some updates on the env

README.md

@@ -27,12 +27,13 @@ Minimal setup to finetune a laptop-friendly Tulu checkpoint with QLoRA and track
 1) Create the env (Conda)
 ```bash
 conda env create -f environment.yml
-conda activate tulu-train
+conda activate deeai
 ```
 2) Add secrets (keep `.env` out of git)
 ```bash
 cp .env.example .env
 # Edit .env with your WANDB_API_KEY / project / entity
+# Optionally set BASE_MODEL_CACHE to choose where HF downloads models
 ```
 3) Verify packages (optional if you prefer pip)
 ```bash
@@ -59,8 +60,13 @@ Key flags:
 - Ensure `WANDB_API_KEY`, `WANDB_PROJECT`, and (optionally) `WANDB_ENTITY` are set in `.env`.
 - Each run captures hyperparameters and metrics; check the W&B UI for live loss curves and checkpoints.
 
+## Model cache location
+- Base model weights download to the Hugging Face cache. You can point downloads to an external directory by setting `BASE_MODEL_CACHE` in `.env` (e.g., `/Volumes/JTQ-s/______GITLAB____/downloaded_base_models`); the script maps this to `HF_HOME`/`TRANSFORMERS_CACHE` before loading models.
+- If `BASE_MODEL_CACHE` is not set, the default HF cache is used (typically `~/.cache/huggingface/hub`).
+
 ## Output
-- Finetuned adapters + tokenizer are written to `outputs/tulu-lora` (configurable via `--output_dir`). Push this to the Hub with `huggingface-cli upload` if desired.
+- Finetuned adapters + tokenizer are written to `outputs/tulu-lora` (configurable via `--output_dir`).
+- `outputs/` is tracked via Git LFS (`.gitattributes`), so weights can be committed and pushed to the Hub. Run `git lfs install` once, then `git add outputs/...` before committing.
 
 ## Troubleshooting
 - OOM? Reduce `max_seq_length`, increase `gradient_accumulation_steps`, or switch to a smaller dataset.

train_tulu.py

@@ -128,8 +128,17 @@ def parse_args() -> ScriptConfig:
     return ScriptConfig(**vars(args))
 
 
+def configure_cache_from_env():
+    """Allow user to redirect HF cache via BASE_MODEL_CACHE env."""
+    cache_dir = os.getenv("BASE_MODEL_CACHE")
+    if cache_dir:
+        os.environ.setdefault("HF_HOME", cache_dir)
+        os.environ.setdefault("TRANSFORMERS_CACHE", cache_dir)
+
+
 def main():
     load_dotenv()
+    configure_cache_from_env()
     cfg = parse_args()
 
     init_wandb(cfg)