siddham0909/trace-rca-qwen3-1.7b

Trace RCA — Qwen/Qwen3-1.7B (GRPO + LoRA)

Trained on the Trace RCA Gym environment.

Training evidence

• —reward_plot.png — reward curves (regenerated every 5 episodes during training)
• —metrics.jsonl — per-episode rich metrics (fault_type, tier, difficulty, reward components)
• —fault_type_metrics.json — per-fault-type rolling success
• —compare.png, compare.md — before/after comparison on pinned scenarios
• —adapter/ — the LoRA adapter (load with PEFT)
• —tb_episodes/ — TensorBoard episode scalars

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-1.7B")
model = PeftModel.from_pretrained(base, "siddham0909/trace-rca-qwen3-1.7b", subfolder="adapter")