Upload Base/Datasets/rag_mcp_sft/FINETUNE_COMMANDS.md with huggingface_hub

Base/Datasets/rag_mcp_sft/FINETUNE_COMMANDS.md

@@ -0,0 +1,54 @@
+# LoRA Finetune Commands
+
+These commands assume you are running from a PowerShell terminal on Windows.
+
+## 1. Clone the workspace repo
+
+```powershell
+git clone <your-luna-repo-url>
+cd LUNA
+```
+
+## 2. Create the environment and install dependencies
+
+```powershell
+python -m venv .venv
+.\.venv\Scripts\Activate.ps1
+python -m pip install --upgrade pip
+pip install -r requirements.txt
+```
+
+## 3. Build the RAG + MCP dataset locally
+
+```powershell
+python .\Base\Datasets\rag_mcp_sft\build_rag_mcp_sft_dataset.py --target-tokens 10000000
+```
+
+## 4. Push the dataset to Hugging Face
+
+```powershell
+$env:HF_TOKEN = "<your_hf_token>"
+python .\Base\Datasets\rag_mcp_sft\push_to_hf.py --repo-id ASTERIZER/LUNA-RAG-MCP-SFT-10M
+```
+
+## 5. Train LoRA adapters on top of the current LUNA SFT checkpoint from Hugging Face
+
+The config already points to the user-requested base checkpoint file:
+
+- repo: `ASTERIZER/LUNA-100M`
+- file: `sft_v1/final/model.pth`
+
+Run:
+
+```powershell
+$env:HF_TOKEN = "<your_hf_token>"
+python .\lora_sft_train.py --config .\rag_mcp_lora_config.yaml
+```
+
+## 6. Optional overrides
+
+```powershell
+python .\lora_sft_train.py --config .\rag_mcp_lora_config.yaml --epochs 3
+python .\lora_sft_train.py --config .\rag_mcp_lora_config.yaml --out_dir .\Base\out\sft\rag_mcp_lora_exp2
+python .\lora_sft_train.py --config .\rag_mcp_lora_config.yaml --pretrained_ckpt .\Base\out\input_models\luna_sft_v1\model.pth
+```