Spaces:
Running on T4
Running on T4
Commit ·
434c6b1
1
Parent(s): faad7f2
Add train.sh startup script and assets folder
Browse files- train.sh: startup script for HF Spaces/Colab/Northflank that checks HF_TOKEN, installs train deps, and runs GRPO training
- assets/: placeholder folder for architecture diagram (architecture.jpg/png)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- assets/README.md +1 -0
- train.sh +33 -0
assets/README.md
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
# Drop architecture.jpg or architecture.png here
|
train.sh
ADDED
|
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
# ============================================================
|
| 3 |
+
# Training startup script for HF Spaces / Colab / Northflank
|
| 4 |
+
#
|
| 5 |
+
# Usage:
|
| 6 |
+
# ./train.sh # full run from config.yaml
|
| 7 |
+
# ./train.sh --steps 5 --episodes 3 # quick smoke test
|
| 8 |
+
# HF_TOKEN=hf_xxx ./train.sh # with inline token
|
| 9 |
+
# ============================================================
|
| 10 |
+
|
| 11 |
+
set -e
|
| 12 |
+
|
| 13 |
+
echo "============================================================"
|
| 14 |
+
echo " Nested RL Envs — GRPO Training"
|
| 15 |
+
echo " Team: Ludes Magnus"
|
| 16 |
+
echo "============================================================"
|
| 17 |
+
|
| 18 |
+
# Check HF_TOKEN
|
| 19 |
+
if [ -z "$HF_TOKEN" ]; then
|
| 20 |
+
echo "ERROR: HF_TOKEN environment variable is not set."
|
| 21 |
+
echo "Set it via: export HF_TOKEN=hf_xxx"
|
| 22 |
+
exit 1
|
| 23 |
+
fi
|
| 24 |
+
|
| 25 |
+
# Install training dependencies if not already installed
|
| 26 |
+
if ! python -c "import unsloth" 2>/dev/null; then
|
| 27 |
+
echo "Installing training dependencies..."
|
| 28 |
+
pip install -q -e ".[train]"
|
| 29 |
+
fi
|
| 30 |
+
|
| 31 |
+
# Run training
|
| 32 |
+
echo "Starting GRPO training..."
|
| 33 |
+
python -m layer1.train "$@"
|