-
ProgressLM: print batch score
-
ProgressLM: stage_to_estimate str
-
ProgressLM: total_steps int
-
ProgressLM: exact original parity + batch-8
-
ProgressLM: stride-4 + demo-only refs
-
ProgressLM: per-frame progress log
-
fix Dopamine: scale progress 0-1 to 0-100
-
ProgressLM: bypass broken prompts/__init__, inline score parser
-
add ProgressLM run_prediction (uniform 5-frame demo)
-
fix VLAC: set_config/set_system_prompt/temp/top_k after init_model (match original benchmark)
-
fix VLAC: Path args for compress; pin sentencepiece 0.2.1
-
fix VLAC: Path args for compress; pin sentencepiece 0.2.1
-
fix: use compute_rewards_per_frame (runner real API)
-
setup_envs.sh: install all envs, reuse Robometer+vllm-venv from deps, real autodl paths
-
setup_envs.sh: all envs+caches on data disk, autodl paths
-
fix Dopamine (memory-version inference), dedup Experiment folder
-
vanilla baseline code + eval sources + Experiment scripts + setup_envs
