nltpt

Enterprise

Activity Feed

AI & ML interests

None defined yet.

Recent Activity

clefourrier authored a paper about 1 month ago

Global MMLU: Understanding and Addressing Cultural and Linguistic Biases in Multilingual Evaluation

reach-vb new activity about 1 month ago

meta-llama/Llama-3.3-70B-Instruct:Update the "How to use" instructions.

reach-vb updated a model about 1 month ago

meta-llama/Llama-3.3-70B-Instruct

View all activity

nltpt's activity

lewtun

posted an update 6 days ago

Post

3161

I was initially pretty sceptical about Meta's Coconut paper [1] because the largest perf gains were reported on toy linguistic problems. However, these results on machine translation are pretty impressive!

https://x.com/casper_hansen_/status/1875872309996855343

Together with the recent PRIME method [2] for scaling RL, reasoning for open models is looking pretty exciting for 2025!

[1] Training Large Language Models to Reason in a Continuous Latent Space (2412.06769)
[2] https://huggingface.co/blog/ganqu/prime

Xenova

posted an update 10 days ago

Post

5528

First project of 2025: Vision Transformer Explorer

I built a web app to interactively explore the self-attention maps produced by ViTs. This explains what the model is focusing on when making predictions, and provides insights into its inner workings! 🤯

Try it out yourself! 👇
webml-community/attention-visualization

Source code: https://github.com/huggingface/transformers.js-examples/tree/main/attention-visualization

lewtun

posted an update 12 days ago

Post

2066

This paper ( HuatuoGPT-o1, Towards Medical Complex Reasoning with LLMs (2412.18925)) has a really interesting recipe for inducing o1-like behaviour in Llama models:

* Iteratively sample CoTs from the model, using a mix of different search strategies. This gives you something like Stream of Search via prompting.
* Verify correctness of each CoT using GPT-4o (needed because exact match doesn't work well in medicine where there are lots of aliases)
* Use GPT-4o to reformat the concatenated CoTs into a single stream that includes smooth transitions like "hmm, wait" etc that one sees in o1
* Use the resulting data for SFT & RL
* Use sparse rewards from GPT-4o to guide RL training. They find RL gives an average ~3 point boost across medical benchmarks and SFT on this data already gives a strong improvement.

Applying this strategy to other domains could be quite promising, provided the training data can be formulated with verifiable problems!

1 reply

Xenova

posted an update 24 days ago

Post

3389

Introducing Moonshine Web: real-time speech recognition running 100% locally in your browser!
🚀 Faster and more accurate than Whisper
🔒 Privacy-focused (no data leaves your device)
⚡️ WebGPU accelerated (w/ WASM fallback)
🔥 Powered by ONNX Runtime Web and Transformers.js

Demo: webml-community/moonshine-web
Source code: https://github.com/huggingface/transformers.js-examples/tree/main/moonshine-web

5 replies

lewtun

posted an update 26 days ago

Post

6706

We outperform Llama 70B with Llama 3B on hard math by scaling test-time compute 🔥

How? By combining step-wise reward models with tree search algorithms :)

We show that smol models can match or exceed the performance of their much larger siblings when given enough "time to think"

We're open sourcing the full recipe and sharing a detailed blog post.

In our blog post we cover:

📈 Compute-optimal scaling: How we implemented DeepMind's recipe to boost the mathematical capabilities of open models at test-time.

🎄 Diverse Verifier Tree Search (DVTS): An unpublished extension we developed to the verifier-guided tree search technique. This simple yet effective method improves diversity and delivers better performance, particularly at large test-time compute budgets.

🧭 Search and Learn: A lightweight toolkit for implementing search strategies with LLMs and built for speed with vLLM

Here's the links:

- Blog post: HuggingFaceH4/blogpost-scaling-test-time-compute

- Code: https://github.com/huggingface/search-and-learn

Enjoy!

2 replies

Narsil

posted an update about 1 month ago

Post

1087

Performance leap: TGI v3 is out. Processes 3x more tokens, 13x faster than vLLM on long prompts. Zero config !

3x more tokens.

By reducing our memory footprint, we’re able to ingest many more tokens and more dynamically than before. A single L4 (24GB) can handle 30k tokens on llama 3.1-8B, while vLLM gets barely 10k. A lot of work went into reducing the footprint of the runtime and its effect are best seen on smaller constrained environments.
13x faster

On long prompts (200k+ tokens) conversation replies take 27.5s in vLLM, while it takes only 2s in TGI. How so ? We keep the initial conversation around, so when a new reply comes in, we can answer almost instantly. The overhead of the lookup is ~5us. Thanks @Dani ël de Kok for the beast data structure.
Zero config

That’s it. Remove all the flags your are using and you’re likely to get the best performance. By evaluating the hardware and model, TGI carefully selects automatic values to give best performance. In production, we don’t have any flags anymore in our deployments. We kept all existing flags around, they may come in handy in niche scenarios.

Read more: https://huggingface.co/docs/text-generation-inference/conceptual/chunking

Xenova

posted an update about 1 month ago

Post

3024

Introducing TTS WebGPU: The first ever text-to-speech web app built with WebGPU acceleration! 🔥 High-quality and natural speech generation that runs 100% locally in your browser, powered by OuteTTS and Transformers.js. 🤗 Try it out yourself!

Demo: webml-community/text-to-speech-webgpu
Source code: https://github.com/huggingface/transformers.js-examples/tree/main/text-to-speech-webgpu
Model: onnx-community/OuteTTS-0.2-500M (ONNX), OuteAI/OuteTTS-0.2-500M (PyTorch)