talk-tag 0.4.0

LLM-powered TalkBank CHAT annotator for speaker-targeted morphosyntactic transcript correction

talk-tag

talk-tag is an adapter-only TalkBank CHAT morphosyntactic error annotator for .cha and .jsonl inputs.

Runtime model contract

The deployment path is fixed:

1. Base model: unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit
2. Adapter: mash-mash/talkbank-morphosyntax-annotator-final-recon_full_comp_preserve_final_seed3407

No merged-model runtime path is used. The package bundles CHAT token augmentation entries and injects them into the tokenizer before adapter loading so tokenizer and checkpoint vocabulary stay aligned.

Install

Python >=3.10 is required.

pip install "talk-tag[runtime]"

Quickstart

Set Hugging Face credentials (required for the fixed base + adapter repositories):

export HF_TOKEN=...

On PowerShell:

$env:HF_TOKEN = "..."

Run preflight checks:

talk-tag doctor

Warm model assets:

talk-tag model pull --device auto

Annotate a folder:

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

Annotate one file:

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

CLI commands

• talk-tag annotate: annotate .cha or .jsonl data.
• talk-tag doctor: run runtime, dependency, and model-access checks.
• talk-tag model pull: pre-download model assets and optionally verify load.

.jsonl inputs require --speaker-field and --text-field.

Inference defaults

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

Documentation and support

• Documentation: https://oliverhennhoefer.github.io/talk-tag/
• Changelog: CHANGELOG.md
• Security policy: SECURITY.md
• Contributing guide: CONTRIBUTING.md

Notebook example

See examples/colab_quickstart.ipynb.