One Year Since the โDeepSeek Momentโ
โข
28
AI & ML interests
The AI community building the future.
Recent Activity
View all activity
Papers
sangkuriang: A pseudo-spectral Python library for Korteweg-de Vries soliton simulation
FineVision: Open Data Is All You Need
Articles
akhaliqย
submitted
a
paper
to Daily Papers
1 day ago
IlyasMoutawwakilย
posted
an
update
1 day ago
Post
1484
After 2 months of refinement, I'm happy to announce that a lot of Transformers' modeling code is now significantly more torch-compile & export-friendly ๐ฅ
Why it had to be done ๐
PyTorch's Dynamo compiler is increasingly becoming the default interoperability layer for ML systems. Anything that relies on torch.export or torch.compile, from model optimization to cross-framework integrations, benefits directly when models can be captured as a single dynamo-traced graph !
Transformers models are now easier to:
โ๏ธ Compile end-to-end with torch.compile backends
๐ฆ Export reliably via torch.export and torch.onnx.export
๐ Deploy to ONNX / ONNX Runtime, Intel Corporation's OpenVINO, NVIDIA AutoDeploy (TRT-LLM), AMD's Quark, Meta's Executorch and more hardware-specific runtimes.
This work aims at unblocking entire TorchDynamo-based toolchains that rely on exporting Transformers across runtimes and accelerators.
We are doubling down on Transformers commitment to be a first-class citizen of the PyTorch ecosystem, more exportable, more optimizable, and easier to deploy everywhere.
There are definitely some edge-cases that we still haven't addressed so don't hesitate to try compiling / exporting your favorite transformers and to open issues / PRs.
PR in the comments ! More updates coming coming soon !
Why it had to be done ๐
PyTorch's Dynamo compiler is increasingly becoming the default interoperability layer for ML systems. Anything that relies on torch.export or torch.compile, from model optimization to cross-framework integrations, benefits directly when models can be captured as a single dynamo-traced graph !
Transformers models are now easier to:
โ๏ธ Compile end-to-end with torch.compile backends
๐ฆ Export reliably via torch.export and torch.onnx.export
๐ Deploy to ONNX / ONNX Runtime, Intel Corporation's OpenVINO, NVIDIA AutoDeploy (TRT-LLM), AMD's Quark, Meta's Executorch and more hardware-specific runtimes.
This work aims at unblocking entire TorchDynamo-based toolchains that rely on exporting Transformers across runtimes and accelerators.
We are doubling down on Transformers commitment to be a first-class citizen of the PyTorch ecosystem, more exportable, more optimizable, and easier to deploy everywhere.
There are definitely some edge-cases that we still haven't addressed so don't hesitate to try compiling / exporting your favorite transformers and to open issues / PRs.
PR in the comments ! More updates coming coming soon !
-
- 1 reply
sergiopaniegoย
posted
an
update
5 days ago
Post
1427
FunctionGemma Tuning Lab is a new no-code tool by
@google
that lets you fine-tune a model directly from the browser, with no coding knowledge required, using TRL behind the scenes.
blog: https://developers.googleblog.com/a-guide-to-fine-tuning-functiongemma/
try it out: google/functiongemma-tuning-lab
This example builds on a more advanced one for learning fine-tuning with SFT using TRL: https://ai.google.dev/gemma/docs/functiongemma/finetuning-with-functiongemma
blog: https://developers.googleblog.com/a-guide-to-fine-tuning-functiongemma/
try it out: google/functiongemma-tuning-lab
This example builds on a more advanced one for learning fine-tuning with SFT using TRL: https://ai.google.dev/gemma/docs/functiongemma/finetuning-with-functiongemma
- 1 reply
sergiopaniegoย
posted
an
update
8 days ago
Post
684
TRL v0.27.0 is out!! ๐ฅณ
It includes GDPO, the latest variant of GRPO for multi-reward RL โจ
GDPO decouples reward normalization to avoid reward collapse and improve per-reward convergence โ developed by
@sliuau @SimonX et al.
Explore the paper: GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization (2601.05242)
Explore the full set of changes here:
https://github.com/huggingface/trl/releases/tag/v0.27.0
It includes GDPO, the latest variant of GRPO for multi-reward RL โจ
GDPO decouples reward normalization to avoid reward collapse and improve per-reward convergence โ developed by
@sliuau @SimonX et al.
