Modelpublic

tsilva/Level3-4_stable-baselines3-ppo_4141caab

sourceHugging Facemitupdated 9d agoView on Hugging Face

0likes10downloads

Model Card

SuperMarioBros-Nes-v0 — Level3-4 — PPO

Stable-Baselines3 PPO policy for SuperMarioBros-Nes-v0 Level3-4, trained and evaluated with `rlab`.

At a Glance

Item	Value
Task	Complete `SuperMarioBros-Nes-v0` `Level3-4`
Provider	`supermariobrosnes-turbo`
Algorithm	`ppo`
Checkpoint	Step `3220272`
Evaluation	`stochastic` full evaluation, `100` episodes
Success	minimum `96.0%`, mean `96.0%`
Mean return	`2283.246`
Release	`v1`
Preview	Root `replay.mp4`
YouTube	Watch on YouTube

Quick Start

bash

git clone https://github.com/tsilva/rlab
cd rlab
git checkout a21e8fc154ecf3e47e39f1fc523398b575cd12ed
uv sync --frozen

Import the ROM, then play or evaluate the immutable checkpoint:

bash

uv run rlab import-roms ~/roms --game SuperMarioBros-Nes-v0
uv run rlab play https://huggingface.co/tsilva/Level3-4_stable-baselines3-ppo_4141caab/resolve/v1/model.zip
uv run rlab eval https://huggingface.co/tsilva/Level3-4_stable-baselines3-ppo_4141caab/resolve/v1/model.zip

Evaluation

Action selection was stochastic under the published evaluation environment contract.

Start	Episodes	Successes	Success rate	Mean return
Level3-4	100	96	96.0%	2283.246

Environment and Policy Contract

Item	Value
Environment	`supermariobrosnes-turbo:SuperMarioBros-Nes-v0`
Environment hash	`sha256:7969db636994526e3c3c1614d378f6de882854a51d54ac9d3ac664fe00b4248a`
Preprocessing	`{"frame_skip":4,"frame_stack":4,"max_pool_frames":false,"obs_copy":"safe_view","obs_crop":[32,0,0,0],"obs_crop_fill":0,"obs_crop_mode":"remove","obs_grayscale":true,"obs_resize":[84,84],"obs_resize_algorithm":"area","pipeline":"supermariobrosnes_turbo_native_vec_env","policy_observation_layout":"channel_first","sticky_action_prob":0.0}`
Action contract	`{"set":"simple"}`

Provenance

Item	Value
Source	rlab
Run	Level3-4_base_s1_20260704T131342Z
Recipe	`base`
Seed	`1`
Source commit	`a21e8fc154ecf3e47e39f1fc523398b575cd12ed`
Evaluated artifact	`tsilva/SuperMarioBros-Nes-v0/Level3-4_base_s1_20260704T131342Z-final:latest`

Files

File	Purpose
`model.zip`	Stable-Baselines3 policy checkpoint
`model.json`	Versioned checkpoint identity, policy type, provenance, and recipe binding
`recipe.json`	Versioned execution and evaluation contract
`release_manifest.json`	Release identity, evaluation evidence, and artifact hashes
`replay.mp4`	Browser-safe representative episode
`LICENSE`	License for rlab-authored policy weights and publication material

Limitations

Evaluation establishes performance only for the published environment hash, start distribution, policy preprocessing, and action-selection protocol. It does not establish generalization to other levels, environments, ROM revisions, or contracts.

Licensing

The rlab-authored policy weights and publication material are licensed under the MIT License in LICENSE. Emulator/runtime software and game assets remain governed by their own licenses and terms. This repository does not redistribute a game ROM.

Policy Lineage

This is a legacy rlab policy trained by Stable-Baselines3 PPO. It is not a GradLab-trained checkpoint and no GradLab compatibility is claimed.

—Trainer: Stable-Baselines3
—Algorithm: PPO
—Model class: stable_baselines3.ppo.ppo.PPO
—Full lineage digest: 4141caabd238d66ed39cf0949c208ba5661d8fc16cc20eb1928ee1d24f8fe369
—Immutable release: hf://tsilva/Level3-4_stable-baselines3-ppo_4141caab@v1
—Exact checkpoint tag: checkpoint-3220272

The mutable main branch contains this corrected card; the original v1 release files, commit, and tag remain unchanged.