[3/n][torchtune integration] add validation logic (#600)

## What does this PR do?
- add validation logic in SFT recipe (validation loss and perplexity)
- add progress bar in both training and validation to better track the
progress on server side (eval has the similar logic)


## Test Plan
validation logic shows up in the Checkpoint training_metric part  
<img width="799" alt="Screenshot 2024-12-12 at 3 21 52 PM"
src="https://github.com/user-attachments/assets/36330ffe-0555-4b2d-93f0-9487dfdf7b4e"
/>

progress bar shows up as 
<img width="476" alt="Screenshot 2024-12-12 at 3 38 11 PM"
src="https://github.com/user-attachments/assets/77306fa2-cb9c-460f-8efc-b41bbe424a7d"
/>
expected

llama_stack/providers/inline/post_training/torchtune/recipes/lora_finetuning_single_device.py

@@ -23,6 +23,7 @@ from llama_stack.providers.inline.post_training.torchtune.common.checkpointer im
 from torch import nn
 from torchtune import utils as torchtune_utils
 from torchtune.training.metric_logging import DiskLogger
+from tqdm import tqdm
 from llama_stack.apis.post_training import *  # noqa
 from llama_stack.distribution.utils.model_utils import model_local_dir
 
@@ -185,11 +186,21 @@ class LoraFinetuningSingleDevice:
         self._model.set_num_output_chunks(self._loss_fn.num_output_chunks)
         log.info("Loss is initialized.")
 
-        self._sampler, self._dataloader = await self._setup_data(
+        self._training_sampler, self._training_dataloader = await self._setup_data(
+            dataset_id=self.training_config.data_config.dataset_id,
             tokenizer=self._tokenizer,
             shuffle=self._shuffle,
             batch_size=self._batch_size,
         )
+
+        if self.training_config.data_config.validation_dataset_id:
+            _, self._validation_dataloader = await self._setup_data(
+                dataset_id=self.training_config.data_config.validation_dataset_id,
+                tokenizer=self._tokenizer,
+                shuffle=False,
+                batch_size=self._batch_size,
+            )
+
         log.info("Dataset and Sampler are initialized.")
 
         # Number of training steps in each epoch depends on the number of batches produced
@@ -197,7 +208,7 @@ class LoraFinetuningSingleDevice:
         # for logging and tracking training state. This should be computed after the dataloader
         # has been setup
         self._steps_per_epoch = (
-            len(self._dataloader) // self._gradient_accumulation_steps
+            len(self._training_dataloader) // self._gradient_accumulation_steps
         )
         if (
             self.max_steps_per_epoch is not None
@@ -316,17 +327,19 @@ class LoraFinetuningSingleDevice:
         return optimizer
 
     async def _setup_data(
-        self, tokenizer: Llama3Tokenizer, shuffle: bool, batch_size: int
+        self,
+        dataset_id: str,
+        tokenizer: Llama3Tokenizer,
+        shuffle: bool,
+        batch_size: int,
     ) -> Tuple[DistributedSampler, DataLoader]:
-        dataset_id = self.training_config.data_config.dataset_id
-
-        async def fetch_rows():
+        async def fetch_rows(dataset_id: str):
             return await self.datasetio_api.get_rows_paginated(
                 dataset_id=dataset_id,
                 rows_in_page=-1,
             )
 
-        all_rows = await fetch_rows()
+        all_rows = await fetch_rows(dataset_id)
         rows = all_rows.rows
 
         # Curretly only support alpaca instruct dataset
@@ -460,9 +473,11 @@ class LoraFinetuningSingleDevice:
             metric_logger = DiskLogger(
                 log_dir=self._output_dir + f"/{self.model_id}-sft-{curr_epoch}"
             )
-            self._sampler.set_epoch(curr_epoch)
+            self._training_sampler.set_epoch(curr_epoch)
+            loss_to_log = 0.0
 
-            for idx, batch in enumerate(self._dataloader):
+            pbar = tqdm(total=self._steps_per_epoch)
+            for idx, batch in enumerate(self._training_dataloader):
                 if (
                     self.max_steps_per_epoch is not None
                     and (idx // self._gradient_accumulation_steps)
@@ -499,6 +514,12 @@ class LoraFinetuningSingleDevice:
                     self.global_step += 1
 
                     loss_to_log = running_loss.item() / num_tokens
+
+                    pbar.update(1)
+                    pbar.set_description(
+                        f"{curr_epoch + 1}|{self.global_step}|Loss: {loss_to_log}"
+                    )
+
                     time_per_step = time.perf_counter() - t0
                     log_dict = {
                         "loss": loss_to_log,
@@ -532,6 +553,44 @@ class LoraFinetuningSingleDevice:
                 post_training_job_id=self.job_uuid,
                 path=checkpoint_path,
             )
+            if self.training_config.data_config.validation_dataset_id:
+                validation_loss, perplexity = await self.validation()
+                training_metrics = PostTrainingMetric(
+                    epoch=curr_epoch,
+                    train_loss=loss_to_log,
+                    validation_loss=validation_loss,
+                    perplexity=perplexity,
+                )
+                checkpoint.training_metrics = training_metrics
             checkpoints.append(checkpoint)
 
         return (memory_stats, checkpoints)
+
+    async def validation(self) -> Tuple[float, float]:
+        total_loss = 0.0
+        total_tokens = 0
+        log.info("Starting validation...")
+        pbar = tqdm(total=len(self._validation_dataloader))
+        for idx, batch in enumerate(self._validation_dataloader):
+            if idx == 10:
+                break
+            torchtune_utils.batch_to_device(batch, self._device)
+
+            # Calculate the number of unmasked tokens in the current batch
+            # and increment the total number of tokens seen in the step
+            num_tokens = (batch["labels"] != self._loss_fn.ignore_index).sum()
+
+            # Loss is normalized by default so we multiply by the number of tokens
+            # This way we can normalize by the total number of tokens if we're accumulating gradients
+            loss = await self._loss_step(batch) * num_tokens
+
+            total_loss += loss
+            total_tokens += num_tokens
+
+            pbar.update(1)
+            pbar.set_description(f"validation step: {idx}")
+
+        mean_loss = total_loss / total_tokens
+        perplexity = torch.exp(torch.tensor(mean_loss))
+
+        return mean_loss, perplexity.item()