Explore the paper: GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization (2601.05242)
Explore the full set of changes here:
https://github.com/huggingface/trl/releases/tag/v0.27.0
akhaliqย
submitted
a
paper
to Daily Papers
8 days ago
akhaliqย
submitted
a
paper
to Daily Papers
10 days ago
sergiopaniegoย
posted
an
update
11 days ago
Post
2920
New REPL environment in OpenEnv available! โจ
Used in the Recursive Language Models (RLM) paper by Alex Zhang.
Ready for inference & post-training using trajectories. Handles long contexts:
> Run Python code in a sandbox
> Make recursive calls to LMs
> Explore data programmatically
> Return final result
Docs: https://meta-pytorch.org/OpenEnv/environments/repl/
Inference script: https://github.com/meta-pytorch/OpenEnv/blob/main/examples/repl_oolong_simple.py
Used in the Recursive Language Models (RLM) paper by Alex Zhang.
Ready for inference & post-training using trajectories. Handles long contexts:
> Run Python code in a sandbox
> Make recursive calls to LMs
> Explore data programmatically
> Return final result
Docs: https://meta-pytorch.org/OpenEnv/environments/repl/
Inference script: https://github.com/meta-pytorch/OpenEnv/blob/main/examples/repl_oolong_simple.py
sergiopaniegoย
posted
an
update
12 days ago
Post
392
Recursive Language Models (RLM) is a new interface for LLMs with cool ideas by Alex Zhang!
โ ๏ธ LLMs struggle with long prompts โ attention overload & lost info
๐ RLMs inspect, split & call themselves on chunks, then aggregate results
โ Handles millions of tokens, reduces noise, improves reasoning
๐ก System prompt guides recursion
๐ฏ RLM trajectories can be used for RL training or distillation (OpenEnv+TRL!!)
We're adding it to OpenEnv (with Kashif Rasul): https://github.com/meta-pytorch/OpenEnv/pull/282
More resources:
> Paper: Recursive Language Models (2512.24601)
> Paper blog: https://alexzhang13.github.io/blog/2025/rlm/
> RLM repo: https://github.com/alexzhang13/rlm
โ ๏ธ LLMs struggle with long prompts โ attention overload & lost info
๐ RLMs inspect, split & call themselves on chunks, then aggregate results
โ Handles millions of tokens, reduces noise, improves reasoning
๐ก System prompt guides recursion
๐ฏ RLM trajectories can be used for RL training or distillation (OpenEnv+TRL!!)
We're adding it to OpenEnv (with Kashif Rasul): https://github.com/meta-pytorch/OpenEnv/pull/282
More resources:
> Paper: Recursive Language Models (2512.24601)
> Paper blog: https://alexzhang13.github.io/blog/2025/rlm/
> RLM repo: https://github.com/alexzhang13/rlm
- 2 replies
nielsrย
submitted
a
paper
to Daily Papers
12 days ago
sergiopaniegoย
posted
an
update
16 days ago
Post
2216
New GRPO + TRL free Colab notebook out! ๐ฅ
Fine-tune 7B+ models on T4 GPUs thanks to a ton of memory optimizations for GRPO
7B model uses only 9.2 GB VRAM (~7ร reduction) ๐คฏ
Try the notebook here ๐ https://colab.research.google.com/github/huggingface/trl/blob/main/examples/notebooks/grpo_trl_lora_qlora.ipynb
Fine-tune 7B+ models on T4 GPUs thanks to a ton of memory optimizations for GRPO
7B model uses only 9.2 GB VRAM (~7ร reduction) ๐คฏ
Try the notebook here ๐ https://colab.research.google.com/github/huggingface/trl/blob/main/examples/notebooks/grpo_trl_lora_qlora.ipynb
akhaliqย
submitted
a
paper
to Daily Papers
16 days ago
sergiopaniegoย
posted
an
update
22 days ago
Post
2572
The list of hands-on notebooks (some beginner-friendly!) to get started with fine-tuning using TRL keeps growing!!
