Today, we're dropping instruct-tuned Zamba2 (2.7B and 1.2B) models! Using aggregate MT-Bench scores: - Our 2.7B outperforms Mistral-7B-Instruct-v0.1 - Our 1.2B outperforms Gemma2-2B-Instruct Both models have the same inference speedups as base Zamba2.

Experience the superior performance of Zyphra's Zamba2-7B model, a 7-billion parameter hybrid marvel, trained on the pioneering Zyda-2 dataset and leading the pack in model efficiency, suitable for both on-device and cloud solutions. ➡️ build.nvidia.com/zyphra/zamba2-… Get Zamba2-7B

Today, in collaboration with @NvidiaAI, we bring you Zamba2-7B – a hybrid-SSM model that outperforms Mistral, Gemma, Llama3 & other leading models in both quality and speed. Zamba2-7B is the leading model for ≤8B weight class. 👇See more in the thread below👇

Yammer is cool, but who is working on an internal corporate version of Polymarket? I need to be able to bet on the outcome of my coworker's decisions.

I will face the tsunami. I will permit it to pass over me and through me. And when it has gone past, I will turn the inner eye to see its path. Where the tsunami has gone there will be nothing. Only I will remain.

We’ve been hard at work with AMD to optimize training for AMD GPUs. Today, we’re sharing a critical milestone towards this goal: FlashAttention-2 (FA2) and Mamba-2 backward kernels on AMD MI300X that surpass NVIDIA H100. We Zyphra are the first to achieve this.

I wish it was cheaper to have an eval where humans spend 10 min talking with a voice model and it gets a point every time it makes the user laugh

audio is so cursed. it really doesn't need to be this way people just suck at writing software

The way R1 says "wait but" is so endearing

Today, we're excited to announce a beta release of Zonos, a highly expressive TTS model with high fidelity voice cloning. We release both transformer and SSM-hybrid models under an Apache 2.0 license. Zonos performs well vs leading TTS providers in quality and expressiveness.

Zyphra speaks! - Two 1.6B TTS Models (Transformer and SSM Hybrid) - Voice cloning - Optimized and cheap API - Apache 2.0, open-weights I could wax poetic about why I think Zonos is great, but listen for yourself.

This was so much fun to work on! Get hyped for v2 coming soon :)

A truly blessed number of H100's to have in a pod 🙏 The holiest time of year calls for holy H100 pods. Please keep this in mind when sourcing your compute.

Grad norms but if they slayed From llm_enjoyer and I