Update weights: MERGE_PROVENANCE.json

Update weights: README.md

Update weights: model-00001-of-00002.safetensors

Model card: precision fixes (inline post-processing location; leaderboard entry timing)

transcribe.py: inline n-gram loop fix (harness loader fetches a fixed sibling list)

Model card: remove score section pending leaderboard results

Model card: clarify ESB = Open ASR Leaderboard test suite

Model card: correct Performance table and architecture note to current measurements

Update v2: README.md

Update v2: ngram_loop_fix.py

Update v2: transcribe.py

Update v2: model-00001-of-00002.safetensors

Add leaderboard reproduction helpers

Add leaderboard reproduction helpers

Add leaderboard reproduction helpers

Add open-asr-leaderboard reproduction script (5.30% 8B / 5.67% 1.7B)

Embed internal leaderboard-parity eval logic in transcribe.py

fix: also suppress right-padding warning from transformers logger

fix: revert pad_left to False (model expects right-padded input + re-pads internally)

fix: use pad_left=True in collator (decoder-only batched generation needs left-padding)

feat: batched generation for ASR (remove bs=1 assertion when no audio_out tokens)

feat: batched generation for ASR (remove bs=1 assertion when no audio_out tokens)

feat: add transcribe_batch for parallel inference (single transcribe wraps it)

fix: dual-mode imports + complete ChatMLDatasetSample + transcribe output parsing

fix: dual-mode imports + complete ChatMLDatasetSample + transcribe output parsing

fix: dual-mode imports + complete ChatMLDatasetSample + transcribe output parsing

fix: dual-mode imports + complete ChatMLDatasetSample + transcribe output parsing

fix: add auto_map + make xcodec imports lazy for trust_remote_code support

fix: add auto_map + make xcodec imports lazy for trust_remote_code support

fix: add auto_map + make xcodec imports lazy for trust_remote_code support

Update model weights and eval results (avg WER 5.67% → 5.36%)

Add ESB benchmark eval results (model-index) for Open ASR Leaderboard

Add README with usage instructions and model card

Fix: use importlib for deepspeed to avoid static import check

Fix: handle higgs_audio_encoder config without global registry

Add model code for trust_remote_code support

Upload folder using huggingface_hub

initial commit