talk-tag 0.2.0

Transcript annotator for speaker-scoped CHAT corpus correction

talk-tag

Adapter-only TalkBank CHAT morphosyntactic error annotator for .cha and .jsonl.

The runtime deployment path is fixed to:

1. Base model: unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit
2. Adapter: mash-mash/Llama_TalkTag_CHAT_error_annotator_adapter

No merged-model runtime path is used.

Install

Python requirement: >=3.10.

pip install "talk-tag[runtime]"

Runtime extras include torch, transformers, and peft.

Hugging Face access

You need Hub access to both repositories above. Set a token before first run:

export HF_TOKEN=...

If token or access is missing, talk-tag doctor/talk-tag model pull will report auth or gated-repo errors.

First-run workflow

1. Check environment:

talk-tag doctor

2. Pull/warm model assets:

talk-tag model pull --device auto

3. Run annotation:

talk-tag annotate \
  --input-dir ./input \
  --output-dir ./output \
  --target-speaker "*CHI" \
  --device auto

Inference defaults

• batch_size = 4
• max_new_tokens = 128
• max_seq_length = 512
• max_context_chars = 1200
• limit = 0
• greedy decoding (do_sample = false)

Supported runtime inputs

• .cha
• .jsonl (requires --speaker-field and --text-field)

Other previously supported formats (.txt, .csv, .json, .xlsx) are rejected in adapter-only deployment mode.

Colab quickstart

See examples/colab_quickstart.ipynb for a minimal setup flow.