Gemma 3 Technical Report
Paper
โข
2503.19786
โข
Published
โข
28
Notebooks-explorers
community
AI & ML interests
Request to join this organization to beta-test notebooks on Hugging Face!
Recent Activity
View all activity
notebooks-explorers's activity
osansevieroย
authored
a
paper
about 11 hours ago
Post
3205
simple guide on the recipe for GRPO on Open-R1 which is built on top of TRL
I think FastAPI wrapper of vLLM with WeightSyncWorker is pretty cool feature. Also, we have many predefined reward functions out of the box!
I think FastAPI wrapper of vLLM with WeightSyncWorker is pretty cool feature. Also, we have many predefined reward functions out of the box!
- 5 replies
Post
3212
So many open releases at Hugging Face past week ๐คฏ recapping all here โคต๏ธ
merve/march-21-releases-67dbe10e185f199e656140ae
๐ Multimodal
> Mistral AI released a 24B vision LM, both base and instruction FT versions, sota ๐ฅ (OS)
> with IBM we released SmolDocling, a sota 256M document parser with Apache 2.0 license (OS)
> SpatialLM is a new vision LM that outputs 3D bounding boxes, comes with 0.5B (QwenVL based) and 1B (Llama based) variants
> SkyWork released SkyWork-R1V-38B, new vision reasoning model (OS)
๐ฌ LLMs
> NVIDIA released new Nemotron models in 49B and 8B with their post-training dataset
> LG released EXAONE, new reasoning models in 2.4B, 7.8B and 32B
> Dataset: Glaive AI released a new reasoning dataset of 22M+ examples
> Dataset: NVIDIA released new helpfulness dataset HelpSteer3
> Dataset: OpenManusRL is a new agent dataset based on ReAct framework (OS)
> Open-R1 team released OlympicCoder, new competitive coder model in 7B and 32B
> Dataset: GeneralThought-430K is a new reasoning dataset (OS)
๐ผ๏ธ Image Generation/Computer Vision
> Roboflow released RF-DETR, new real-time sota object detector (OS) ๐ฅ
> YOLOE is a new real-time zero-shot object detector with text and visual prompts ๐ฅน
> Stability AI released Stable Virtual Camera, a new novel view synthesis model
> Tencent released Hunyuan3D-2mini, new small and fast 3D asset generation model
> ByteDance released InfiniteYou, new realistic photo generation model
> StarVector is a new 8B model that generates svg from images
> FlexWorld is a new model that expands 3D views (OS)
๐ค Audio
> Sesame released CSM-1B new speech generation model (OS)
๐ค Robotics
> NVIDIA released GR00T, new robotics model for generalized reasoning and skills, along with the dataset
*OS ones have Apache 2.0 or MIT license
๐ Multimodal
> Mistral AI released a 24B vision LM, both base and instruction FT versions, sota ๐ฅ (OS)
> with IBM we released SmolDocling, a sota 256M document parser with Apache 2.0 license (OS)
> SpatialLM is a new vision LM that outputs 3D bounding boxes, comes with 0.5B (QwenVL based) and 1B (Llama based) variants
> SkyWork released SkyWork-R1V-38B, new vision reasoning model (OS)
๐ฌ LLMs
> NVIDIA released new Nemotron models in 49B and 8B with their post-training dataset
> LG released EXAONE, new reasoning models in 2.4B, 7.8B and 32B
> Dataset: Glaive AI released a new reasoning dataset of 22M+ examples
> Dataset: NVIDIA released new helpfulness dataset HelpSteer3
> Dataset: OpenManusRL is a new agent dataset based on ReAct framework (OS)
> Open-R1 team released OlympicCoder, new competitive coder model in 7B and 32B
> Dataset: GeneralThought-430K is a new reasoning dataset (OS)
๐ผ๏ธ Image Generation/Computer Vision
> Roboflow released RF-DETR, new real-time sota object detector (OS) ๐ฅ
> YOLOE is a new real-time zero-shot object detector with text and visual prompts ๐ฅน
> Stability AI released Stable Virtual Camera, a new novel view synthesis model
> Tencent released Hunyuan3D-2mini, new small and fast 3D asset generation model
> ByteDance released InfiniteYou, new realistic photo generation model
> StarVector is a new 8B model that generates svg from images
> FlexWorld is a new model that expands 3D views (OS)
๐ค Audio
> Sesame released CSM-1B new speech generation model (OS)
๐ค Robotics
> NVIDIA released GR00T, new robotics model for generalized reasoning and skills, along with the dataset
*OS ones have Apache 2.0 or MIT license
Post
2097
We introduce FAT5 (Flash Attention T5) โก
An implementation of T5 in PyTorch with UL2 objective optimized for GPGPU for both training and inference thanks to 13 different optimizations.