โข SFT
โข GRPO
โข Tool calling & agents
โข RL environments with OpenEnv
โข LLMs and VLMs
โจ Many run on FREE Colab, making it super easy to get started fast!
https://github.com/huggingface/trl/tree/main/examples/notebooks
โข SFT
โข GRPO
โข Tool calling & agents
โข RL environments with OpenEnv
โข LLMs and VLMs
โจ Many run on FREE Colab, making it super easy to get started fast!
https://github.com/huggingface/trl/tree/main/examples/notebooks
akhaliqย
submitted
2
papers
to Daily Papers
22 days ago
sergiopaniegoย
posted
an
update
25 days ago
Post
467
As the year comes to an end, itโs a good moment to catch up on some of the best long-form pieces published by the
@huggingface
team.
Iโve gathered them all here if you want to read or save them for later:
https://huggingface.co/collections/sergiopaniego/research-and-long-form-blog-posts
Iโve gathered them all here if you want to read or save them for later:
https://huggingface.co/collections/sergiopaniego/research-and-long-form-blog-posts
sergiopaniegoย
posted
an
update
26 days ago
Post
2296
This super detailed tutorial by
@Paulescu
is pure gold ๐ช "Fine-tuning a Small Language Model for browser control with GRPO and OpenEnv"
LFM2-350M ( @LiquidAI ) + BrowserGym (OpenEnv) + GRPO (TRL) for learning browser control ๐ค
https://paulabartabajo.substack.com/p/fine-tuning-lfm2-350m-for-browser
LFM2-350M ( @LiquidAI ) + BrowserGym (OpenEnv) + GRPO (TRL) for learning browser control ๐ค
https://paulabartabajo.substack.com/p/fine-tuning-lfm2-350m-for-browser
craffelย
authored
a
paper
29 days ago
nielsrย
submitted
a
paper
to Daily Papers
about 1 month ago
sergiopaniegoย
posted
an
update
about 1 month ago
Post
486
if youโre on holidays ๐ and want some reading, here are blogs I contributed to this year:
๐ฆ VLMs in TRL: https://huggingface.co/blog/trl-vlm-alignment
๐ฆ VLMs in 2025: https://huggingface.co/blog/vlms-2025
๐พ tokenization in transformers v5: https://huggingface.co/blog/tokenizers
๐ธ faster transformers: https://huggingface.co/blog/faster-transformers
๐ฆ VLMs in TRL: https://huggingface.co/blog/trl-vlm-alignment
๐ฆ VLMs in 2025: https://huggingface.co/blog/vlms-2025
๐พ tokenization in transformers v5: https://huggingface.co/blog/tokenizers
๐ธ faster transformers: https://huggingface.co/blog/faster-transformers
sergiopaniegoย
posted
an
update
about 1 month ago
Post
2002
The Christmas holidays are here! ๐
Thinking about learning something new in AI?
@huggingface offers 12 FREE courses covering all the relevant topics, for every level of experience. A great challenge for the holidays (and worth saving for later ๐)
Letโs explore them!
๐ง ๐๐๐ ๐๐ผ๐๐ฟ๐๐ฒ: large language models with HF tools
https://huggingface.co/learn/llm-course
๐ค ๐๐ด๐ฒ๐ป๐๐ ๐๐ผ๐๐ฟ๐๐ฒ: build and deploy AI agents
https://huggingface.co/learn/agents-course
๐จ ๐๐ถ๐ณ๐ณ๐๐๐ถ๐ผ๐ป ๐๐ผ๐๐ฟ๐๐ฒ: diffusion models with ๐ค Diffusers
https://huggingface.co/learn/diffusion-course
๐ ๐๐๐ฑ๐ถ๐ผ ๐๐ผ๐๐ฟ๐๐ฒ: transformers for audio tasks
https://huggingface.co/learn/audio-course
๐ฎ ๐๐ฒ๐ฒ๐ฝ ๐ฅ๐ ๐๐ผ๐๐ฟ๐๐ฒ: deep reinforcement learning
https://huggingface.co/learn/deep-rl-course
๐๏ธ ๐๐ผ๐บ๐บ๐๐ป๐ถ๐๐ ๐๐ผ๐บ๐ฝ๐๐๐ฒ๐ฟ ๐ฉ๐ถ๐๐ถ๐ผ๐ป ๐๐ผ๐๐ฟ๐๐ฒ: modern computer vision with HF
https://huggingface.co/learn/computer-vision-course
๐ฆพ ๐ฅ๐ผ๐ฏ๐ผ๐๐ถ๐ฐ๐ ๐๐ผ๐๐ฟ๐๐ฒ (๐๐ฒ๐ฅ๐ผ๐ฏ๐ผ๐): learning-based robotics
https://huggingface.co/learn/robotics-course
๐งฉ ๐ ๐๐ฃ ๐๐ผ๐๐ฟ๐๐ฒ: Model Context Protocol explained
https://huggingface.co/learn/mcp-course
๐งช ๐ ๐ฆ๐บ๐ผ๐น ๐๐ผ๐๐ฟ๐๐ฒ: post-training AI models
https://huggingface.co/learn/a-smol-course
๐น๏ธ ๐ ๐ ๐ณ๐ผ๐ฟ ๐๐ฎ๐บ๐ฒ๐: AI in game development
https://huggingface.co/learn/ml-for-games-course
๐ง ๐ ๐ ๐ณ๐ผ๐ฟ ๐ฏ๐: machine learning for 3D data
https://huggingface.co/learn/ml-for-3d-course
๐ ๐ข๐ฝ๐ฒ๐ป-๐ฆ๐ผ๐๐ฟ๐ฐ๐ฒ ๐๐ ๐๐ผ๐ผ๐ธ๐ฏ๐ผ๐ผ๐ธ: practical AI notebooks
https://huggingface.co/learn/cookbook
All of them can be found here: https://huggingface.co/learn
Thinking about learning something new in AI?
