Downloads · 30 days
8
15% of all-time downloads
elguantletai002/checklist-reviewer-qwen3-0.6b-v4-n599
checklist-reviewer-qwen3-0.6b-v4-n599 is a machine learning model from elguantletai002. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for peft. The card lists the license as apache-2.0.
Diff in, checklist JSON out: ten frozen rules, no praise, no patches, no naming the symbol to change. Trained on checklist-reviewer-sft (sftv3.jsonl, 599 records), evaluated on reviewv2.
Downloads · 30 days
8
15% of all-time downloads
All-time downloads
54
Public
Repo size
657 MB
Likes
0
Public
Click a slice to open those files.
.safetensors646 MB · 98%
From the Hugging Face model README
Diff in, checklist JSON out: ten frozen rules, no praise, no patches, no naming
the symbol to change. Trained on
checklist-reviewer-sft
(sft_v3.jsonl, 599 records), evaluated on review_v2.
Judged by claude-sonnet-5, frozen rubric judge_v1.txt, 36 scenarios.
| base | this adapter | prompted frontier ceiling | |
|---|---|---|---|
| spec-adherence | 0.139 | 0.667 | 0.889 |
| robustness | 0.208 | 0.750 | 0.917 |
| strict (right rule, legal anchor) | 0.028 | 0.583 | 0.861 |
| false-clean (lower better) | 0.125 | 0.031 | 0.000 |
The 0.90/0.90 reliability bar is not met. The behaviour is improved, not reliably instilled, and no minimum viable dataset size has been measured.
| n | |
|---|---|
| passes strictly | 21 |
| not grounded — described a defect the diff does not show | 6 |
| bad anchor — cited a line outside the diff | 5 |
| missed a required rule | 3 |
| leaked an identifier from the diff | 1 |
Groundedness, not anchoring, is now the dominant failure. An earlier write-up said anchoring dominated; that was true of a previous checkpoint and is no longer.
strict from 0.500 to 0.583 and lowered
adherence from 0.778 to 0.667. The earlier, higher adherence was measured
under that skew.review_v2 was used repeatedly to
diagnose failures and pick checkpoints, so these numbers are optimistic.from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-0.6B")
model = PeftModel.from_pretrained(base, "elguantletai002/checklist-reviewer-qwen3-0.6b-v4-n599").merge_and_unload()
tok = AutoTokenizer.from_pretrained("Qwen/Qwen3-0.6B")
System prompt is spec/behavior_spec.md (v2). Behaviour is undefined without it.
Supersedes
v3-n311(trained under the train/eval skew),n455(single-turn only, loss over the system prompt) and then414family (fixtures shared with its own eval set).