Link real GitHub training scripts gist

CADFORGE_BLOG.md

@@ -7,7 +7,7 @@ Detailed judge-facing blog with screenshots, diagrams, references, and all run i
 ## Judge Reproducibility Links
 
 - **GitHub repo:** [sanjuhs/open-env-meta-final-hackathon](https://github.com/sanjuhs/open-env-meta-final-hackathon)
-- **Training script bundle / gist-ready source:** [training/GITHUB_GIST_TRAINING_SCRIPTS.md](https://github.com/sanjuhs/open-env-meta-final-hackathon/blob/main/training/GITHUB_GIST_TRAINING_SCRIPTS.md)
+- **GitHub Gist: training scripts:** [CADForge OpenEnv SFT/GRPO scripts](https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb)
 - **Google Colab smoke notebook:** [cadforge_openenv_training_colab.ipynb](https://colab.research.google.com/github/sanjuhs/open-env-meta-final-hackathon/blob/main/training/cadforge_openenv_training_colab.ipynb)
 - **Raw training logs and evidence:** [sanjuhs/cadforge-training-evidence](https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence)
 

README.md

@@ -22,7 +22,7 @@ The agent receives a design request, writes a complete CadQuery Python file, and
 ## Judge-Facing Links
 
 - **GitHub repo:** [sanjuhs/open-env-meta-final-hackathon](https://github.com/sanjuhs/open-env-meta-final-hackathon)
-- **Training script bundle / gist-ready source:** [training/GITHUB_GIST_TRAINING_SCRIPTS.md](https://github.com/sanjuhs/open-env-meta-final-hackathon/blob/main/training/GITHUB_GIST_TRAINING_SCRIPTS.md)
+- **GitHub Gist: training scripts:** [CADForge OpenEnv SFT/GRPO scripts](https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb)
 - **Raw training logs and evidence:** [sanjuhs/cadforge-training-evidence](https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence)
 - Training notebook on this HF Space: [training/cadforge_openenv_training_colab.ipynb](training/cadforge_openenv_training_colab.ipynb)
 - Open the same notebook in Google Colab: [Colab training notebook](https://colab.research.google.com/github/sanjuhs/open-env-meta-final-hackathon/blob/main/training/cadforge_openenv_training_colab.ipynb)

docs/detailed-blog/cadforge-detailed-blog.md

@@ -11,7 +11,7 @@ CADForge is an OpenEnv reinforcement-learning environment for code-CAD. The agen
 | GitHub repo | [sanjuhs/open-env-meta-final-hackathon](https://github.com/sanjuhs/open-env-meta-final-hackathon) |
 | Google Colab smoke notebook | [cadforge_openenv_training_colab.ipynb](https://colab.research.google.com/github/sanjuhs/open-env-meta-final-hackathon/blob/main/training/cadforge_openenv_training_colab.ipynb) |
 | HF Space | [sanjuhs/cadforge-cadquery-openenv](https://huggingface.co/spaces/sanjuhs/cadforge-cadquery-openenv) |
-| Training script bundle / gist-ready source | [training/GITHUB_GIST_TRAINING_SCRIPTS.md](https://github.com/sanjuhs/open-env-meta-final-hackathon/blob/main/training/GITHUB_GIST_TRAINING_SCRIPTS.md) |
+| GitHub Gist: training scripts | [CADForge OpenEnv SFT/GRPO scripts](https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb) |
 | Raw logs and evidence archive | [sanjuhs/cadforge-training-evidence](https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence) |
 
 The full SFT and GRPO runs were executed on a RunPod H200 through distinct production scripts. The Colab notebook is the public judge-runnable smoke path: it validates OpenEnv, loads the public dataset, runs the CadQuery reward backend, and launches tiny SFT/GRPO checks using the same source files.

server/app.py

@@ -353,7 +353,7 @@ SPACE_HTML = r'''
           <a class="button" href="https://huggingface.co/spaces/sanjuhs/cadforge-cadquery-openenv/blob/main/CADFORGE_BLOG.md" target="_blank">Read mini-blog</a>
           <a class="button" href="https://huggingface.co/spaces/sanjuhs/cadforge-cadquery-openenv/blob/main/docs/detailed-blog/cadforge-detailed-blog.md" target="_blank">Detailed blog</a>
           <a class="button" href="https://github.com/sanjuhs/open-env-meta-final-hackathon" target="_blank">Training code</a>
-          <a class="button" href="https://github.com/sanjuhs/open-env-meta-final-hackathon/blob/main/training/GITHUB_GIST_TRAINING_SCRIPTS.md" target="_blank">Script gist bundle</a>
+          <a class="button" href="https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb" target="_blank">Training scripts Gist</a>
           <a class="button" href="https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence" target="_blank">Training logs</a>
         </div>
       </div>
@@ -396,7 +396,7 @@ SPACE_HTML = r'''
       <div class="repro-links">
         <a class="repro-link" href="https://github.com/sanjuhs/open-env-meta-final-hackathon" target="_blank">GitHub repo</a>
         <a class="repro-link" href="https://colab.research.google.com/github/sanjuhs/open-env-meta-final-hackathon/blob/main/training/cadforge_openenv_training_colab.ipynb" target="_blank">Open Colab notebook</a>
-        <a class="repro-link" href="https://github.com/sanjuhs/open-env-meta-final-hackathon/blob/main/training/GITHUB_GIST_TRAINING_SCRIPTS.md" target="_blank">Training script bundle</a>
+        <a class="repro-link" href="https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb" target="_blank">Training scripts Gist</a>
         <a class="repro-link" href="https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence" target="_blank">HF logs + evidence</a>
       </div>
     </section>

