Qihui (Isaac) Yang

[Andrew Ho @andrewho03]

https://t.co/v7RlDg1cTV

[Qihui (Isaac) Yang @rocksoul183]

We see our approach extendable to many generative modeling approaches!

[Qihui (Isaac) Yang @rocksoul183]

Here comes our new Instrument Generation model! 🎹

Presenting FlowSynth: A new paradigm for flow matching that enables test-time search to achieve SOTA results. Work done @SmuleLabs @smule.

Paper Link: https://t.co/hXmVtJxoqd
We are prepping open-sourcing!
(Threads 👇)

[Qihui (Isaac) Yang @rocksoul183]

We reach SOTA instrument generation performance with it!

[Qihui (Isaac) Yang @rocksoul183]

DFM works well without test-time search, and even better with test-time search.

[Qihui (Isaac) Yang @rocksoul183]

Unlike standard flow matching formulated as a deterministic ODE, DFM is a spin on Rectified Flow Matching that learns a distribution at each step.

[AK @_akhaliq]

Llasa

Scaling Train-Time and Inference-Time Compute for Llama-based Speech Synthesis

[Qiuqiang Kong @QiuqiangK]

A codebase for learning TTS/music_generation with LLM in 1 hour: https://t.co/hsLGcXWc9s

[Paul Couvert @itsPaulAi]

Wait, so we now have a 100% open source model that's better than GPT-4o?!

DeepSeek v3 is even superior to Claude Sonnet 3.5 for code according to multiple benchmarks.

Already available for free to everyone 🧵

[Scott H. Hawley @drscotthawley]

New tutorial! I spent 3 weeks realizing flow-matching/rectified flows can be viewed in a simple way that end-runs the usual pages of math:
"Basic physics provides a 'straight, fast' way to get up to speed with flow-based generative models"
Colab included! https://t.co/NHscQI7qCX

[Big Tech Alert @BigTechAlert]

🆕 @karpathy has started following @drscotthawley