fix: update HF token, echo it at runtime, export both HF_TOKEN and HUGGING_FACE_HUB_TOKEN

scripts/model_opt_nvfp4_full.py

@@ -40,8 +40,8 @@ KV_CACHE_QUANT = "fp8_cast"
 # 2.8TB RAM is enough for the 3TB model (with memory-mapped loading)
 EXTRA_FLAGS = "--trust_remote_code --use_seq_device_map --gpu_max_mem_percentage 0.7"
 
-# HF token for gated calibration datasets (nvidia/Nemotrons-Post-Training-Dataset-v2)
-HF_TOKEN = "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"
+# HF token for gated calibration datasets (nvidia/Nemotron-Post-Training-Dataset-v2)
+HF_TOKEN = "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"
 
 # Output dir follows modelopt convention: <model>_<quant>_kv_<kv_quant>
 # We override the model name to make the strategy clear
@@ -53,7 +53,9 @@ LOG_FILE = f"/root/nvidia-meeting/modelopt_{QUANT}.log"
 # ── Run ─────────────────────────────────────────────────────────────────────
 cmd = f"""cd {SCRIPT_DIR} && \\
 . /root/nvidia-meeting/venv/bin/activate && \\
-HF_TOKEN={HF_TOKEN} \\
+export HF_TOKEN={HF_TOKEN} && \\
+export HUGGING_FACE_HUB_TOKEN={HF_TOKEN} && \\
+echo "HF_TOKEN=$HF_TOKEN" && \\
 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True \\
 bash scripts/huggingface_example.sh \\
     --model {MODEL} \\
@@ -66,6 +68,7 @@ bash scripts/huggingface_example.sh \\
 print(f"Running: {QUANT} quantization on {MODEL}")
 print(f"Output: {OUTPUT_NAME}")
 print(f"Log: {LOG_FILE}")
+print(f"HF_TOKEN: {HF_TOKEN}")
 print(f"Command:\n{cmd}\n")
 
 ret = subprocess.call(cmd, shell=True)