update deprecated env var (#14975)

Signed-off-by: Maanu Grover <maanug@nvidia.com>

Author: maanug-nv

scripts/performance/executors.py

@@ -69,7 +69,7 @@ def slurm_executor(
     """
     PERF_ENV_VARS = {
         "TORCH_NCCL_AVOID_RECORD_STREAMS": "1",  # Disable caching NCCL communication buffer memory
-        "TRANSFORMERS_OFFLINE": "1",  # Enable online downloads from HuggingFace
+        "HF_HUB_OFFLINE": "1",  # Enable online downloads from HuggingFace
         "TOKENIZERS_PARALLELISM": "False",  # Restrict warning message prints
         "NCCL_NVLS_ENABLE": "0",  # Disable NVLink SHARP to save memory
         "NVTE_FLASH_ATTN": "1",  # Enable Flash Attention, which is needed to enable cuDNN fused attention
@@ -104,7 +104,7 @@ def slurm_executor(
         PERF_ENV_VARS["NEMO_HOME"] = nemo_home
         mounts.extend([f"{nemo_home}:{nemo_home}"])
     if hf_token is not None:
-        PERF_ENV_VARS.update({"HF_TOKEN": hf_token, "TRANSFORMERS_OFFLINE": "0"})
+        PERF_ENV_VARS.update({"HF_TOKEN": hf_token, "HF_HUB_OFFLINE": "0"})
 
     PERF_ENV_VARS |= custom_env_vars
     mounts.extend(custom_mounts)

scripts/performance/utils.py

@@ -44,7 +44,7 @@ def hf_tokenizer(model_name: str) -> run.Config[AutoTokenizer]:
         f"`AutoTokenizer` first searches for tokenizer files locally stored in {DEFAULT_NEMO_HOME}.",
         "(from env var `NEMO_HOME`- can be changed using '-nh/--nemo_home' CLI arg).",
         "If files are missing locally, `AutoTokenizer` will try downloading from HuggingFace. In this case-",
-        "make sure env vars 'TRANSFORMERS_OFFLINE':'0' and 'HF_TOKEN':'<token_value>' are set in your sbatch script.",
+        "make sure env vars 'HF_HUB_OFFLINE':'0' and 'HF_TOKEN':'<token_value>' are set in your sbatch script.",
         "Both of these will be set automatically if you provide '-hf/--hf_token' CLI arg.",
     ]
     logging.warning(" ".join(log_msg))