SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model
Paper
โข
2502.02737
โข
Published
โข
58
HuggingFaceM4
Enterprise
company
AI & ML interests
None defined yet.
Recent Activity
View all activity
HuggingFaceM4's activity
anditoย
authored
a
paper
about 1 hour ago
lewtunย
authored
a
paper
about 1 hour ago
thomwolfย
authored
a
paper
about 1 hour ago
lvwerraย
authored
a
paper
about 4 hours ago
Post
5530
Introducing ๐ผ๐ฝ๐ฒ๐ป ๐๐ฒ๐ฒ๐ฝ-๐ฅ๐ฒ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต by Hugging Face! ๐ฅ
OpenAI's latest agentic app Deep Research seems really good... But it's closed, as usual.
โฑ๏ธ So with a team of cracked colleagues, we set ourselves a 24hours deadline to replicate and open-source Deep Research! โฑ๏ธ
โก๏ธ We built open-Deep-Research, an entirely open agent that can: navigate the web autonomously, scroll and search through pages, download and manipulate files, run calculation on data...
We aimed for the best performance: are the agent's answers really rigorous?
On GAIA benchmark, Deep Research had 67% accuracy on the validation set.
โก๏ธ open Deep Research is at 55% (powered by o1), it is:
- the best pass@1 solution submitted
- the best open solution ๐ช๐ช
And it's only getting started ! Please jump in, drop PRs, and let's bring it to the top !
Read the blog post ๐ https://huggingface.co/blog/open-deep-research
OpenAI's latest agentic app Deep Research seems really good... But it's closed, as usual.
โฑ๏ธ So with a team of cracked colleagues, we set ourselves a 24hours deadline to replicate and open-source Deep Research! โฑ๏ธ
โก๏ธ We built open-Deep-Research, an entirely open agent that can: navigate the web autonomously, scroll and search through pages, download and manipulate files, run calculation on data...
We aimed for the best performance: are the agent's answers really rigorous?
On GAIA benchmark, Deep Research had 67% accuracy on the validation set.
โก๏ธ open Deep Research is at 55% (powered by o1), it is:
- the best pass@1 solution submitted
- the best open solution ๐ช๐ช
And it's only getting started ! Please jump in, drop PRs, and let's bring it to the top !
Read the blog post ๐ https://huggingface.co/blog/open-deep-research
Post
420
I was quoted in an article about the French Lucie AI in La Presse. While I love the name for obvious reasons ๐ there were still a lot of problems with the model and how and when it was deployed. Nevertheless seeing new smaller models being developed is an exciting direction for the next years of AI development to come!
https://www.lapresse.ca/affaires/techno/2025-02-02/radioscopie/lucie-l-ia-francaise-qui-ne-passe-pas-le-test.php
Also fun to see my comments in French.
https://www.lapresse.ca/affaires/techno/2025-02-02/radioscopie/lucie-l-ia-francaise-qui-ne-passe-pas-le-test.php
Also fun to see my comments in French.
Post
1576
Seeing AI develop has been a wild ride, from trying to explain why we'd bother to generate a single sentence with a *neural network* to explaining that AI is not a magic, all-knowing box. The recent weeks and months have been a lot of talking about how AI works; to policy makers, to other developers, but also and mainly friends and family without a technical background.
Yesterday, the first provisions of the EU AI Act came into force, and one of the the key highlights are the AI literacy requirements for organisations deploying AI systems. This isn't just a box-ticking exercise. Ensuring that employees and stakeholders understand AI systems is crucial for fostering responsible and transparent AI development. From recognising biases to understanding model limitations, AI literacy empowers individuals to engage critically with these technologies and make informed decisions.
In the context of Hugging Face, AI literacy has many facets: allowing more people to contribute to AI development, providing courses and documentation to ensuring access is possible, and accessible AI tools that empower users to better understand how AI systems function. This isn't just a regulatory milestone; itโs an opportunity to foster a culture where AI literacy becomes foundational, enabling stakeholders to recognise biases, assess model limitations, and engage critically with technology.
