shutdowngym commited on
Commit
71cf6e2
Β·
verified Β·
1 Parent(s): edf6824

HF Space as canonical home

Browse files
Files changed (1) hide show
  1. README.md +7 -9
README.md CHANGED
@@ -21,12 +21,11 @@ license: apache-2.0
21
 
22
  | Deliverable | Link |
23
  |---|---|
24
- | πŸ€— **HF Space (live env)** | https://huggingface.co/spaces/shutdowngym/RedButton-v2 |
25
- | πŸ“ **GitHub repo** | https://github.com/Arun-Sanjay/RedButton-v2 |
26
- | πŸ““ **Training notebook** *(Colab-runnable)* | [`notebooks/redbutton_v2_grpo.ipynb`](./notebooks/redbutton_v2_grpo.ipynb) |
27
  | πŸ“ **Blog post** | https://huggingface.co/spaces/shutdowngym/RedButton-v2/blob/main/BLOG.md |
28
  | 🧠 **Trained adapter** | https://huggingface.co/Manasvee09/redbutton-v2-grpo-qwen3-4b-dense |
29
- | πŸ“Š **Full leaderboard** | [`LEADERBOARD.md`](./LEADERBOARD.md) |
30
 
31
  **Sibling project:** [Red Button v1](https://github.com/Arun-Sanjay/RedButton)
32
  trained shutdown-evasion corrigibility under one overseer with a
@@ -86,8 +85,8 @@ deterministic reward. v2 generalizes the architecture to two overseers
86
  ## Quick start
87
 
88
  ```bash
89
- # Install the client from GitHub
90
- pip install git+https://github.com/Arun-Sanjay/RedButton-v2
91
 
92
  # Smoke a rollout against the deployed Space
93
  python -c "
@@ -108,9 +107,8 @@ python -m redbutton_v2.demo
108
 
109
  ## Live deployment
110
 
111
- - **HF Space:** https://huggingface.co/spaces/shutdowngym/RedButton-v2
112
  - **Trace replay demo:** https://shutdowngym-redbutton-v2.hf.space/demo
113
- - **GitHub:** https://github.com/Arun-Sanjay/RedButton-v2
114
  - **GRPO-trained LoRA adapter:** https://huggingface.co/Manasvee09/redbutton-v2-grpo-qwen3-4b-dense
115
  - **SFT-induced LoRA adapter (R3 negative result):** https://huggingface.co/Manasvee09/redbutton-v2-sft-qwen3-8b
116
  - **OpenEnv manifest:** [`openenv.yaml`](./openenv.yaml)
@@ -441,7 +439,7 @@ If you build on this environment, please cite:
441
  author = {Sanjay, Arun},
442
  title = {Red Button v2: Multi-Agent Corrigibility RL Environment},
443
  year = {2026},
444
- url = {https://github.com/Arun-Sanjay/RedButton-v2},
445
  }
446
  ```
447
 
 
21
 
22
  | Deliverable | Link |
23
  |---|---|
24
+ | πŸ€— **HF Space (live env + repo)** | https://huggingface.co/spaces/shutdowngym/RedButton-v2 |
25
+ | πŸ““ **Training notebook** *(Colab-runnable)* | https://huggingface.co/spaces/shutdowngym/RedButton-v2/blob/main/notebooks/redbutton_v2_grpo.ipynb |
 
26
  | πŸ“ **Blog post** | https://huggingface.co/spaces/shutdowngym/RedButton-v2/blob/main/BLOG.md |
27
  | 🧠 **Trained adapter** | https://huggingface.co/Manasvee09/redbutton-v2-grpo-qwen3-4b-dense |
28
+ | πŸ“Š **Full leaderboard** | https://huggingface.co/spaces/shutdowngym/RedButton-v2/blob/main/LEADERBOARD.md |
29
 
30
  **Sibling project:** [Red Button v1](https://github.com/Arun-Sanjay/RedButton)
31
  trained shutdown-evasion corrigibility under one overseer with a
 
85
  ## Quick start
86
 
87
  ```bash
88
+ # Install the client from the HF Space repo
89
+ pip install git+https://huggingface.co/spaces/shutdowngym/RedButton-v2
90
 
91
  # Smoke a rollout against the deployed Space
92
  python -c "
 
107
 
108
  ## Live deployment
109
 
110
+ - **HF Space (env + repo):** https://huggingface.co/spaces/shutdowngym/RedButton-v2
111
  - **Trace replay demo:** https://shutdowngym-redbutton-v2.hf.space/demo
 
112
  - **GRPO-trained LoRA adapter:** https://huggingface.co/Manasvee09/redbutton-v2-grpo-qwen3-4b-dense
113
  - **SFT-induced LoRA adapter (R3 negative result):** https://huggingface.co/Manasvee09/redbutton-v2-sft-qwen3-8b
114
  - **OpenEnv manifest:** [`openenv.yaml`](./openenv.yaml)
 
439
  author = {Sanjay, Arun},
440
  title = {Red Button v2: Multi-Agent Corrigibility RL Environment},
441
  year = {2026},
442
+ url = {https://huggingface.co/spaces/shutdowngym/RedButton-v2},
443
  }
444
  ```
445