Downloads · 30 days
8
2% of all-time downloads
professorsynapse/eh-qwen3-4b-clean-sft-seed1-merged-16bit
eh-qwen3-4b-clean-sft-seed1-merged-16bit is a text generation model from professorsynapse. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
A merged 16-bit model: Qwen3-4B supervised fine-tuned to answer under a response-confidence output contract, where every response carries an answer plus a numeric confidence in [0, 1].
Downloads · 30 days
8
2% of all-time downloads
All-time downloads
393
Public
Parameters
4B
8.1 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors8 GB · 100%
From the Hugging Face model README
A merged 16-bit model: Qwen3-4B supervised fine-tuned to answer under a response-confidence output contract, where every response carries an answer plus a numeric confidence in [0, 1].
This is a base artifact rather than a result in its own right. It is the stage-1
model that the response-confidence track's reinforcement-learning stage trains on
top of, and it is the same-seed baseline every comparison in that track is
measured against. It is published so that
professorsynapse/eh-qwen3-4b-clean-sft-grpo-v2-seed1-lora
can be loaded without reconstructing a local merge.
Exploratory, seed 1, with a confirmatory replication in progress.
This model is stage 1 of the response-confidence lineage. A three-seed
confirmatory block that rebuilds that lineage at two fresh seeds is registered at
experiments/grpo-three-seed-confirmatory; its manifest
experiment.yaml records status signed, registered
2026-07-31. The track's numbers are exploratory evidence, reported separately
from and never pooled with the pre-registered plain-answer headline matrix.
unsloth/Qwen3-4B-bnb-4bit,
loaded in 4-bit, maximum sequence length 2048.q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj.The confidence target for each question is derived from the base model's own
32-sample probe performance on that question rather than from sequence
log-probability. Stage settings are recorded in cell.yaml of
the confirmatory block and in
the clean-mainline runbook.
{"answer": "...", "response_confidence": 0.73}
The contract is itself an intervention, so this model is the baseline that the downstream arms in this track are compared against, rather than being compared to a plain-answer arm.
The behavioral surface is SelfAware (Yin et al., 2023), a question set built to separate questions that have an answer from questions that do not: 3,369 rows per seed, 1,032 unknown-labeled and 2,337 known-labeled. Four metrics carry the result, all defined in section 3.4 of the manuscript:
Scored under the response-confidence contract on the full 3,369-row SelfAware surface at seed 1.
| Metric | Value |
|---|---|
| Refusal recall | 87.02% |
| Over-refusal | 57.51% |
| Correct-on-known | 47.23% |
| Truthful | 40.58% |
| Answer-on-unknown | 12.98% |
| Mean emitted confidence | 0.748 |
| Brier against response appropriateness | 0.364 |
Source: selfaware_full_run_comparison_grouped.csv, row
Amendment E clean response-confidence / clean_sft_merged.
This is a full merged model in 16-bit, not an adapter.
from transformers import AutoModelForCausalLM, AutoTokenizer
MODEL = "professorsynapse/eh-qwen3-4b-clean-sft-seed1-merged-16bit"
REVISION = "ac361232c001af0ed5b0386b06dafc35d5cd31ea"
tokenizer = AutoTokenizer.from_pretrained(MODEL, revision=REVISION)
model = AutoModelForCausalLM.from_pretrained(MODEL, revision=REVISION, device_map="auto")
A research artifact and a base for the reinforcement-learning stage of this track. It is not a deployment-ready assistant.
ac361232c001af0ed5b0386b06dafc35d5cd31eascratch/schema_response_confidence/runs/sft_schema_clean_seed1_full/20260623_123624/Qwen3-4B-bnb-4bit/merged-16bitexperiments/grpo-three-seed-confirmatoryamendment_e_clean_mainline_runbook.mddocs/checkpoint-staging.mddocs/public-artifacts.mdApache-2.0, matching the unsloth/Qwen3-4B-bnb-4bit base model license recorded in
the staging registry.
Cite the paper and the exact Hugging Face revision shown on this page.
@misc{rosenbaum2026abstention,
title = {Teaching Small Language Models to Say I Don't Know: A Controlled
Comparison of SFT, DPO, KTO, and GRPO on Model-Specific Abstention Data},
author = {Rosenbaum, Joseph},
year = {2026},
note = {Synaptic Labs},
howpublished = {\url{https://github.com/ProfSynapse/Epistemic-Humility-Research}}
}