Embedding these principles into daily practice, and eventually extending our learnings in AI literacy to the general public, is essential for building trustworthy AI that aligns with societal values.
Yesterday, the first provisions of the EU AI Act came into force, and one of the the key highlights are the AI literacy requirements for organisations deploying AI systems. This isn't just a box-ticking exercise. Ensuring that employees and stakeholders understand AI systems is crucial for fostering responsible and transparent AI development. From recognising biases to understanding model limitations, AI literacy empowers individuals to engage critically with these technologies and make informed decisions.
In the context of Hugging Face, AI literacy has many facets: allowing more people to contribute to AI development, providing courses and documentation to ensuring access is possible, and accessible AI tools that empower users to better understand how AI systems function. This isn't just a regulatory milestone; itโs an opportunity to foster a culture where AI literacy becomes foundational, enabling stakeholders to recognise biases, assess model limitations, and engage critically with technology.
Embedding these principles into daily practice, and eventually extending our learnings in AI literacy to the general public, is essential for building trustworthy AI that aligns with societal values.
- 1 reply
Post
291
From ancient medical ethics to modern AI challenges, the journey of consent represents one of humanity's most fascinating ethical evolutions. In my latest blog post, I explore how we've moved from medical paternalism to a new frontier where AI capabilities force us to rethink consent.
The "consent gap" in AI is real: while we can approve initial data use, AI systems can generate countless unforeseen applications of our personal information. It's like signing a blank check without knowing all possible amounts that could be filled in.
Should we reimagine consent for the AI age? Perhaps we need dynamic consent systems that evolve alongside AI capabilities, similar to how healthcare transformed from physician-centered authority to patient autonomy.
Curious to hear your thoughts: how can we balance technological innovation with meaningful user sovereignty over digital identity?
Read more: https://huggingface.co/blog/giadap/evolution-of-consent
The "consent gap" in AI is real: while we can approve initial data use, AI systems can generate countless unforeseen applications of our personal information. It's like signing a blank check without knowing all possible amounts that could be filled in.
Should we reimagine consent for the AI age? Perhaps we need dynamic consent systems that evolve alongside AI capabilities, similar to how healthcare transformed from physician-centered authority to patient autonomy.
Curious to hear your thoughts: how can we balance technological innovation with meaningful user sovereignty over digital identity?
Read more: https://huggingface.co/blog/giadap/evolution-of-consent
Post
3712
This week in open AI was ๐ฅ Let's recap! ๐ค
merve/january-31-releases-679a10669bd4030090c5de4d
LLMs ๐ฌ
> Huge: AllenAI released new Tรผlu models that outperform DeepSeek R1 using Reinforcement Learning with Verifiable Reward (RLVR) based on Llama 3.1 405B ๐ฅ
> Mistral AI is back to open-source with their "small" 24B models (base & SFT), with Apache 2.0 license ๐ฑ
> Alibaba Qwen released their 1M context length models Qwen2.5-Instruct-1M, great for agentic use with Apache 2.0 license ๐ฅ
> Arcee AI released Virtuoso-medium, 32.8B LLMs distilled from DeepSeek V3 with dataset of 5B+ tokens
> Velvet-14B is a new family of 14B Italian LLMs trained on 10T tokens in six languages
> OpenThinker-7B is fine-tuned version of Qwen2.5-7B-Instruct on OpenThoughts dataset
VLMs & vision ๐
> Alibaba Qwen is back with Qwen2.5VL, amazing new capabilities ranging from agentic computer use to zero-shot localization ๐ฅ
> NVIDIA released new series of Eagle2 models with 1B and 9B sizes
> DeepSeek released Janus-Pro, new any-to-any model (image-text generation from image-text input) with MIT license
> BEN2 is a new background removal model with MIT license!
