Upload pod_train.py with huggingface_hub

pod_train.py

@@ -18,12 +18,12 @@ print(f"Data: {HF_DATA_REPO}")
 print(f"Output: {HF_MODEL_REPO}")
 start_time = time.time()
 
-# Install dependencies — upgrade torch first (RunPod image has 2.4.0, need 2.6+ for set_submodule)
+# Install dependencies — pin exact compatible versions for torch 2.4.x (RunPod image)
+# transformers<4.46 avoids set_submodule (needs torch 2.5+)
+# trl<0.12 avoids processing_class kwarg (needs transformers 4.46+)
 subprocess.check_call([sys.executable, "-m", "pip", "install", "-q",
-    "torch>=2.6.0", "--index-url", "https://download.pytorch.org/whl/cu124"])
-subprocess.check_call([sys.executable, "-m", "pip", "install", "-q",
-    "transformers", "peft", "datasets", "accelerate", "bitsandbytes",
-    "huggingface_hub", "trl", "runpod"])
+    "transformers==4.45.2", "peft==0.12.0", "datasets",
+    "accelerate", "bitsandbytes", "huggingface_hub", "trl==0.11.4", "runpod"])
 
 from datasets import load_dataset
 from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments, BitsAndBytesConfig
@@ -74,8 +74,8 @@ model.print_trainable_parameters()
 # Training args
 output_dir = f"/workspace/{ADAPTER_NAME}-lora"
 num_examples = len(dataset)
-batch_size = 4
-grad_accum = 4  # effective batch = 16
+batch_size = 2
+grad_accum = 8  # effective batch = 16
 num_epochs = 3 if num_examples < 5000 else (2 if num_examples < 20000 else 1)
 warmup = min(100, num_examples // (batch_size * grad_accum))