The main one is that we have designed a CUDA kernel to expand the Flash Attention by @tridao with RPE biases and supports other PE such as RoPE, ALiBi or FIRE.
The result kernel is 2 times faster than a SPDA implementation.
We also use Triton kernels to optimize certain parts of the architecture, such as the cross-entropy and RMSNorm layer.
The various kernels have been carefully built to be compatible with BF16 and torch.compile to go even faster and achieve efficient pretraining.
All other optimizations are described in a ๐ subsequent blog post available on @huggingface ๐ค: CATIE-AQ/FAT5-report.
This methodology enabled us to efficiently pretrain as a proof of concept a FAT5 with 147M parameters in French in a reasonable time (1,461H for 419B tokens), with limited resources (1 A100 i.e. a computational budget of ~ โฌ1,900) and a low carbon footprint (13.5kg eq CO2).
The model's weights are also available on Hugging Face: CATIE-AQ/FAT5-small.
Not very useful in practice, it's a PoC and not an instructed model (it's planned for later).
All the code is available on GitHub if you want to pretrain your own model in your own language or for a specific domain: https://github.com/catie-aq/flashT5 โญ
Ending by indicating that was a joint project with @BorisAlbar at hf.co/CATIE-AQ.
An implementation of T5 in PyTorch with UL2 objective optimized for GPGPU for both training and inference thanks to 13 different optimizations.
The main one is that we have designed a CUDA kernel to expand the Flash Attention by @tridao with RPE biases and supports other PE such as RoPE, ALiBi or FIRE.
The result kernel is 2 times faster than a SPDA implementation.
We also use Triton kernels to optimize certain parts of the architecture, such as the cross-entropy and RMSNorm layer.
The various kernels have been carefully built to be compatible with BF16 and torch.compile to go even faster and achieve efficient pretraining.
All other optimizations are described in a ๐ subsequent blog post available on @huggingface ๐ค: CATIE-AQ/FAT5-report.
This methodology enabled us to efficiently pretrain as a proof of concept a FAT5 with 147M parameters in French in a reasonable time (1,461H for 419B tokens), with limited resources (1 A100 i.e. a computational budget of ~ โฌ1,900) and a low carbon footprint (13.5kg eq CO2).
The model's weights are also available on Hugging Face: CATIE-AQ/FAT5-small.
Not very useful in practice, it's a PoC and not an instructed model (it's planned for later).
All the code is available on GitHub if you want to pretrain your own model in your own language or for a specific domain: https://github.com/catie-aq/flashT5 โญ
Ending by indicating that was a joint project with @BorisAlbar at hf.co/CATIE-AQ.
Post
2493
Mistral AI Small 3.1 24B is not only commercial free but also the best model in a single GPU deployment.
I packed up all the information you need to know in a single picture. Hope this helps! :)
I packed up all the information you need to know in a single picture. Hope this helps! :)
- 1 reply
mrfakenameย
posted
an
update
10 days ago
Post
2516
GGUF quants (text-only) for the new Mistral Small 3.1 24B are now live:
mrfakename/mistral-small-3.1-24b-instruct-2503-gguf
mrfakename/mistral-small-3.1-24b-instruct-2503-gguf
mrfakenameย
posted
an
update
11 days ago
Post
2200
Converted the new Mistral Small 3.1 models to HF format (currently text-only, no vision):
Instruct: mrfakename/mistral-small-3.1-24b-instruct-2503-hf
Base: mrfakename/mistral-small-3.1-24b-base-2503-hf
GGUF quants coming soon!
Instruct: mrfakename/mistral-small-3.1-24b-instruct-2503-hf
Base: mrfakename/mistral-small-3.1-24b-base-2503-hf
GGUF quants coming soon!
Post
1556
Gemma 3 Release in a nutshell
(seems like function calling is not supported whereas the announcement said so)
(seems like function calling is not supported whereas the announcement said so)
Post
2685
Important notice ๐จ
For Inference Providers who have built support for our Billing API (currently: Fal, Novita, HF-Inference โ with more coming soon), we've started enabling Pay as you go (=PAYG)
What this means is that you can use those Inference Providers beyond the free included credits, and they're charged to your HF account.
You can see it on this view: any provider that does not have a "Billing disabled" badge, is PAYG-compatible.
For Inference Providers who have built support for our Billing API (currently: Fal, Novita, HF-Inference โ with more coming soon), we've started enabling Pay as you go (=PAYG)
What this means is that you can use those Inference Providers beyond the free included credits, and they're charged to your HF account.
