WebShop 路 Qwen3.5-4B
Public WebShop experiment checkpoints based on Qwen3.5-4B. Original model directories and stored tensor precision are preserved from the previous dataset-type archive. Student step85 remains in its original sharded layout; Teacher step100 is the merged FP32 model. FP32 weights may be loaded in BF16 for experiments.
No original base-model weights are included. This migration changes repository type and naming only; it does not merge, convert or retrain models. Select the required model subdirectory when downloading.
Original archive metadata is preserved under provenance/legacy-checkpoint-archive/. _manifests/repository_consolidation.json lists original paths, source revision and checksums.
Private experiment records: https://huggingface.co/datasets/dingwenjun/WebShop-Qwen3.5-4B-records