Basic LORA trainer, doesn't seem to make any speed difference

pdelfin/train/config/qwen2vl-2b.yaml

@@ -3,8 +3,8 @@ model:
   arch: causal
 
 wandb:
-  project: refine
-  entity: pdf-qwen2vl
+  project: pdelfin
+  entity: ai2-llm
 
 # TODO This is not used
 format:
@@ -93,10 +93,10 @@ hparams:
   gradient_checkpointing: true
   clip_grad_norm: 1.0
   learning_rate: 3e-4
-  max_steps: 10000
+  max_steps: 200
   pad_multiple_of: 16
   log_every_steps: 5
-  eval_every_steps: 250
+  eval_every_steps: 100
   optim: adamw_torch
   lr_scheduler: cosine
   weight_decay: 0.01
@@ -118,7 +118,7 @@ lora:
       - down_proj
 
 save:
-  path: s3://ai2-tylerm-experimental/experiments/rephrase/v1/models/lucas
-  save_every_steps: 500
+  path: s3://ai2-oe-data/jakep/experiments/qwen2vl-pdf/v1/models/
+  save_every_steps: 100
 
 max_workers: 1

pdelfin/train/train.py

@@ -123,6 +123,8 @@ def run_train(config: TrainConfig):
 
     accelerator = accelerate.Accelerator()
 
+    setup_environment(aws_config=config.aws, wandb_config=config.wandb, WANDB_RUN_GROUP=run_name.group)
+
     train_ds = build_batch_query_response_vision_dataset(
                         query_glob_path="s3://ai2-oe-data/jakep/openai_batch_data_v2_mini/*.jsonl",
                         response_glob_path="s3://ai2-oe-data/jakep/openai_batch_done_v2_mini/*.json",
@@ -133,10 +135,25 @@ def run_train(config: TrainConfig):
     )
     processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct")
 
+    if config.lora is not None:
+        peft_config = LoraConfig(
+            r=config.lora.rank,
+            lora_alpha=config.lora.alpha,
+            lora_dropout=config.lora.dropout,
+            bias=config.lora.bias,  # pyright: ignore
+            task_type=config.lora.task_type,
+            target_modules=list(config.lora.target_modules),
+        )
+        model = get_peft_model(model=model, peft_config=peft_config)
+        log_trainable_parameters(model=model, logger=logger)
+
     train_ds = train_ds.with_transform(partial(batch_prepare_data_for_qwen2_training, processor=processor))
     print(train_ds)
     print("---------------")
 
+    save_path = join_path("", config.save.path, run_name.run)
+
+    save_config(config, join_path("", save_path, "config.yaml"))  # pyright: ignore
 
     with TemporaryDirectory() as output_dir:
 
@@ -177,22 +194,31 @@ def run_train(config: TrainConfig):
 
         # Set the collator
         collator = partial(packing_collator, pad_multiple_of=config.hparams.pad_multiple_of, do_shrink=False)
-        #checkpoint_callback = CheckpointUploadCallback(save_path=save_path, logger=logger)
-
-        # # Initialize Trainer
-        # trainer = Trainer(
-        #     model=model,
-        #     args=training_args,
-        #     train_dataset=train_ds,
-        #     #eval_dataset=formatted_dataset["validation"],  # pyright: ignore
-        #     tokenizer=processor.tokenizer,
-        #     #data_collator=collator,
-        #     #callbacks=[checkpoint_callback],
-        # )
-
-
-        # # Train the model
-        # trainer.train()  # pyright: ignore
+        checkpoint_callback = CheckpointUploadCallback(save_path=save_path, logger=logger)
+
+        # Initialize Trainer
+        trainer = Trainer(
+            model=model,
+            args=training_args,
+            train_dataset=train_ds,
+            #eval_dataset=formatted_dataset["validation"],  # pyright: ignore
+            tokenizer=processor.tokenizer,
+            #data_collator=collator,
+            #callbacks=[checkpoint_callback],
+        )
+
+        # Could not get this to work
+        # if get_rank() == 0:
+        #     # this is a hack to add script and peft config to wandb config
+        #     update_wandb_config(config, trainer, model)
+
+        # Train the model
+        trainer.train()  # pyright: ignore
+
+        with get_local_dir(join_path("", save_path, "best")) as best_dir:
+            model.save_pretrained(best_dir)
+            tokenizer.tokenizer.save_pretrained(best_dir)
+            logger.info("Saved best model to %s", best_dir)
 
         # Uncomment to test speed of data loader
         # train_dataloader = DataLoader(train_ds, batch_size=1, num_workers=2, shuffle=False)