chore: add hf cli

Author: fin-griffin

sweep/README.md

@@ -5,11 +5,12 @@ LoRA rank, learning rate, batch size, and target layer configurations.
 
 ## Setup on the H100
 
-**1. Clone the repo and install dependencies**
+**1. Clone the repo, install dependencies, and activate the venv**
 
 ```bash
 git clone <repo-url> && cd voice
 uv sync
+source .venv/bin/activate
 ```
 
 **2. Set environment variables**
@@ -24,15 +25,15 @@ WANDB_API_KEY=...
 **3. Log in to Hugging Face and W&B**
 
 ```bash
-uv run huggingface-cli login
-uv run wandb login
+huggingface-cli login
+wandb login
 ```
 
 **4. Verify the grid looks correct**
 
 ```bash
 cd sweep
-uv run python sweep.py --dry-run
+python sweep.py --dry-run
 ```
 
 ## Running the sweep
@@ -42,9 +43,9 @@ Start a tmux session so the sweep survives SSH disconnection:
 ```bash
 tmux new -s sweep
 cd sweep
-uv run python sweep.py
+python sweep.py
 # Override GPU if needed (default is device 3):
-# uv run python sweep.py --device 0
+# python sweep.py --device 0
 # Ctrl+B, D  →  detach (safe to close laptop)
 # tmux attach -t sweep  →  reconnect later
 ```
@@ -57,7 +58,7 @@ Progress is logged to `sweep.log` and to the W&B project `VOICE`.
 it will skip completed runs and retry failed ones:
 
 ```bash
-uv run python sweep.py
+python sweep.py
 ```
 
 ## Outputs

sweep/sweep.py

@@ -9,9 +9,9 @@
 
 Usage (inside tmux on the H100)::
 
-    uv run python sweep.py
-    uv run python sweep.py --dry-run        # print grid without training
-    uv run python sweep.py --device 0       # override GPU (default: 3)
+    python sweep.py
+    python sweep.py --dry-run   # print grid without training
+    python sweep.py --device 0  # override GPU (default: 3)
 """
 
 import argparse