Document v1.0.2 PUCT+PLN deployment
Browse files
README.md
CHANGED
|
@@ -188,4 +188,61 @@ held-out evaluation results.
|
|
| 188 |
- Checkpoints may contain optimizer state and other retained training metadata.
|
| 189 |
PyTorch checkpoint loading can execute pickle content; load only trusted files.
|
| 190 |
|
|
|
|
|
|
|
| 191 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 188 |
- Checkpoints may contain optimizer state and other retained training metadata.
|
| 189 |
PyTorch checkpoint loading can execute pickle content; load only trusted files.
|
| 190 |
|
| 191 |
+
<!-- deployment-v1.0.2-start -->
|
| 192 |
+
## Deployment: v1.0.2
|
| 193 |
|
| 194 |
+
Selected best.pt from PUCT+PLN training run 20261108-083020.
|
| 195 |
+
This is a serving bundle, not a complete training-resume archive.
|
| 196 |
+
|
| 197 |
+
| Identifier | Value |
|
| 198 |
+
|---|---|
|
| 199 |
+
| Repository | `iCogLabs-Dev/Lean-Proof-Search-Checkpoints` |
|
| 200 |
+
| Version | `v1.0.2` |
|
| 201 |
+
| Artifact commit | `264df74f6abb33e19a5467664699ad07eb955d89` |
|
| 202 |
+
| Manifest SHA-256 | `ed08e144022c6faf6a565d8e706371a8539bf406f9a756553e6ac42682033fa5` |
|
| 203 |
+
| Application commit | `2271bdb346d05b749bce773e55a962a6c810acb7` |
|
| 204 |
+
|
| 205 |
+
[Browse the pinned release](https://huggingface.co/iCogLabs-Dev/Lean-Proof-Search-Checkpoints/tree/264df74f6abb33e19a5467664699ad07eb955d89/deployments/mcts/v1.0.2).
|
| 206 |
+
|
| 207 |
+
### Compatible runtime
|
| 208 |
+
|
| 209 |
+
Use application commit `2271bdb346d05b749bce773e55a962a6c810acb7` and its Dockerfile and locked dependencies.
|
| 210 |
+
The image supplies Lean, the custom Pantograph REPL/client, Mathlib, and PLN.
|
| 211 |
+
The REPL source identity and compiled Mathlib must match the declaration inventory.
|
| 212 |
+
|
| 213 |
+
[Docker deployment guide](https://github.com/iCog-Labs-Dev/Mathematics-ai-atp/blob/2271bdb346d05b749bce773e55a962a6c810acb7/docker/HUGGING_FACE_DEPLOYMENT.md).
|
| 214 |
+
|
| 215 |
+
### Download and verify
|
| 216 |
+
|
| 217 |
+
From that application checkout, in its installed Python environment:
|
| 218 |
+
|
| 219 |
+
python -m maths_ai.gnn_inference.scripts.fetch_mcts_deployment \
|
| 220 |
+
--repo iCogLabs-Dev/Lean-Proof-Search-Checkpoints \
|
| 221 |
+
--revision 264df74f6abb33e19a5467664699ad07eb955d89 \
|
| 222 |
+
--version v1.0.2 \
|
| 223 |
+
--manifest-sha256 ed08e144022c6faf6a565d8e706371a8539bf406f9a756553e6ac42682033fa5 \
|
| 224 |
+
--output "$HOME/mcts-deployments/v1.0.2"
|
| 225 |
+
|
| 226 |
+
Keep both the downloaded directory symlink and its hidden sibling backing directory.
|
| 227 |
+
|
| 228 |
+
### CPU or GPU serving
|
| 229 |
+
|
| 230 |
+
After building the matching image, mount the downloaded directories:
|
| 231 |
+
|
| 232 |
+
export MATHS_AI_ARTIFACTS="$HOME/mcts-deployments/v1.0.2/artifacts"
|
| 233 |
+
export MATHS_AI_CONFIG="$HOME/mcts-deployments/v1.0.2/config"
|
| 234 |
+
|
| 235 |
+
For CPU:
|
| 236 |
+
|
| 237 |
+
MATHS_AI_DEVICE=cpu docker compose up -d mcts-api
|
| 238 |
+
|
| 239 |
+
Alternatively, on an NVIDIA-capable Docker host:
|
| 240 |
+
|
| 241 |
+
MATHS_AI_DEVICE=cuda:0 docker compose --profile gpu up -d mcts-api-gpu
|
| 242 |
+
|
| 243 |
+
Choose one service. Both use the same artifacts.
|
| 244 |
+
Mount paths must be accessible to the Docker daemon.
|
| 245 |
+
|
| 246 |
+
Verify /ready and a kernel-certified /prove result after startup.
|
| 247 |
+
Successful upload does not establish successful Docker or GPU execution.
|
| 248 |
+
<!-- deployment-v1.0.2-end -->
|