AI for Science Is a Governed Loop, Not a Single Model
AI for Science is a governed workflow that combines domain skills, context, adapted models, and evaluation to turn scientific questions into reproducible evidence.
Hugging FaceMaryam Motamedi, Piotr Żelasko, Kunal Dhawan, Nikolay Karpov, Nune Tadevosyan, Meline Mkrtchyan, Nithin Rao Koluguri, FrancescoAnnouncement
NVIDIA released Nemotron 3.5 Transcribe, an open-weight, promptable speech model that performs transcription, speaker attribution, translation, and audio Q&A in a single checkpoint. It transcribes 40 languages, translates 31 to or from English, and attributes speakers in 13, handling recordings up to an hour long with up to eight speakers. The model leads the Hugging Face Open ASR Leaderboard for English at 5.33% average WER and tops David AI’s multilingual diarization leaderboard with an 18.26% diarization error rate. It uses a hybrid Mamba–Transformer mixture-of-experts architecture with about 4B active parameters out of 30B total, running on a single NVIDIA GPU. The accompanying Granary v2 pipeline and NeMo Speech fine-tuning tools allow developers to adapt the model to domain vocabulary, demonstrated by a medical case study that cut word error rate by 45.8% on held-out speakers.
NVIDIA released Nemotron 3.5 Transcribe, an open-weight, promptable speech model that performs transcription, speaker attribution, translation, and audio Q&A in a single checkpoint. It transcribes 40 languages, translates 31 to or from English, and attributes speakers in 13. The model uses a hybrid Mamba–Transformer mixture-of-experts architecture with about 4B active parameters out of 30B total. It runs on a single NVIDIA GPU and can be served with vLLM or run locally with NeMo-Speech.cpp. The accompanying Granary v2 pipeline and NeMo Speech fine-tuning tools allow developers to adapt the model to domain vocabulary and transcript conventions.
NVIDIA released Nemotron 3.5 Transcribe, an open-weight, promptable speech model that performs transcription, speaker attribution, translation, and audio Q&A in a single checkpoint. It transcribes 40 languages, translates 31 to or from English, and attributes speakers in 13. The model uses a hybrid Mamba–Transformer mixture-of-experts architecture with about 4B active parameters out of 30B total. It runs on a single NVIDIA GPU and can be served with vLLM or run locally with NeMo-Speech.cpp. The accompanying Granary v2 pipeline and NeMo Speech fine-tuning tools allow developers to adapt the model to domain vocabulary and transcript conventions.
Nemotron 3.5 Transcribe is a single open-weight model that handles transcription, speaker attribution, translation, and audio Q&A, with support for 40 transcription languages and 13 speaker-attribution languages.
The model leads the Hugging Face Open ASR Leaderboard for English at 5.33% average WER and tops David AI’s multilingual diarization leaderboard with an 18.26% diarization error rate.
Fine-tuning with Granary v2 and NeMo Speech tools can substantially reduce word error rate on domain-specific speech, as shown by a 45.8% relative WER reduction on held-out speakers in a medical case study.
Nemotron 3.5 Transcribe transcribes 40 languages, translates 31 of them to or from English, and attributes speakers in 13 languages. Audio question answering is trained in English only.
The model runs on a single NVIDIA GPU, from Jetson AGX Thor and DGX Spark to Blackwell data-center GPUs. On one NVIDIA RTX PRO 6000 Blackwell Server Edition GPU with NVFP4 precision, it processes 100 hours of audio in under three minutes on 30-second clips at high concurrency.
Yes, developers can guide output with natural language prompts, such as specifying transcription with speaker labels, translation, or audio Q&A. Prompts can include term lists with near-miss distractors so the model uses a spelling only when the audio supports it.
Summaries are generated from the original by a language model. Check the original before relying on it.
See the idea
AI for Science is a governed workflow that combines domain skills, context, adapted models, and evaluation to turn scientific questions into reproducible evidence.
Keep reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
One useful email. Unsubscribe anytime.