ALFWorld-Qwen3-8B-Adaptive-Random-Each

ALFWorld-Qwen3-8B-Adaptive-Random-Each

A Qwen/Qwen3-8B policy trained as a multi-turn search agent (Search-R1 style) with Process-GRPO: a process reward model (Olmo-3-7B-Think verifier) scores each turn, with per-(group, turn-position) advantage normalization and verifier prompts that include the retrieved tool responses and the gold answer.

The model at the repository root is the final policy (training step 400). Intermediate checkpoints are provided under step_<STEP>/ subfolders.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

# Final model (repo root)
model = AutoModelForCausalLM.from_pretrained("wckwan/ALFWorld-Qwen3-8B-Adaptive-Random-Each")
tokenizer = AutoTokenizer.from_pretrained("wckwan/ALFWorld-Qwen3-8B-Adaptive-Random-Each")

# An intermediate checkpoint
model_step = AutoModelForCausalLM.from_pretrained("wckwan/ALFWorld-Qwen3-8B-Adaptive-Random-Each", subfolder="step_400")

Checkpoints

Root: final policy at step 400.

• (none)

Training summary (step 400)

• Process-reward score mean ≈ 0.93
• Searches per trajectory ≈ 2.6 (non-collapsed, diverse multi-search policy)
• Training-batch accuracy ≈ 0.49