Spaces:
Sleeping
Sleeping
Clarify how reviewers can run and inspect outputs
Browse files
README.md
CHANGED
|
@@ -217,6 +217,51 @@ Multiple **independent** reward signals (per `help_guide.md §7`) — reduces re
|
|
| 217 |
|
| 218 |
**Final episode score** = weighted blend of phase scores; see `server/rewards.py`.
|
| 219 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 220 |
## Setup
|
| 221 |
|
| 222 |
### Prerequisites
|
|
@@ -228,8 +273,8 @@ Multiple **independent** reward signals (per `help_guide.md §7`) — reduces re
|
|
| 228 |
### Local Development
|
| 229 |
|
| 230 |
```bash
|
| 231 |
-
git clone
|
| 232 |
-
cd api_contract_validator
|
| 233 |
pip install -e .
|
| 234 |
|
| 235 |
uvicorn server.app:app --host 0.0.0.0 --port 7860 --reload
|
|
|
|
| 217 |
|
| 218 |
**Final episode score** = weighted blend of phase scores; see `server/rewards.py`.
|
| 219 |
|
| 220 |
+
## How To Run And See Output
|
| 221 |
+
|
| 222 |
+
If you are reviewing this submission, the fastest path is:
|
| 223 |
+
|
| 224 |
+
1. **Try the hosted environment**:
|
| 225 |
+
|
| 226 |
+
```bash
|
| 227 |
+
curl https://pushpam14-api-contract-validator.hf.space/health
|
| 228 |
+
|
| 229 |
+
curl -X POST https://pushpam14-api-contract-validator.hf.space/reset \
|
| 230 |
+
-H "Content-Type: application/json" \
|
| 231 |
+
-d '{"task_name":"trace_downstream_blast_radius","seed":1}'
|
| 232 |
+
```
|
| 233 |
+
|
| 234 |
+
2. **Inspect the committed training outputs**:
|
| 235 |
+
|
| 236 |
+
- Reward curve: [`results/reward_curve.png`](results/reward_curve.png)
|
| 237 |
+
- Before/after chart: [`results/before_after.png`](results/before_after.png)
|
| 238 |
+
- Training proof: [`results/TRAINING_RUN_PROOF.md`](results/TRAINING_RUN_PROOF.md)
|
| 239 |
+
- Full training log: [`results/training_full_log.txt`](results/training_full_log.txt)
|
| 240 |
+
- Score files: [`../baseline_72b_v2_scores.json`](../baseline_72b_v2_scores.json), [`../baseline_7b_scores.json`](../baseline_7b_scores.json), [`../trained_scores.json`](../trained_scores.json)
|
| 241 |
+
|
| 242 |
+
3. **Re-run locally if desired**:
|
| 243 |
+
|
| 244 |
+
```bash
|
| 245 |
+
git clone https://github.com/kumarpushpam17-personal/Hackathon
|
| 246 |
+
cd Hackathon/api_contract_validator
|
| 247 |
+
pip install -e .
|
| 248 |
+
uvicorn server.app:app --host 0.0.0.0 --port 7860
|
| 249 |
+
```
|
| 250 |
+
|
| 251 |
+
In another terminal:
|
| 252 |
+
|
| 253 |
+
```bash
|
| 254 |
+
curl http://localhost:7860/health
|
| 255 |
+
openenv validate
|
| 256 |
+
```
|
| 257 |
+
|
| 258 |
+
4. **Re-run inference or training**:
|
| 259 |
+
|
| 260 |
+
- Baseline/trained inference: [`inference.py`](inference.py)
|
| 261 |
+
- Colab training notebook: [`training/grpo_colab.ipynb`](training/grpo_colab.ipynb)
|
| 262 |
+
- HF Jobs launcher used for the submitted run: [`training/run_in_hf_jobs.py`](training/run_in_hf_jobs.py)
|
| 263 |
+
- Full training instructions: [`training/README.md`](training/README.md)
|
| 264 |
+
|
| 265 |
## Setup
|
| 266 |
|
| 267 |
### Prerequisites
|
|
|
|
| 273 |
### Local Development
|
| 274 |
|
| 275 |
```bash
|
| 276 |
+
git clone https://github.com/kumarpushpam17-personal/Hackathon
|
| 277 |
+
cd Hackathon/api_contract_validator
|
| 278 |
pip install -e .
|
| 279 |
|
| 280 |
uvicorn server.app:app --host 0.0.0.0 --port 7860 --reload
|