Upload train_question_generator.py with huggingface_hub

train_question_generator.py

@@ -1,13 +1,11 @@
 # /// script
-# dependencies = ["trl>=0.12.0", "peft>=0.13.0", "datasets", "transformers>=4.45.0", "accelerate", "bitsandbytes", "huggingface_hub"]
+# dependencies = ["trl>=0.20.0", "peft>=0.13.0", "datasets", "transformers>=4.45.0", "accelerate", "bitsandbytes", "huggingface_hub"]
 # ///
 
 import os
 from datasets import load_dataset
 from peft import LoraConfig
 from trl import SFTTrainer, SFTConfig
-from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
-import torch
 
 # Authenticate
 from huggingface_hub import login
@@ -26,26 +24,8 @@ if eval_dataset:
     print(f"Eval samples: {len(eval_dataset)}")
 
 # Using Qwen2.5-7B for question generation
-print("Loading model: Qwen/Qwen2.5-7B-Instruct...")
 model_id = "Qwen/Qwen2.5-7B-Instruct"
-
-# 4-bit quantization for fitting on A10G
-bnb_config = BitsAndBytesConfig(
-    load_in_4bit=True,
-    bnb_4bit_quant_type="nf4",
-    bnb_4bit_compute_dtype=torch.bfloat16,
-)
-
-tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
-if tokenizer.pad_token is None:
-    tokenizer.pad_token = tokenizer.eos_token
-
-model = AutoModelForCausalLM.from_pretrained(
-    model_id,
-    quantization_config=bnb_config,
-    device_map="auto",
-    trust_remote_code=True,
-)
+print(f"Using model: {model_id}")
 
 # LoRA config - slightly higher rank for more complex task
 peft_config = LoraConfig(
@@ -57,7 +37,7 @@ peft_config = LoraConfig(
     task_type="CAUSAL_LM",
 )
 
-# Training config - removed max_seq_length, gradient_checkpointing from SFTConfig
+# Training config - modern TRL API
 training_args = SFTConfig(
     output_dir="./question-generator-output",
     num_train_epochs=2,
@@ -73,17 +53,17 @@ training_args = SFTConfig(
     push_to_hub=True,
     hub_model_id="KevinKeller/cognitive-question-generator-qwen2.5-7b",
     report_to="none",
+    max_seq_length=8192,
+    load_in_4bit=True,  # Enable 4-bit quantization
 )
 
 print("Starting training...")
 trainer = SFTTrainer(
-    model=model,
+    model=model_id,  # Pass model name, not loaded model
     train_dataset=train_dataset,
     eval_dataset=eval_dataset,
     peft_config=peft_config,
-    processing_class=tokenizer,  # Renamed from 'tokenizer' in TRL 0.12+
     args=training_args,
-    max_seq_length=8192,
 )
 
 trainer.train()