You can see it on this view: any provider that does not have a "Billing disabled" badge, is PAYG-compatible.
- 2 replies
davanstrienย
posted
an
update
29 days ago
Post
2795
๐ Introducing "Hugging Face Dataset Spotlight" ๐
I'm excited to share the first episode of our AI-generated podcast series focusing on nice datasets from the Hugging Face Hub!
This first episode explores mathematical reasoning datasets:
- SynthLabsAI/Big-Math-RL-Verified: Over 250,000 rigorously verified problems spanning multiple difficulty levels and mathematical domains
- open-r1/OpenR1-Math-220k: 220,000 math problems with multiple reasoning traces, verified for accuracy using Math Verify and Llama-3.3-70B models.
- facebook/natural_reasoning: 1.1 million general reasoning questions carefully deduplicated and decontaminated from existing benchmarks, showing superior scaling effects when training models like Llama3.1-8B-Instruct.
Plus a bonus segment on bespokelabs/bespoke-manim!
https://www.youtube.com/watch?v=-TgmRq45tW4
I'm excited to share the first episode of our AI-generated podcast series focusing on nice datasets from the Hugging Face Hub!
This first episode explores mathematical reasoning datasets:
- SynthLabsAI/Big-Math-RL-Verified: Over 250,000 rigorously verified problems spanning multiple difficulty levels and mathematical domains
- open-r1/OpenR1-Math-220k: 220,000 math problems with multiple reasoning traces, verified for accuracy using Math Verify and Llama-3.3-70B models.
- facebook/natural_reasoning: 1.1 million general reasoning questions carefully deduplicated and decontaminated from existing benchmarks, showing superior scaling effects when training models like Llama3.1-8B-Instruct.
Plus a bonus segment on bespokelabs/bespoke-manim!
https://www.youtube.com/watch?v=-TgmRq45tW4
davanstrienย
posted
an
update
29 days ago
Post
3635
Quick POC: Turn a Hugging Face dataset card into a short podcast introducing the dataset using all open models.
I think I'm the only weirdo who would enjoy listening to something like this though ๐
Here is an example for eth-nlped/stepverify
I think I'm the only weirdo who would enjoy listening to something like this though ๐
Here is an example for eth-nlped/stepverify
- 2 replies
davanstrienย
posted
an
update
about 1 month ago
Post
2607
Hacked together a way to log trl GRPO training completions to a ๐ค dataset repo. This allows you to:
- Track rewards from multiple reward functions
- Treat the completion and rewards from training as a "proper" dataset and do EDA
- Share results for open science
The implementation is super hacky, but I'm curious if people would find this useful.
To push completions to the Hub, you just need two extra parameters:
Example dataset: davanstrien/test-logs
Colab: https://colab.research.google.com/drive/1wzBFPVthRYYTp-mEYlznLg_e_0Za1M3g
- Track rewards from multiple reward functions
- Treat the completion and rewards from training as a "proper" dataset and do EDA
- Share results for open science
The implementation is super hacky, but I'm curious if people would find this useful.
To push completions to the Hub, you just need two extra parameters:
log_completions=True
log_completions_hub_repo='your-username/repo-name'Example dataset: davanstrien/test-logs
Colab: https://colab.research.google.com/drive/1wzBFPVthRYYTp-mEYlznLg_e_0Za1M3g
Post
6428
Google just released PaliGemma 2 Mix: new versatile instruction vision language models ๐ฅ
> Three new models: 3B, 10B, 28B with res 224, 448 ๐
> Can do vision language tasks with open-ended prompts, understand documents, and segment or detect anything ๐คฏ
Read more https://huggingface.co/blog/paligemma2mix
Try the demo google/paligemma2-10b-mix
All models are here google/paligemma-2-mix-67ac6a251aaf3ee73679dcc4
> Three new models: 3B, 10B, 28B with res 224, 448 ๐
> Can do vision language tasks with open-ended prompts, understand documents, and segment or detect anything ๐คฏ
Read more https://huggingface.co/blog/paligemma2mix
Try the demo google/paligemma2-10b-mix
All models are here google/paligemma-2-mix-67ac6a251aaf3ee73679dcc4
davanstrienย
posted
an
update
about 1 month ago
Post
2248
Dataset descriptions for trending Hugging Face datasets? Powered by a Smol model
davanstrien/Smol-Hub-tldr
Post
7302
I was thinking i need to step up my game on training Flux LoRas models, time to have some fun ! โ๏ธ
Expect a new drop per week on aesthetics that catched my attention, here are 3 of them that worked really well !