training/GITHUB_GIST_TRAINING_SCRIPTS.md

@@ -1,6 +1,8 @@
 # CADForge Training Scripts Bundle
 
-This is the gist-ready source bundle for the CADForge OpenEnv submission.
+This is the source bundle mirrored into the public CADForge OpenEnv training-script Gist:
+
+https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb
 
 **Main GitHub repo:** https://github.com/sanjuhs/open-env-meta-final-hackathon  
 **HF Space:** https://huggingface.co/spaces/sanjuhs/cadforge-cadquery-openenv  
@@ -42,9 +44,9 @@ The production run used shell wrappers so each stage could run in the right Pyth
 - Strict GRPO eval report: https://huggingface.co/spaces/sanjuhs/cadforge-cadquery-openenv/blob/main/training/eval/qwen35-9b-cadforge-grpo-strict-build-20260426-strict-build/eval_report.md
 - Detailed technical blog: https://huggingface.co/spaces/sanjuhs/cadforge-cadquery-openenv/blob/main/docs/detailed-blog/cadforge-detailed-blog.md
 
-## Optional: Create the GitHub Gist
+## Recreate or Update the GitHub Gist
 
-After authenticating GitHub CLI with gist permission:
+After authenticating GitHub CLI with gist permission, this is the command used to create the public Gist:
 
 ```bash
 gh auth login --scopes gist,repo

training/cadforge_openenv_training_colab.ipynb

@@ -11,7 +11,7 @@
     "**GitHub repo:** [https://github.com/sanjuhs/open-env-meta-final-hackathon](https://github.com/sanjuhs/open-env-meta-final-hackathon)  \n",
     "**Run this notebook in Google Colab:** [https://colab.research.google.com/github/sanjuhs/open-env-meta-final-hackathon/blob/main/training/cadforge_openenv_training_colab.ipynb](https://colab.research.google.com/github/sanjuhs/open-env-meta-final-hackathon/blob/main/training/cadforge_openenv_training_colab.ipynb)  \n",
     "**HF Space:** [https://huggingface.co/spaces/sanjuhs/cadforge-cadquery-openenv](https://huggingface.co/spaces/sanjuhs/cadforge-cadquery-openenv)  \n",
-    "**Training script bundle / gist-ready source:** [https://github.com/sanjuhs/open-env-meta-final-hackathon/blob/main/training/GITHUB_GIST_TRAINING_SCRIPTS.md](https://github.com/sanjuhs/open-env-meta-final-hackathon/blob/main/training/GITHUB_GIST_TRAINING_SCRIPTS.md)  \n",
+    "**GitHub Gist: training scripts:** [https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb](https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb)  \n",
     "**Raw training logs + evidence archive:** [https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence](https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence)\n",
     "\n",
     "## What actually ran where\n",
@@ -59,7 +59,7 @@
     "- Hugging Face Space repo: https://huggingface.co/spaces/sanjuhs/cadforge-cadquery-openenv\n",
     "\n",
     "- GitHub repo, exact source: https://github.com/sanjuhs/open-env-meta-final-hackathon\n",
-    "- Training script bundle / gist-ready source: https://github.com/sanjuhs/open-env-meta-final-hackathon/blob/main/training/GITHUB_GIST_TRAINING_SCRIPTS.md\n",
+    "- GitHub Gist: training scripts: https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb\n",
     "- Raw training logs and evidence bundle: https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence\n",
     "- RunPod/H200 clarification: full 2B/9B SFT and GRPO runs were executed as distinct scripts on RunPod; this Colab is the public smoke rerun path.\n",
     "- Training dataset: https://huggingface.co/datasets/sanjuhs/cadforge-cadquery-agentic-traces\n",
@@ -524,7 +524,7 @@
     "\n",
     "## RunPod H200 production scripts\n",
     "\n",
-    "The real training was launched on RunPod H200 through shell wrappers and Python scripts, then backed up to Hugging Face. Judges can inspect the source bundle here: https://github.com/sanjuhs/open-env-meta-final-hackathon/blob/main/training/GITHUB_GIST_TRAINING_SCRIPTS.md\n",
+    "The real training was launched on RunPod H200 through shell wrappers and Python scripts, then backed up to Hugging Face. Judges can inspect the public training-script Gist here: https://gist.github.com/sanjuhs/10596f688e8b4560910a3b1b137bfeeb\n",
     "\n",
     "Raw logs and charts are here: https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence\n"
    ]