CoolFace
Modelpublic

burtenshaw/openenv-echo-world-model-distilgpt2-seed2

sourceHugging Faceapache-2.0updated 3mo agoView on Hugging Face
0likes12downloads
Model Card

OpenEnv ECHO World Model

This checkpoint was trained with examples/echo_world_model/train_echo.py to predict OpenEnv terminal environment outputs from verifier-free ECHO loss.

Training Metrics

metricvalue
best_step35
heldout_ce_after0.26466473937034607
heldout_ce_before6.246147632598877
heldout_ce_delta-5.981482893228531
heldout_ce_improvement_pct95.76275242056317
heldout_token_acc_after0.8421052631578947
heldout_token_acc_before0.10526315789473684
lr5e-05
modeldistilgpt2
seed2
steps60