fffiloni/cute-comic-800
fffiloni/carbo-800
fffiloni/oniric-750
Expect a new drop per week on aesthetics that catched my attention, here are 3 of them that worked really well !
fffiloni/cute-comic-800
fffiloni/carbo-800
fffiloni/oniric-750
Post
4773
Your weekly recap of open AI is here, and it's packed with models!
merve/feb-14-releases-67af876b404cc27c6d837767
๐ Multimodal
> OpenGVLab released InternVideo 2.5 Chat models, new video LMs with long context
> AIDC released Ovis2 model family along with Ovis dataset, new vision LMs in different sizes (1B, 2B, 4B, 8B, 16B, 34B), with video and OCR support
> ColQwenStella-2b is a multilingual visual retrieval model that is sota in it's size
> Hoags-2B-Exp is a new multilingual vision LM with contextual reasoning, long context video understanding
๐ฌ LLMs
A lot of math models!
> Open-R1 team released OpenR1-Math-220k large scale math reasoning dataset, along with Qwen2.5-220K-Math fine-tuned on the dataset, OpenR1-Qwen-7B
> Nomic AI released new Nomic Embed multilingual retrieval model, a MoE with 500 params with 305M active params, outperforming other models
> DeepScaleR-1.5B-Preview is a new DeepSeek-R1-Distill fine-tune using distributed RL on math
> LIMO is a new fine-tune of Qwen2.5-32B-Instruct on Math
๐ฃ๏ธ Audio
> Zonos-v0.1 is a new family of speech recognition models, which contains the model itself and embeddings
๐ผ๏ธ Vision and Image Generation
> We have ported DepthPro of Apple to transformers for your convenience!
> illustrious-xl-v1.0 is a new illustration generation model
๐ Multimodal
> OpenGVLab released InternVideo 2.5 Chat models, new video LMs with long context
> AIDC released Ovis2 model family along with Ovis dataset, new vision LMs in different sizes (1B, 2B, 4B, 8B, 16B, 34B), with video and OCR support
> ColQwenStella-2b is a multilingual visual retrieval model that is sota in it's size
> Hoags-2B-Exp is a new multilingual vision LM with contextual reasoning, long context video understanding
๐ฌ LLMs
A lot of math models!
> Open-R1 team released OpenR1-Math-220k large scale math reasoning dataset, along with Qwen2.5-220K-Math fine-tuned on the dataset, OpenR1-Qwen-7B
> Nomic AI released new Nomic Embed multilingual retrieval model, a MoE with 500 params with 305M active params, outperforming other models
> DeepScaleR-1.5B-Preview is a new DeepSeek-R1-Distill fine-tune using distributed RL on math
> LIMO is a new fine-tune of Qwen2.5-32B-Instruct on Math
๐ฃ๏ธ Audio
> Zonos-v0.1 is a new family of speech recognition models, which contains the model itself and embeddings
๐ผ๏ธ Vision and Image Generation
> We have ported DepthPro of Apple to transformers for your convenience!
> illustrious-xl-v1.0 is a new illustration generation model
- 3 replies
davanstrienย
posted
an
update
about 1 month ago
Post
1939
How do you make 1M+ Hugging Face models & datasets more discoverable?
davanstrien/Smol-Hub-tldr!
I fine-tuned HuggingFaceTB/SmolLM2-360M to generate one-line summaries from a model or dataset README.
Its own self-description?
"A model for generating concise summaries of model & dataset cards from the Hugging Face Hub"
The goal? Make it easier to find the right models and datasets for your specific needs. It's already powering a semantic search for datasets Space.
It's still a WIP but thanks to @loubnabnl , @anton-l , @eliebak et al, for cooking such a nice base model for fine-tuning small, efficient models for specific domains and tasks. ๐
davanstrien/Smol-Hub-tldr!
I fine-tuned HuggingFaceTB/SmolLM2-360M to generate one-line summaries from a model or dataset README.
Its own self-description?
"A model for generating concise summaries of model & dataset cards from the Hugging Face Hub"
The goal? Make it easier to find the right models and datasets for your specific needs. It's already powering a semantic search for datasets Space.