Audio ๐ฃ๏ธ
> YuE is a new open-source music generation foundation model, lyrics-to-song generation
Codebase ๐ฉ๐ปโ๐ป
> We are open-sourcing our SmolVLM training and eval codebase! https://github.com/huggingface/smollm/tree/main/vision
> Open-R1 is open-source reproduction of R1 by @huggingface science team https://huggingface.co/blog/open-r1
LLMs ๐ฌ
> Huge: AllenAI released new Tรผlu models that outperform DeepSeek R1 using Reinforcement Learning with Verifiable Reward (RLVR) based on Llama 3.1 405B ๐ฅ
> Mistral AI is back to open-source with their "small" 24B models (base & SFT), with Apache 2.0 license ๐ฑ
> Alibaba Qwen released their 1M context length models Qwen2.5-Instruct-1M, great for agentic use with Apache 2.0 license ๐ฅ
> Arcee AI released Virtuoso-medium, 32.8B LLMs distilled from DeepSeek V3 with dataset of 5B+ tokens
> Velvet-14B is a new family of 14B Italian LLMs trained on 10T tokens in six languages
> OpenThinker-7B is fine-tuned version of Qwen2.5-7B-Instruct on OpenThoughts dataset
VLMs & vision ๐
> Alibaba Qwen is back with Qwen2.5VL, amazing new capabilities ranging from agentic computer use to zero-shot localization ๐ฅ
> NVIDIA released new series of Eagle2 models with 1B and 9B sizes
> DeepSeek released Janus-Pro, new any-to-any model (image-text generation from image-text input) with MIT license
> BEN2 is a new background removal model with MIT license!
Audio ๐ฃ๏ธ
> YuE is a new open-source music generation foundation model, lyrics-to-song generation
Codebase ๐ฉ๐ปโ๐ป
> We are open-sourcing our SmolVLM training and eval codebase! https://github.com/huggingface/smollm/tree/main/vision
> Open-R1 is open-source reproduction of R1 by @huggingface science team https://huggingface.co/blog/open-r1
- 1 reply
Post
2433
Now you can launch a code agent directly from your terminal!
โจ ๐๐๐๐๐๐๐๐๐ "๐๐๐๐ ๐๐๐๐" directly launches a CodeAgent
โถ๏ธ This also works with web agents (replace ๐๐๐๐๐๐๐๐๐ with ๐ ๐๐๐๐๐๐๐) thanks to @merve !
๐พ Another treat from smolagents release 1.7.0:
Now agents have a memory mechanism, enabling many possibilities like replaying the last run with ๐๐๐๐๐.๐๐๐๐๐๐ข(), thank you @clefourrier !
Check the release notes here ๐ https://github.com/huggingface/smolagents/releases/tag/v1.7.0
โจ ๐๐๐๐๐๐๐๐๐ "๐๐๐๐ ๐๐๐๐" directly launches a CodeAgent
โถ๏ธ This also works with web agents (replace ๐๐๐๐๐๐๐๐๐ with ๐ ๐๐๐๐๐๐๐) thanks to @merve !
๐พ Another treat from smolagents release 1.7.0:
Now agents have a memory mechanism, enabling many possibilities like replaying the last run with ๐๐๐๐๐.๐๐๐๐๐๐ข(), thank you @clefourrier !
Check the release notes here ๐ https://github.com/huggingface/smolagents/releases/tag/v1.7.0
davanstrienย
posted
an
update
8 days ago
Post
1683
Why choose between strong LLM reasoning and efficient models?
Use DeepSeek to generate high-quality training data, then distil that knowledge into ModernBERT answerdotai/ModernBERT-base for fast, efficient classification.
Blog post: https://danielvanstrien.xyz/posts/2025/deepseek/distil-deepseek-modernbert.html
Use DeepSeek to generate high-quality training data, then distil that knowledge into ModernBERT answerdotai/ModernBERT-base for fast, efficient classification.
