprised by the fact that ONLY 1 epoch (assuming same number of training tokens I think?) of data rephrased 10 times has better accuracy than 10 epochs of the same data rephrased once.
> They do rewriting for Math and Knowledge, for Math they apply the ShallowMath recipe and instruct the model to rephrase in a "learning note" style
> They talk about diversity and probably have some internal stuff/eval to test that, as always still a bit unclear for me how to properly measure that.

The infra is also very nice, quick summary:
> PP=16 (1F1B schedule, a bit custom), EP=16, zero1
> No FP8 computation but for storage of specific layers, selective recomputation for inexpensive block, activation offloading to CPU
eliebak 
authored a paper 2 months ago
alielfilali01 
authored a paper 2 months ago
1024m 
authored a paper 2 months ago
alielfilali01 
posted an update 3 months ago
1024m 
authored 3 papers 4 months ago
eliebak 
authored a paper 4 months ago
eliebak 
posted an update 5 months ago
view post
Post
2066
Google just dropped an exciting technical report for the brand-new Gemma3 model! 🚀 Here are my personal notes highlighting the most intriguing architectural innovations, design choices, and insights from this release:

1) Architecture choices:
> No more softcaping, replace by QK-Norm
> Both Pre AND Post Norm
> Wider MLP than Qwen2.5, ~ same depth
> SWA with 5:1 and 1024 (very small and cool ablation on the paper!)
> No MLA to save KV cache, SWA do the job!

2) Long context
> Only increase the rope in the global layer (to 1M)
> Confirmation that it's harder to do long context for smol models, no 128k for the 1B
> Pretrained with 32k context? seems very high
> No yarn nor llama3 like rope extension

3) Distillation
> Only keep te first 256 logits for the teacher
> Ablation on the teacher gap (tl;dr you need some "patience" to see that using a small teacher is better)
> On policy distillation yeahh (by
@agarwl_
et al), not sure if the teacher gap behave the same here, curious if someone have more info?

4) Others
> Checkpoint with QAT, that's very cool
> RL using improve version of BOND, WARM/WARP good excuse to look at
@ramealexandre
papers
> Only use Zero3, no TP/PP if i understand correctly ?
> Training budget relatively similar than gemma2
  • 1 reply
·
alielfilali01 
posted an update 6 months ago
view post
Post
1066
🚨 Arabic LLM Evaluation 🚨

Few models join the ranking of https://huggingface.co/spaces/inceptionai/AraGen-Leaderboard Today.

The new MistralAI model, Saba, is quite impressive, Top10 ! Well done @arthurmensch and team.

Sadly Mistral did not follow its strategy about public weights this time, we hope this changes soon and we get the model with a permissive license.

We added other Mistral models and apparently, we have been sleeping on mistralai/Mistral-Large-Instruct-2411 !

Another impressive model that joined the ranking today is ALLaM-AI/ALLaM-7B-Instruct-preview. After a long wait finally ALLaM is here and it is IMPRESSIVE given its size !

ALLaM is ranked on OALL/Open-Arabic-LLM-Leaderboard as well.
eliebak 
authored 2 papers 6 months ago
alielfilali01 
posted an update 7 months ago
view post
Post
2149
3C3H AraGen Leaderboard welcomes today deepseek-ai/DeepSeek-V3 and 12 other models (including the late gpt-3.5 💀) to the ranking of best LLMs in Arabic !


Observations:
- DeepSeek-v3 ranked 3rd and only Open model among the top 5 !

- A 14B open model ( Qwen/Qwen2.5-14B-Instruct) outperforms gpt-3.5-turbo-0125 (from last year). This shows how much we came in advancing and supporting Arabic presence within the LLM ecosystem !

- Contrary to what observed in likelihood-acc leaderboards (like OALL/Open-Arabic-LLM-Leaderboard) further finetuned models like maldv/Qwentile2.5-32B-Instruct actually decreased the performance compared to the original model Qwen/Qwen2.5-32B-Instruct.
It's worth to note that the decrease is statiscally insignificant which imply that at best, the out-domain finetuning do not really hurts the model original capabilities acquired during pretraining.
Previous work addressed this (finetuning VS pretraining) but more investigation in this regard is required (any PhDs here ? This could be your question ...)


Check out the latest rankings: https://huggingface.co/spaces/inceptionai/AraGen-Leaderboard
alielfilali01 
posted an update 7 months ago
view post
Post
2061
~75% on the challenging GPQA with only 40M parameters 🔥🥳

GREAT ACHIEVEMENT ! Or is it ?

This new Work, "Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation", take out the mystery about many models i personally suspected their results. Speacially on leaderboards other than the english one, Like the Open Arabic LLM Leaderbaord OALL/Open-Arabic-LLM-Leaderboard.

The authors of this work, first started by training a model on the GPQA data, which, unsurprisingly, led to the model achieving 100% performance.

Afterward, they trained what they referred to as a 'legitimate' model on legitimate data (MedMCQA). However, they introduced a distillation loss from the earlier, 'cheated' model.

What they discovered was fascinating: the knowledge of GPQA leaked through this distillation loss, even though the legitimate model was never explicitly trained on GPQA during this stage.

This raises important questions about the careful use of distillation in model training, especially when the training data is opaque. As they demonstrated, it’s apparently possible to (intentionally or unintentionally) leak test data through this method.

Find out more: Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation (2412.15255)
  • 1 reply
·
alielfilali01 
posted an update 8 months ago
view post
Post
3553
Unpopular opinion: Open Source takes courage to do !

Not everyone is brave enough to release what they have done (the way they've done it) to the wild to be judged !
It really requires a high level of "knowing wth are you doing" ! It's kind of a super power !

Cheers to the heroes here who see this!
·
alielfilali01 
posted an update 8 months ago
view post
Post
1602
Apparently i forgot to put this here !

Well, this is a bit late but consider given our recent blog a read if you are interested in Evaluation.

You don't have to be into Arabic NLP in order to read it, the main contribution we are introducing is a new evaluation measure for NLG. We made the fisrt application of this measure on Arabic for now and we will be working with colleagues from the community to expand it to other languages.

Blog:
Rethinking LLM Evaluation with 3C3H: AraGen Benchmark and Leaderboard
https://huggingface.co/blog/leaderboard-3c3h-aragen

Space:
https://huggingface.co/spaces/inceptionai/AraGen-Leaderboard

Give it a read and let me know your thoughts 🤗
alielfilali01 
posted an update 9 months ago
view post
Post
2262
Unpopular opinion : o1-preview is more stupid than 4o and Qwen2.5-72B-Instruct in extremely underrated !
  • 2 replies
·
1024m 
authored a paper 10 months ago