Excited to release Higgs-Llama-3-70B, the first model in our Higgs family of LLM for role playing - Post-trained from Llama-3-base - Excluded benchmark data (including training examples) from our fine-tuning data - Blog: boson.ai/higgs-opensour… - 🤗: huggingface.co/bosonai/Higgs-…

Congrats to the Qwen2 release👏 We tuned Qwen1 models for our customers last year, they were very happy!😄 Higgs-Llama-3-70B and Qwen2-70B-Instruct reported very comparable results😆😆😆 Results by joining benchmark tables with Llama-3-70B-Instruct as the key

Excited to share Higgs-V2, improved both general and roleplaying abilities. The performance boost comes from the in-house built reward model. More at boson.ai/higgs-v2/

BosonAI

Boson AI's new audio generation model is out: emotionally expressive voice, long form audio, multi-speaker cloning, directable audio (e.g. music, laughter), trained on 10M+ hours of audio, and best of all, open source. boson.ai/blog/higgs-aud… - BosonAI Xingjian Shi Mu Li

1/3 We are excited to announce our opensource release of #BosonAI #HiggsAudio v2 (boson.ai/blog/higgs-aud…), a SToA multimodal LLM for audio generation. #HiggsAudio v2 is the first OSS large audio model excels at multi-speakers cloning, emotionally competent voice generation.

2/3 #HiggsAudiov2 is trained on a massive corpus of over 10M hours of high-quality audio data, annotated using BosonAI’s #ASR, and #LLM models. on #Github: github.com/boson-ai/higgs… + on #HuggingFace huggingface.co/bosonai/higgs-…).

3/3 HiggsAudiov2 adopts Dual-FFN architecture that is capable of handling text and audio tokens jointly. Higgs Audio v2 has achieved state-of-art performance, beats “gpt‑4o‑mini‑tts” with 75.7% win rate on “Emotions” and 55.7 % on “Questions” in EmergentTTS‑Eval.

Multi-speaker conversations can be tricky for AI models. #HiggsAudio makes it easy. #HiggsAudio v2 supports end2end voice clone and generation, see live #HiggsAudio translating and cloning simultaneously youtu.be/br9zBkpcaco. #HiggsAudio inference can run on a Jetson Orin Nano

NEW: Higgs Audio V2 from BosonAI open, unified TTS model w/ voice cloning, beats GPT 4o mini tts and ElevenLabs v2 🔥 > Trained on 10M hours (speech, music, events) > Built on top of Llama 3.2 3B > Works real-time and on edge > Beats GPT-4o-mini-tts, ElevenLabs v2 in prosody

New TTS bomb dropped on Hugging Face. It has multispeaker support and the output quality looks amazing 🤯

#HiggsAudio V2 by BosonAI is now the #1 trending audio model on Hugging Face - try it out on huggingface.co/spaces/smola/h…

toast of the town

youtu.be/VGFP_xjWLoE Higgs-audio whipped up something relaxing —a fully AI-generated track with lyrics, beat, and chorus all in one, zero-shot.