Blog post: https://danielvanstrien.xyz/posts/2025/deepseek/distil-deepseek-modernbert.html
dylanebertย
posted
an
update
9 days ago
Post
2913
I made a 1 minute video explaining the DeepSeek situation
R1: deepseek-ai/DeepSeek-R1
Janus Pro: deepseek-ai/Janus-Pro-7B
R1: deepseek-ai/DeepSeek-R1
Janus Pro: deepseek-ai/Janus-Pro-7B
- 3 replies
Post
3549
๐ง๐ต๐ฒ ๐๐๐ฏ ๐๐ฒ๐น๐ฐ๐ผ๐บ๐ฒ๐ ๐ฒ๐
๐๐ฒ๐ฟ๐ป๐ฎ๐น ๐ถ๐ป๐ณ๐ฒ๐ฟ๐ฒ๐ป๐ฐ๐ฒ ๐ฝ๐ฟ๐ผ๐๐ถ๐ฑ๐ฒ๐ฟ๐!
โ Hosting our own inference was not enough: now the Hub 4 new inference providers: fal, Replicate, SambaNova Systems, & Together AI.
Check model cards on the Hub: you can now, in 1 click, use inference from various providers (cf video demo)
Their inference can also be used through our Inference API client. There, you can use either your custom provider key, or your HF token, then billing will be handled directly on your HF account, as a way to centralize all expenses.
๐ธ Also, PRO users get 2$ inference credits per month!
Read more in the announcement ๐ https://huggingface.co/blog/inference-providers
โ Hosting our own inference was not enough: now the Hub 4 new inference providers: fal, Replicate, SambaNova Systems, & Together AI.
Check model cards on the Hub: you can now, in 1 click, use inference from various providers (cf video demo)
Their inference can also be used through our Inference API client. There, you can use either your custom provider key, or your HF token, then billing will be handled directly on your HF account, as a way to centralize all expenses.
๐ธ Also, PRO users get 2$ inference credits per month!
Read more in the announcement ๐ https://huggingface.co/blog/inference-providers
- 1 reply
davanstrienย
posted
an
update
9 days ago
Post
1823
Updated the ColPali Query Generator Space
davanstrien/ColPali-Query-Generator to use
Qwen/Qwen2.5-VL-7B-Instruct.
Given an input image, it generates several queries along with explanations to justify them. This approach can generate synthetic data for fine-tuning ColPali models.
Given an input image, it generates several queries along with explanations to justify them. This approach can generate synthetic data for fine-tuning ColPali models.
Post
6906
AI is not a zero-sum game. Open-source AI is the tide that lifts all boats!
davanstrienย
posted
an
update
10 days ago
Post
1981
๐ Big step for multilingual AI data!
The Hugging Face community has rated educational content in languages spoken by 1.6 billion people! New additions:
โข Japanese
โข Italian
โข Old High German
Learn more and contribute: https://huggingface.co/blog/davanstrien/fineweb2-community
These ratings can help enhance training data for major world languages.
The Hugging Face community has rated educational content in languages spoken by 1.6 billion people! New additions:
โข Japanese
โข Italian
โข Old High German
Learn more and contribute: https://huggingface.co/blog/davanstrien/fineweb2-community
These ratings can help enhance training data for major world languages.
- 1 reply
Post
9876
We are reproducing the full DeepSeek R1 data and training pipeline so everybody can use their recipe. Instead of doing it in secret we can do it together in the open!
๐งช Step 1: replicate the R1-Distill models by distilling a high-quality reasoning corpus from DeepSeek-R1.
๐ง Step 2: replicate the pure RL pipeline that DeepSeek used to create R1-Zero. This will involve curating new, large-scale datasets for math, reasoning, and code.
๐ฅ Step 3: show we can go from base model -> SFT -> RL via multi-stage training.
