ALFWorld 路 Qwen3.5-4B
Public trained checkpoints for ALFWorld experiments with Qwen3.5-4B. This repository contains multiple model variants; select the required subdirectory when downloading or loading.
| Directory | Contents | Stored precision |
|---|---|---|
grpo/ |
GRPO experiment model | BF16 |
single-opd/ |
Single-OPD experiment model | BF16 |
retireopd/ |
RetireOPD experiment model | BF16 |
teacher95-historical-bf16/ |
Teacher, step 95 | BF16 |
student100-fp32/ |
Earlier Student, step 100 | FP32 |
Student FP32 weights can be loaded in BF16 for experiments. Consolidation does not convert or modify model weights. Original base-model weights are excluded.
The earlier dataset-type checkpoint archive was copied here without changing its file bytes. provenance/legacy-checkpoint-archive/ preserves its README, license, attributes, and checkpoint manifest; _manifests/repository_consolidation.json records the source revision and per-file checksums.
Private experiment code, data, and records: https://huggingface.co/datasets/dingwenjun/ALFWorld-Qwen3.5-4B-records