Likich/open-coding-zephyr7b-single_code-qlora

zephyr7b QLoRA Open-Coding Adapter

This PEFT adapter fine-tunes HuggingFaceH4/zephyr-7b-beta to produce exactly one concise open code for an input utterance or qualitative text segment.

Output schema

Task mode: single_code.

{"code": "short analytical label"}

Held-out verification

• —Rows: 100
• —Valid JSON rate: 1.000
• —Non-empty rate: 1.000
• —Exact set match: 0.170
• —Mean set F1: 0.170
• —Average generated codes: 1.000
• —Verification passed: True

Exact match is reported as a format and regression diagnostic, not as a complete measure of open-code quality. Valid abstractive labels may differ in wording.

Loading

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model = AutoModelForCausalLM.from_pretrained("HuggingFaceH4/zephyr-7b-beta")
tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-beta")
model = PeftModel.from_pretrained(base_model, "Likich/open-coding-zephyr7b-single_code-qlora")

The repository contains adapter weights, tokenizer metadata, training metadata, held-out verification metrics, and sample predictions. It does not contain the full base-model weights.