Follow along: https://github.com/huggingface/open-r1
๐งช Step 1: replicate the R1-Distill models by distilling a high-quality reasoning corpus from DeepSeek-R1.
๐ง Step 2: replicate the pure RL pipeline that DeepSeek used to create R1-Zero. This will involve curating new, large-scale datasets for math, reasoning, and code.
๐ฅ Step 3: show we can go from base model -> SFT -> RL via multi-stage training.
Follow along: https://github.com/huggingface/open-r1
- 5 replies
Post
4987
Oof, what a week! ๐ฅต So many things have happened, let's recap!
merve/jan-24-releases-6793d610774073328eac67a9
Multimodal ๐ฌ
- We have released SmolVLM -- tiniest VLMs that come in 256M and 500M, with it's retrieval models ColSmol for multimodal RAG ๐
- UI-TARS are new models by ByteDance to unlock agentic GUI control ๐คฏ in 2B, 7B and 72B
- Alibaba DAMO lab released VideoLlama3, new video LMs that come in 2B and 7B
- MiniMaxAI released Minimax-VL-01, where decoder is based on MiniMax-Text-01 456B MoE model with long context
- Dataset: Yale released a new benchmark called MMVU
- Dataset: CAIS released Humanity's Last Exam (HLE) a new challenging MM benchmark
LLMs ๐
- DeepSeek-R1 & DeepSeek-R1-Zero: gigantic 660B reasoning models by DeepSeek, and six distilled dense models, on par with o1 with MIT license! ๐คฏ
- Qwen2.5-Math-PRM: new math models by Qwen in 7B and 72B
- NVIDIA released AceMath and AceInstruct, new family of models and their datasets (SFT and reward ones too!)
Audio ๐ฃ๏ธ
- Llasa is a new speech synthesis model based on Llama that comes in 1B,3B, and 8B
- TangoFlux is a new audio generation model trained from scratch and aligned with CRPO
Image/Video/3D Generation โฏ๏ธ
- Flex.1-alpha is a new 8B pre-trained diffusion model by ostris similar to Flux
- tencent released Hunyuan3D-2, new 3D asset generation from images
Multimodal ๐ฌ
- We have released SmolVLM -- tiniest VLMs that come in 256M and 500M, with it's retrieval models ColSmol for multimodal RAG ๐
- UI-TARS are new models by ByteDance to unlock agentic GUI control ๐คฏ in 2B, 7B and 72B
- Alibaba DAMO lab released VideoLlama3, new video LMs that come in 2B and 7B
- MiniMaxAI released Minimax-VL-01, where decoder is based on MiniMax-Text-01 456B MoE model with long context
- Dataset: Yale released a new benchmark called MMVU
- Dataset: CAIS released Humanity's Last Exam (HLE) a new challenging MM benchmark
LLMs ๐
- DeepSeek-R1 & DeepSeek-R1-Zero: gigantic 660B reasoning models by DeepSeek, and six distilled dense models, on par with o1 with MIT license! ๐คฏ
- Qwen2.5-Math-PRM: new math models by Qwen in 7B and 72B
- NVIDIA released AceMath and AceInstruct, new family of models and their datasets (SFT and reward ones too!)
Audio ๐ฃ๏ธ
- Llasa is a new speech synthesis model based on Llama that comes in 1B,3B, and 8B
- TangoFlux is a new audio generation model trained from scratch and aligned with CRPO
Image/Video/3D Generation โฏ๏ธ
- Flex.1-alpha is a new 8B pre-trained diffusion model by ostris similar to Flux
- tencent released Hunyuan3D-2, new 3D asset generation from images
- 7 replies
dylanebertย
posted
an
update
13 days ago
Post
624
โ๏ธ Convert
i've created a simple space to convert .ply gaussian splat files to .splat format
dylanebert/ply-to-splat
.ply to .splati've created a simple space to convert .ply gaussian splat files to .splat format
dylanebert/ply-to-splat