Real-world work rebuilt as deterministic RL simulations from seed data. Part 1: PortSimEnv v1, berth planning at the Port of Barcelona.
AI & ML interests
Open RL Environments at Scale
Recent Activity
Try the three environments, follow the tutorial, and explore the datasets, models and historical RL/SFT results.
-
The ultimate guide to multi-harness RL
🔀206Train open models with RL inside real agent harnesses
-
Multi-harness RL slides
🎛9Watch a talk on training coding agents with RL environments
-
SmolDataEnvs Multi-harness | SETA Whitebox
🧪Answer dataset questions and view scoring results
-
SmolDataEnvs Multi-harness | Native OpenCode
🧪Run OpenCode tasks and view grading results
All 7,780 of Xiaomi's MiMo-V2.6 RL environments as Harbor tasks, set up and graded like Xiaomi's harness.
Repo2RLEnv coding and terminal RL environments in Harbor format. Datasets include per-task quality labels, provenance and generation economics.
-
Geoguesser Environment
🌍1Interact with a GeoGuessr‑style environment through custom actions
-
FineEnvs/geoguesser-tasks
RL Environment • Updated • 3.65k • 388 -
FineEnvs/geoguesser-qwen3.5-4b-grpo
Image-Text-to-Text • Updated • 61 -
FineEnvs/geoguesser-qwen3.5-4b-grpo-v3
Image-Text-to-Text • Updated • 74
Deterministic data-analysis agent tasks from the jupyter-agent dataset — verified answers, no LLM judge. Harbor env suites, plain dataset & SFT.
OpenEnv environments for multilingual OCR (22 languages) and speech recognition (102 languages), with GRPO recipes and Kannada runs.
-
Nayana Multilingual OCR
📖OpenEnv document OCR, layout and VQA over 22 languages
-
FLEURS Multilingual ASR
🎙1OpenEnv speech recognition over 102 FLEURS languages
-
FineEnvs/gemma-4-E4B-it-kannada-ocr-grpo
Image-Text-to-Text • Updated • 29 -
FineEnvs/gemma-4-E4B-it-kannada-asr-grpo
Automatic Speech Recognition • Updated • 28
One Wordle game packaged for OpenEnv, ORS and NeMo Gym, from FineEnvs' 00-environments-101: the same env, three RL frameworks side by side.
5.5K+ RL tasks for hill-climbing small models in code and data science. Deterministic grading, no LLM judge.
LaTeX OCR environment, model, dataset, and five-run training comparison across four models, including unstable and stabilized Gemma.
An RL environment where the agent paints by writing p5.brush sketches, rewarded by an aesthetic preference model looking at the render.
-
FineEnvs/watercolour-grpo-hps-only
Reinforcement Learning • Updated • 24 • 1 -
FineEnvs/watercolour-reference-pool
RL Environment • Updated • 178 • 1.92k • 2 -
FineEnvs/watercolour-rollouts-hps-only
RL Environment • Updated • 470 • 822 -
FineEnvs/watercolour-grpo-judge-led
Reinforcement Learning • Updated • 26 • 1
A curated collection of articles, guides, tutorials, slides, and resources for learning how to build, train, and evaluate RL environments for Agents
-
The ultimate guide to RL environments: building and scaling them in the LLM era
📝258Building and scaling RL environments for LLM training
-
RL Environments 101 — Slides
🎞43Explore RL Environments 101 slide presentation
-
How to turn a game into an RL environment
🌍71From an idea to a trained 4B, with the dead ends left in
-
The ultimate guide to multi-harness RL
🔀206Train open models with RL inside real agent harnesses
Real-world work rebuilt as deterministic RL simulations from seed data. Part 1: PortSimEnv v1, berth planning at the Port of Barcelona.
OpenEnv environments for multilingual OCR (22 languages) and speech recognition (102 languages), with GRPO recipes and Kannada runs.
-
Nayana Multilingual OCR
📖OpenEnv document OCR, layout and VQA over 22 languages
-
FLEURS Multilingual ASR
🎙1OpenEnv speech recognition over 102 FLEURS languages
-
FineEnvs/gemma-4-E4B-it-kannada-ocr-grpo
Image-Text-to-Text • Updated • 29 -
FineEnvs/gemma-4-E4B-it-kannada-asr-grpo
Automatic Speech Recognition • Updated • 28
Try the three environments, follow the tutorial, and explore the datasets, models and historical RL/SFT results.
-
The ultimate guide to multi-harness RL
🔀206Train open models with RL inside real agent harnesses
-
Multi-harness RL slides
🎛9Watch a talk on training coding agents with RL environments
-
SmolDataEnvs Multi-harness | SETA Whitebox
🧪Answer dataset questions and view scoring results
-
SmolDataEnvs Multi-harness | Native OpenCode
🧪Run OpenCode tasks and view grading results
One Wordle game packaged for OpenEnv, ORS and NeMo Gym, from FineEnvs' 00-environments-101: the same env, three RL frameworks side by side.
All 7,780 of Xiaomi's MiMo-V2.6 RL environments as Harbor tasks, set up and graded like Xiaomi's harness.
5.5K+ RL tasks for hill-climbing small models in code and data science. Deterministic grading, no LLM judge.
Repo2RLEnv coding and terminal RL environments in Harbor format. Datasets include per-task quality labels, provenance and generation economics.
LaTeX OCR environment, model, dataset, and five-run training comparison across four models, including unstable and stabilized Gemma.
-
Geoguesser Environment
🌍1Interact with a GeoGuessr‑style environment through custom actions
-
FineEnvs/geoguesser-tasks
RL Environment • Updated • 3.65k • 388 -
FineEnvs/geoguesser-qwen3.5-4b-grpo
Image-Text-to-Text • Updated • 61 -
FineEnvs/geoguesser-qwen3.5-4b-grpo-v3
Image-Text-to-Text • Updated • 74
An RL environment where the agent paints by writing p5.brush sketches, rewarded by an aesthetic preference model looking at the render.
-
FineEnvs/watercolour-grpo-hps-only
Reinforcement Learning • Updated • 24 • 1 -
FineEnvs/watercolour-reference-pool
RL Environment • Updated • 178 • 1.92k • 2 -
FineEnvs/watercolour-rollouts-hps-only
RL Environment • Updated • 470 • 822 -
FineEnvs/watercolour-grpo-judge-led
Reinforcement Learning • Updated • 26 • 1
Deterministic data-analysis agent tasks from the jupyter-agent dataset — verified answers, no LLM judge. Harbor env suites, plain dataset & SFT.
A curated collection of articles, guides, tutorials, slides, and resources for learning how to build, train, and evaluate RL environments for Agents
-
The ultimate guide to RL environments: building and scaling them in the LLM era
📝258Building and scaling RL environments for LLM training
-
RL Environments 101 — Slides
🎞43Explore RL Environments 101 slide presentation
-
How to turn a game into an RL environment
🌍71From an idea to a trained 4B, with the dead ends left in
-
The ultimate guide to multi-harness RL
🔀206Train open models with RL inside real agent harnesses