Jade Copet

We release CWM, a 32B LLM for code reasoning, agents, and world modeling research! 📊 Tech Report ai.meta.com/research/publi… 🤖Model Weights ai.meta.com/resources/mode… 🤗 Hugging Face huggingface.co/facebook/cwm 🧑‍💻 Inference Code github.com/facebookresear… Check it out

We’re excited to share our preparedness report on Code World Model (CWM), FAIR’s latest open-weight model for code generation and reasoning. This report was developed by the SEAL team and the AI Security team, marking our first external publication since part of SEAL joined Meta just 1.5 months ago. It reflects our ongoing commitment to safety and alignment as we look ahead to future open-source and frontier models. We’re actively working to broaden our evaluation coverage and improve our processes, and we welcome collaboration and feedback from the broader research community. For more details, please read the full report: ai.meta.com/research/publi…

SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolution arxiv.org/abs/2502.18449 by @YuxiangWei9 @sidawxyz and the whole team! Get started with your favorite model here github.com/facebookresear…

LLMs for code should do much better if they can iterate on tests -- but they don't. Our new work (RLEF) addresses this with execution feedback at RL *training time* to use execution feedback at *inference time*. arxiv.org/abs/2410.02089 is just out! 1/6

Very happy to announce that my paper “Audio Conditioning for Music Generation via Discrete Bottleneck Features“ done with @honualx @adiyossLC @jadecopet and Axel Roebel has been accepted at ISMIR24. Paper: arxiv.org/abs/2407.12563 Sample: musicgenstyle.github.io Code: soon

Join us tomorrow to learn about our recent work on multimodality!

AudioSeal training code is now available inside the beautiful audiocraft repo 🚀 github.com/facebookresear… You can now train your own audio watermarking models and define your own tradeoff between fidelity, robustness and message capacity based on your needs.

Introducing Meta Llama 3: the most capable openly available LLM to date. Today we’re releasing 8B & 70B models that deliver on new capabilities such as improved reasoning and set a new state-of-the-art for models of their sizes. Today's release includes the first two Llama 3 models — in the coming months we expect to introduce new capabilities, longer context windows, additional model sizes and enhanced performance + the Llama 3 research paper for the community to learn from our work. More details ➡️ go.fb.me/i2y41n Download Llama 3 ➡️ go.fb.me/ct2xko

Today we present AudioSeal, a proactive solution for the detection of voice cloning based on localised watermarking. It relies on 2 jointly trained models: an imperceptible watermark generator and a detector with sample level precision. 1/n

We released a 70B version of CodeLlama today! Trained on 1T tokens, it is a much stronger base model for coding tasks. I look forward to seeing what the community will do with it! :)

MAGNeT by Meta text-to-music and text-to-audio huggingface.co/papers/2401.04… MAGNeT, a masked generative sequence modeling method that operates directly over several streams of audio tokens. Unlike prior work, MAGNeT is comprised of a single-stage, non-autoregressive transformer. During training, we predict spans of masked tokens obtained from a masking scheduler, while during inference we gradually construct the output sequence using several decoding steps. To further enhance the quality of the generated audio, we introduce a novel rescoring method in which, we leverage an external pre-trained model to rescore and rank predictions from MAGNeT, which will be then used for later decoding steps. Lastly, we explore a hybrid version of MAGNeT, in which we fuse between autoregressive and non-autoregressive models to generate the first few seconds in an autoregressive manner while the rest of the sequence is being decoded in parallel. We demonstrate the efficiency of MAGNeT for the task of text-to-music and text-to-audio generation and conduct an extensive empirical evaluation, considering both objective metrics and human studies. The proposed approach is comparable to the evaluated baselines, while being significantly faster (x7 faster than the autoregressive baseline). Through ablation studies and analysis, we shed light on the importance of each of the components comprising MAGNeT, together with pointing to the trade-offs between autoregressive and non-autoregressive modeling, considering latency, throughput, and generation quality.

New unsupervised lofi AI beats dropped 👀 jbilcke-hf-ai-tube.hf.space/watch?v=f8d004…

How to make machines understand the world, reason, plan, and learn as efficiently as animals humans? Meta-FAIR is hiring scientists to work on this question. We call this Advanced Machine Intelligence (AMI). Help us build the next generation of AI. Open positions are available in all our locations in NORAM (Bay Area, Seattle, Montreal, New York, Pittsburgh) and EMEA (Paris, London, Tel Aviv). FAIR believes in open research. If you are at NeurIPS, come see us at the Meta booth. Just click here: metacareers.com/jobs/872265983…

Joint work with @FelixKreuk, @itai_gat, @talremez, David Kant, @adiyossLC, @syhw and @honualx [6/6]

MusicGen is publicly available! 📄 Paper arxiv.org/abs/2306.05284 💻 Code (MIT) & pretrained models (CC-by-NC) available in AudioCraft github.com/facebookresear… 🤖 HF demo huggingface.co/spaces/faceboo… 🎶 More Samples ai.honu.io/papers/musicge… [5/6]

We present MusicGen at #NeurIPS2023 on Thu at 10.45am. Poster #603 "Simple and Controllable Music Generation”. MusicGen is a state-of-the-art text-to-music generation model with optional control over the melody. It can generate high-quality samples, both mono and stereo. 🧵[1/6]

Curious about using NLP pretrained LLMs for speech? Turns out you can just fine tune them. Poster number #543, 5pm to 7pm local time with @MichaelHassid. See you there ;) See @adiyossLC great recap for more 👇