Conversation
- merge.py: reads adapter_config.json for base model ID, loads with device_map='cpu' (avoids ROCm segfault during merge_and_unload), saves standalone WhisperForConditionalGeneration in safetensors format alongside processor artifacts so output dir is fully self-contained - test_finetune_merge.py: 15 tests covering read_base_model_id, dry-run, missing dir, full mock pipeline, and _print_summary - pyproject.toml: register listenr-merge and listenr-asr entry points
- Add 'merge' service: CPU-only (no GPU passthrough), HIP_VISIBLE_DEVICES=-1
to prevent ROCm HSA from initialising and causing segfault
- Fix all volume entries: remove nested ${VAR:-${HOME}/path} defaults that
podman-compose cannot parse; use plain ${VAR} with required .env file
- Add :z SELinux label to all bind mounts (required on Fedora/RHEL hosts)
- .env.example: add LISTENR_MERGED; uncomment all vars with /home/you/ placeholders
Loads a merged WhisperForConditionalGeneration and runs it against audio clips from manifest.jsonl, printing side-by-side comparisons of the original Whisper transcription and the fine-tuned model output. - --keyword WORD (repeatable): filter to clips where WORD appears in corrected_transcription; marks each clip HIT/MISS and prints a recall summary table at the end - --n N: number of clips to test (default 20) - --min-duration S: skip clips shorter than S seconds - --audio PATH: transcribe a single file instead of manifest clips - JSONL parser uses raw_decode to handle objects concatenated on one line (a known manifest write bug that silently drops entries with json.loads)
Documents the RuntimeError that occurs when podman-compose encounters
nested ${VAR:-${HOME}/path} syntax, explains why plain ${VAR} is used
instead, and shows the sed one-liner to bootstrap .env from .env.example.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Take the weights from fine-tuning step and combine back into ASR model of choice.
Create a quick check to actually test if the new model handles keywords/improvements for sanities sake
add more documentation, fix previous utilities, clean up generally and ensure docker setup actually uses
.envand update project readme now that we are end to end 🥳