@huggingface offers 12 FREE courses covering all the relevant topics, for every level of experience. A great challenge for the holidays (and worth saving for later ๐)
Letโs explore them!
๐ง ๐๐๐ ๐๐ผ๐๐ฟ๐๐ฒ: large language models with HF tools
https://huggingface.co/learn/llm-course
๐ค ๐๐ด๐ฒ๐ป๐๐ ๐๐ผ๐๐ฟ๐๐ฒ: build and deploy AI agents
https://huggingface.co/learn/agents-course
๐จ ๐๐ถ๐ณ๐ณ๐๐๐ถ๐ผ๐ป ๐๐ผ๐๐ฟ๐๐ฒ: diffusion models with ๐ค Diffusers
https://huggingface.co/learn/diffusion-course
๐ ๐๐๐ฑ๐ถ๐ผ ๐๐ผ๐๐ฟ๐๐ฒ: transformers for audio tasks
https://huggingface.co/learn/audio-course
๐ฎ ๐๐ฒ๐ฒ๐ฝ ๐ฅ๐ ๐๐ผ๐๐ฟ๐๐ฒ: deep reinforcement learning
https://huggingface.co/learn/deep-rl-course
๐๏ธ ๐๐ผ๐บ๐บ๐๐ป๐ถ๐๐ ๐๐ผ๐บ๐ฝ๐๐๐ฒ๐ฟ ๐ฉ๐ถ๐๐ถ๐ผ๐ป ๐๐ผ๐๐ฟ๐๐ฒ: modern computer vision with HF
https://huggingface.co/learn/computer-vision-course
๐ฆพ ๐ฅ๐ผ๐ฏ๐ผ๐๐ถ๐ฐ๐ ๐๐ผ๐๐ฟ๐๐ฒ (๐๐ฒ๐ฅ๐ผ๐ฏ๐ผ๐): learning-based robotics
https://huggingface.co/learn/robotics-course
๐งฉ ๐ ๐๐ฃ ๐๐ผ๐๐ฟ๐๐ฒ: Model Context Protocol explained
https://huggingface.co/learn/mcp-course
๐งช ๐ ๐ฆ๐บ๐ผ๐น ๐๐ผ๐๐ฟ๐๐ฒ: post-training AI models
https://huggingface.co/learn/a-smol-course
๐น๏ธ ๐ ๐ ๐ณ๐ผ๐ฟ ๐๐ฎ๐บ๐ฒ๐: AI in game development
https://huggingface.co/learn/ml-for-games-course
๐ง ๐ ๐ ๐ณ๐ผ๐ฟ ๐ฏ๐: machine learning for 3D data
https://huggingface.co/learn/ml-for-3d-course
๐ ๐ข๐ฝ๐ฒ๐ป-๐ฆ๐ผ๐๐ฟ๐ฐ๐ฒ ๐๐ ๐๐ผ๐ผ๐ธ๐ฏ๐ผ๐ผ๐ธ: practical AI notebooks
https://huggingface.co/learn/cookbook
All of them can be found here: https://huggingface.co/learn