Dataset Viewer

The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

WebShop OLMo-3-7B-Instruct GRPO: held-out evaluation at response 2048

Held-out WebShop evaluation rows for wckwan/Webshop-Olmo3-7B-Outcome-2048 (merged_hf_actor_gs20 through gs400, every 20 steps).

These rows replace the earlier GRPO rows for this cell (fingerprint 9c807b420392). Those were evaluated at max_response_length=512, a quarter of the trained cap of 2048, so the policy's generations were truncated. Do not mix the two series.

Layout

global_step_N/result.json         # metrics, diagnostics, goal set, config, fingerprint, runtime
global_step_N/trajectories.jsonl  # per-episode trajectories
global_step_N/eval_seed0.log      # eval log
MANIFEST.sha256                   # sha256sum -c compatible

Protocol

standalone_eval preset webshop_olmo3 (Instruct), identical to the GiGPO, SEED and pivot rows of this cell:

field value
env_name Webshop (small catalog, synthetic goals)
env_seed 0
val_batch_size 500 goals, 4 shards of 125 (default --val-shard-size)
max_steps 15
max_prompt_length / max_response_length 4096 / 2048
history_length 2
temperature / top_p / top_k 0.4 / 1.0 / -1
history_action_char_cap None (uncapped)

Action parser: verl-agent commit 8abb4e2d (includes 451b4969: the action is read after the final </think>). Runtime: vLLM 0.11.0, data parallel 2 (TP=1), engine seed 0.

Goal set

sha256 of json.dumps(...)
tasks.goal_indices 3c6e0c423c7234e8a59d9a8f1193f9e8be9d3c8d3874647c1ad7e54635dba05e
tasks.shards 0bc99e50663610731defb44055ace1d15f490037d6a11eea4b70d3752e042b27

Every row is checked by scripts/check_result_webshop.py, which pins the goal-set hashes, the 4x125 sharding and the 4096/2048 caps.

Config fingerprint

fingerprint
these rows (native) 6448dfad6bbdf8435a3bbd73b9822b19de99b45419d8921622329869f1065c38
legacy-equivalent (GiGPO baseline) 060ad196f1509810037101e0b3d3c3bed1cac46ee0ac26a4e9190634367a8070

The only difference between the two is the history_action_char_cap key, added in commit a668c6b6 and set to None here, which leaves behaviour unchanged. Removing it reproduces the GiGPO baseline hash exactly. Aggregate against other arms with --allow-mixed.

Downloads last month
137