CoolFace
Modelpublic

RESEARCH-EMPRM/emprm-v2-stageB_rung5_reasoning_s0

sourceHugging Faceapache-2.0updated 20d agoView on Hugging Face
0likes8downloads
Model Card

EM-PRM v2 — stageB_rung5_reasoning_s0

LoRA adapter for Qwen/Qwen3-VL-8B-Instruct from the paper EM-PRM: Evidence-Mediated Process Rewards for Robust Multimodal Reasoning (EM-PRM v2 experiment ladder).

Component ablation: rung-5 recipe trained on reasoning negatives only, seed 0.

Training

  • —LoRA rank 64, alpha 128, dropout 0.05, target modules downproj, gateproj, kproj, oproj, qproj, upproj, v_proj; vision tower frozen; bfloat16.
  • —Seed 0, learning rate 5e-05, effective batch 2×4, one epoch.
  • —Training data, pair sets and every gate artifact are in the mirror RESEARCH-EMPRM/emprm-v2 (dataset repo; results/runs_v2/train/stageB_rung5_reasoning_s0/) and the paper bundle under backdata/.

Pre-registered gates (development half)

  • —Held-out relational FlipAcc (operation subset): 0.7262
  • —Forced-evidence acceptance at 0.5: false 0.0, true 0.9475
  • —Chart-disjoint pair gain against the v1 head: +0.1161 (95% interval [0.0744, 0.1573])

Load

python
from transformers import AutoModelForImageTextToText, AutoProcessor
from peft import PeftModel
base = AutoModelForImageTextToText.from_pretrained("Qwen/Qwen3-VL-8B-Instruct", dtype="bfloat16", device_map="cuda")
model = PeftModel.from_pretrained(base, "RESEARCH-EMPRM/emprm-v2-stageB_rung5_reasoning_s0")
processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-8B-Instruct")

adapter_config.json records the local path the adapter was trained from; pass the base model explicitly as above. Scoring prompts (bank extraction, claim extraction, claim support, ranking) are the ones in work/scripts/eval_bon.py of the mirror.

Provenance

Trained in the EM-PRM v2 repository; every number quoted in the paper is traceable to planning/V2_PLAN.md and the generated tables in the mirror.