Hn title generator

examples/hn_title_generator/skypilot-reference-grpo-trainer.yaml

@@ -0,0 +1,44 @@
+# To launch, run the following command from the root directory of the art repository:
+# `uv run sky launch examples/hn_title_generator/skypilot-reference-grpo-trainer.yaml --cluster=kyle-hn-title-generator-002 --env-file=.env --yes --retry-until-up --down --idle-minutes-to-autostop 60`
+
+resources:
+  image_id: pytorch/pytorch:2.5.1-cuda12.4-cudnn9-devel
+  cloud: runpod
+  region: US
+  accelerators:
+    - "H100-SXM"
+
+workdir: .
+
+envs:
+  HF_HUB_ENABLE_HF_TRANSFER: 1
+  VLLM_CONFIGURE_LOGGING: 0
+  REWARD_MODEL_URL:
+
+setup: |
+  apt-get update && apt-get install -y git
+
+  curl -LsSf https://astral.sh/uv/install.sh | sh
+
+  # Source the environment to make uv available
+  source $HOME/.local/bin/env
+
+  uv pip install --system \
+    unsloth==2025.3.19 \
+    vllm==0.8.2 \
+    bitsandbytes==0.45.4 \
+    datasets==3.3.2 \
+    s3fs==2024.12.0 \
+    hf-transfer==0.1.9 \
+    typer==0.15.2 \
+    fastapi==0.115.11 \
+    python-dotenv==1.0.1 \
+    polars==1.24.0 \
+    wandb==0.19.8 \
+    git+https://github.com/corbt/panza.git \
+
+  echo "Setup complete"
+
+run: |
+  echo "Running train_grpo.py"
+  uv run python examples/hn_title_generator/reference_grpo_trainer.py

examples/hn_title_generator/skypilot.yaml

@@ -0,0 +1,20 @@
+# To launch, run the following command from the root directory of the art repository:
+# `uv run sky launch examples/hn_title_generator/skypilot.yaml --cluster=kyle-hn-title-generator-001 --env-file=.env --yes --retry-until-up --down --idle-minutes-to-autostop 60`
+
+workdir: .
+resources:
+  accelerators: ["H100-SXM:1"]
+envs:
+  HF_HUB_ENABLE_HF_TRANSFER: 1
+  REWARD_MODEL_URL:
+
+setup: |
+  curl -LsSf https://astral.sh/uv/install.sh | sh
+
+  source $HOME/.local/bin/env
+
+  uv sync
+
+run: |
+  echo "Running training script..."
+  uv run python examples/hn_title_generator/train.py