Instructions to use OpenMOSS-Team/MOSS-Transcribe-Diarize with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use OpenMOSS-Team/MOSS-Transcribe-Diarize with Transformers:
# Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("OpenMOSS-Team/MOSS-Transcribe-Diarize", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
Add Open ASR Leaderboard results
Final guarded Open ASR results are now uploaded and verified.
- Dataset revision:
b6bdcd0beb34f8975dc659796176d88f43aff502 - Cleaned 7 sets: 74,842 / 74,842 samples
- AVG7 WER: 5.17%
- Aggregate RTFx: 294.02
- Evaluator Space:
Cb1ock/open-asr-leaderboard-moss-transcribe-diarize - Results bucket:
Cb1ock/moss-transcribe-diarize-results - Batch size: 256 on H200
The PR YAML has been refreshed to these final numbers. All seven QZ validation jobs completed with the final guarded evaluator code.
The final guarded manifests are now also mirrored to the public personal bucket Cb1ock/open-asr-leaderboard-moss-transcribe-diarize-results. Anonymous download verification passed for all 7 files (74,842 rows), with SHA256 matching the locally scored manifests. The original private bucket was left intact.
OpenMOSS-Team maintainers: this PR is ready to merge. The YAML diff contains the final guarded current-normalizer scores (AVG7 5.17, aggregate RTFx 294.02). The evaluator Space is under Cb1ock, and all 7 manifests are anonymously readable from Cb1ock/open-asr-leaderboard-moss-transcribe-diarize-results with 74,842/74,842 rows and matching SHA256 checksums.
Closing this submission for now; we are not proceeding with the leaderboard submission.
Reopening this submission; proceeding with the Open ASR Leaderboard submission.
GitHub Open ASR Leaderboard submission is now open: https://github.com/huggingface/open_asr_leaderboard/pull/192
The submission branch was synchronized with the latest upstream main before opening the PR. The evaluator Space, public result bucket, and final 7-set scores remain unchanged.