Update train.py

train.py

@@ -1,9 +1,9 @@
 # ==============================================
-# Stephen Model Fine-Tuning Script (LoRA + PEFT) — RunPod Ready
-# Updated: Aug 2025
+# Stephen Model Fine-Tuning Script (LoRA + PEFT)
+# Optimized for JSONL dataset
 # ==============================================
 
-# !pip install -q "transformers>=4.44.0" "datasets" "peft>=0.12.0" accelerate bitsandbytes sentencepiece huggingface_hub
+!pip install -q "transformers>=4.44.0" "datasets" "peft>=0.12.0" accelerate bitsandbytes sentencepiece huggingface_hub
 
 import os
 from datetime import datetime
@@ -38,13 +38,15 @@ login(token=HF_TOKEN, add_to_git_credential=True)
 log(f"Logged in as: {whoami()['name']} ✅")
 
 # ==============================================
-# 2. Load Dataset
+# 2. Load JSONL Dataset from HF
 # ==============================================
 dataset_name = "dgtalbug/stephen-dataset"
-log(f"Loading dataset: {dataset_name}...")
-dataset = load_dataset(dataset_name)
-log(f"Dataset loaded. Splits: {list(dataset.keys())}")
-log(f"First example: {dataset['train'][0]}")
+data_file = "stephen.jsonl"
+
+log(f"Loading dataset: {dataset_name}/{data_file} ...")
+dataset = load_dataset(dataset_name, data_files=data_file, split="train")
+log(f"Dataset loaded — {len(dataset)} rows")
+log(f"First example: {dataset[0]}")
 
 # ==============================================
 # 3. Load Base Model & Tokenizer
@@ -88,14 +90,14 @@ log("Configuring LoRA...")
 lora_config = LoraConfig(
     r=16,
     lora_alpha=32,
-    target_modules=["q_proj", "v_proj", "k_proj", "o_proj"],  # StableCode safe
+    target_modules=["q_proj", "v_proj", "k_proj", "o_proj"],  # Safe for StableCode
     lora_dropout=0.05,
     bias="none",
     task_type="CAUSAL_LM"
 )
 model = get_peft_model(model, lora_config)
 
-# ✅ Fix requires_grad warning
+# ✅ Enable training for LoRA params
 for name, param in model.named_parameters():
     if "lora" in name:
         param.requires_grad = True
@@ -103,21 +105,27 @@ for name, param in model.named_parameters():
 log("LoRA config applied ✅")
 
 # ==============================================
-# 5. Tokenize Dataset
+# 5. Tokenize Dataset (JSONL)
 # ==============================================
 log("Tokenizing dataset...")
 
+# Detect correct text key
+first_row = dataset[0]
+if "text" in first_row:
+    text_key = "text"
+elif "prompt" in first_row:
+    text_key = "prompt"
+else:
+    text_key = list(first_row.keys())[0]  # fallback
+
+log(f"Using text key: '{text_key}'")
+
 def tokenize_fn(example):
-    text = f"{example.get('character', '')}: {example.get('line', '')}".strip()
-    return tokenizer(text, truncation=True, padding="max_length", max_length=512)
+    return tokenizer(example[text_key], truncation=True, padding="max_length", max_length=512)
 
-tokenized_datasets = dataset.map(
-    tokenize_fn,
-    batched=True,
-    remove_columns=dataset["train"].column_names
-)
+tokenized_dataset = dataset.map(tokenize_fn, batched=True, remove_columns=dataset.column_names)
 log("Tokenization complete ✅")
-log(f"Tokenized sample: {tokenized_datasets['train'][0]}")
+log(f"Tokenized sample: {tokenized_dataset[0]}")
 
 # ==============================================
 # 6. Data Collator
@@ -152,13 +160,12 @@ training_args = TrainingArguments(
 log("Training arguments ready ✅")
 
 # ==============================================
-# 8. Custom Trainer — Prevent ValueError
+# 8. Custom Trainer
 # ==============================================
 class NoUnpackTrainer(Trainer):
-    """Custom Trainer that avoids Hugging Face's unpacking bug."""
+    """Avoids unpacking bug in HF Trainer."""
     def compute_loss(self, model, inputs, return_outputs=False, num_items_in_batch=None):
         inputs = self._prepare_inputs(inputs)
-        # Ensure labels exist for LM
         if "labels" not in inputs:
             inputs["labels"] = inputs["input_ids"].clone()
         outputs = model(**inputs)
@@ -169,13 +176,13 @@ log("Initializing Trainer...")
 trainer = NoUnpackTrainer(
     model=model,
     args=training_args,
-    train_dataset=tokenized_datasets["train"],
+    train_dataset=tokenized_dataset,
     data_collator=data_collator
 )
 log("Trainer initialized ✅")
 
 # ==============================================
-# 9. Train & Push to Hub (Checkpoint Safe)
+# 9. Train & Push to Hub
 # ==============================================
 log("Starting training...")
 last_ckpt = None