wonsupshin's picture
De-anonymize model card for public release
e63b3a7 verified
|
Raw
History Blame Contribute Delete
990 Bytes
---
base_model: google/gemma-3-27b-it
language:
- ko
- en
license: cc-by-nc-4.0
tags:
- speech-to-sft
- emnlp-2026-industry
---
# expertQA_gemma3-27b_exp1_1_medical_seed2
SFT checkpoint from the EMNLP 2026 Industry Track submission *A Factorial Ablation of a Speech-to-SFT Pipeline: Differential Effects on Data Quality and Downstream Transfer*.
| Field | Value |
|---|---|
| Pipeline condition | **Exp 1-1** (Phase 0 only — transcript refinement) |
| Domain | medical |
| Seed | 2 |
| Base model | [google/gemma-3-27b-it](https://huggingface.co/google/gemma-3-27b-it) |
| Training | LoRA (rank 16, α=32, QLoRA 4-bit, lr 2e-4, 3 epochs) |
| Upstream STT | In-house STT (paper main pipeline) |
| License | CC BY-NC 4.0 (research and non-commercial use only) |
**Intended use**: research and non-commercial use only, matching the consent scope of the source audio.
**Companion repository** (code, configs, prompts, sample QA): <https://github.com/flitto/speech-to-sft-ablation-paper>