It's still a WIP but thanks to @loubnabnl , @anton-l , @eliebak et al, for cooking such a nice base model for fine-tuning small, efficient models for specific domains and tasks. ๐
davanstrienย
posted
an
update
about 1 month ago
Post
1376
Made some significant updates to my ๐ค semantic datasets search app. If you love falling into a wiki black hole, you might like this...
https://huggingface.co/spaces/librarian-bots/huggingface-datasets-semantic-search
https://huggingface.co/spaces/librarian-bots/huggingface-datasets-semantic-search
Post
3148
Interesting releases in open AI this week, let's recap ๐ค
merve/feb-7-releases-67a5f7d7f172d8bfe0dd66f4
๐ค Robotics
> Pi0, first open-source foundation vision-language action model was released in Le Robot (Apache 2.0)
๐ฌ LLMs
> Groundbreaking: s1 is simpler approach to test-time scaling, the release comes with small s1K dataset of 1k question-reasoning trace pairs (from Gemini-Thinking Exp) they fine-tune Qwen2.5-32B-Instruct to get s1-32B, outperforming o1-preview on math ๐คฏ s1-32B and s1K is out!
> Adyen released DABstep, a new benchmark along with it's leaderboard demo for agents doing data analysis
> Krutrim released Krutrim-2 instruct, new 12B model based on NeMo12B trained and aligned on Indic languages, a new multilingual sentence embedding model (based on STSB-XLM-R), and a translation model for Indic languages
๐ Multimodal
> PKU released Align-DS-V, a model aligned using their new technique called LLF for all modalities (image-text-audio), along with the dataset Align Anything
> OLA-7B is a new any-to-any model by Tencent that can take text, image, video, audio data with context window of 32k tokens and output text and speech in English and Chinese
> Krutrim released Chitrarth, a new vision language model for Indic languages and English
๐ผ๏ธ Vision
> BiRefNet_HR is a new higher resolution BiRefNet for background removal
๐ฃ๏ธ Audio
> kyutai released Hibiki, it's a real-time speech-to-speech translation model ๐คฏ it's available for French-English translation
> Krutrim released Dhwani, a new STT model for Indic languages
> They also release a new dataset for STT-TTS
๐ผ๏ธ Image Generation
> Lumina released Lumina-Image-2.0, a 2B parameter-flow based DiT for text to image generation
> Tencent released Hunyuan3D-2, a 3D asset generation model based on DiT and Hunyuan3D-Paint
> boreal-hl-v1 is a new boring photorealistic image generation LoRA based on Hunyuan
๐ค Robotics
> Pi0, first open-source foundation vision-language action model was released in Le Robot (Apache 2.0)
๐ฌ LLMs
> Groundbreaking: s1 is simpler approach to test-time scaling, the release comes with small s1K dataset of 1k question-reasoning trace pairs (from Gemini-Thinking Exp) they fine-tune Qwen2.5-32B-Instruct to get s1-32B, outperforming o1-preview on math ๐คฏ s1-32B and s1K is out!
> Adyen released DABstep, a new benchmark along with it's leaderboard demo for agents doing data analysis
> Krutrim released Krutrim-2 instruct, new 12B model based on NeMo12B trained and aligned on Indic languages, a new multilingual sentence embedding model (based on STSB-XLM-R), and a translation model for Indic languages
๐ Multimodal
> PKU released Align-DS-V, a model aligned using their new technique called LLF for all modalities (image-text-audio), along with the dataset Align Anything
> OLA-7B is a new any-to-any model by Tencent that can take text, image, video, audio data with context window of 32k tokens and output text and speech in English and Chinese
> Krutrim released Chitrarth, a new vision language model for Indic languages and English
๐ผ๏ธ Vision
> BiRefNet_HR is a new higher resolution BiRefNet for background removal
๐ฃ๏ธ Audio
> kyutai released Hibiki, it's a real-time speech-to-speech translation model ๐คฏ it's available for French-English translation
> Krutrim released Dhwani, a new STT model for Indic languages
> They also release a new dataset for STT-TTS
๐ผ๏ธ Image Generation
> Lumina released Lumina-Image-2.0, a 2B parameter-flow based DiT for text to image generation
> Tencent released Hunyuan3D-2, a 3D asset generation model based on DiT and Hunyuan3D-Paint
> boreal-hl-v1 is a new boring photorealistic image generation LoRA based on Hunyuan
Post
2345
IBM released
ibm-granite/granite-vision-3.1-2b-preview, a small vision LM with impressive performance on different tasks ๐ฎ๐ฅ
it comes with transformers and vLLM support from the get-go ๐
you can run it in Colab T4, so I built a notebook to put it to test, find it here: https://github.com/merveenoyan/smol-vision/blob/main/inference_gists/IBM_Granite_Vision.ipynb
it comes with transformers and vLLM support from the get-go ๐
you can run it in Colab T4, so I built a notebook to put it to test, find it here: https://github.com/merveenoyan/smol-vision/blob/main/inference_gists/IBM_Granite_Vision.ipynb