refine

llama_stack/apis/post_training/post_training.py

@@ -68,6 +68,7 @@ class TrainingConfig(BaseModel):
     n_epochs: int
     max_steps_per_epoch: int
     gradient_accumulation_steps: int
+    max_validation_steps: int
     data_config: DataConfig
     optimizer_config: OptimizerConfig
     efficiency_config: Optional[EfficiencyConfig] = None

llama_stack/providers/inline/post_training/torchtune/common/utils.py

@@ -52,9 +52,9 @@ class ModelConfig(BaseModel):
 
 class DatasetSchema(BaseModel):
     alpaca: List[Dict[str, ParamType]]
-    instruct: Dict[str, ParamType]
-    chat_sharegpt: Dict[str, ParamType]
-    chat_openai: Dict[str, ParamType]
+    instruct: List[Dict[str, ParamType]]
+    chat_sharegpt: List[Dict[str, ParamType]]
+    chat_openai: List[Dict[str, ParamType]]
 
 
 MODEL_CONFIGS: Dict[str, ModelConfig] = {
@@ -96,16 +96,22 @@ EXPECTED_DATASET_SCHEMA = DatasetSchema(
             ColumnName.output.value: StringType(),
         },
     ],
-    instruct={
-        ColumnName.input.value: StringType(),
-        ColumnName.output.value: StringType(),
-    },
-    chat_sharegpt={
-        ColumnName.conversations.value: StringType(),
-    },
-    chat_openai={
-        ColumnName.messages.value: StringType(),
-    },
+    instruct=[
+        {
+            ColumnName.input.value: StringType(),
+            ColumnName.output.value: StringType(),
+        }
+    ],
+    chat_sharegpt=[
+        {
+            ColumnName.conversations.value: StringType(),
+        }
+    ],
+    chat_openai=[
+        {
+            ColumnName.messages.value: StringType(),
+        }
+    ],
 )
 
 BuildLoraModelCallable = Callable[..., torch.nn.Module]

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

@@ -126,8 +126,7 @@ class LoraFinetuningSingleDevice:
 
         self._output_dir = str(DEFAULT_CHECKPOINT_DIR)
 
-        # self.seed = training.set_seed(seed=config.torch_seed)
-        self.seed = 42
+        self.seed = training.set_seed(seed=config.torch_seed)
         self.epochs_run = 0
         self.total_epochs = training_config.n_epochs
         self._data_format = training_config.data_config.data_format
@@ -140,6 +139,7 @@ class LoraFinetuningSingleDevice:
         self.global_step = 0
 
         self._gradient_accumulation_steps = training_config.gradient_accumulation_steps
+        self.max_validation_steps = training_config.max_validation_steps
 
         self._clip_grad_norm = 1.0
         self._enable_activation_checkpointing = (
@@ -366,7 +366,6 @@ class LoraFinetuningSingleDevice:
             column_map=self._column_map,
         )
         data_transform = await utils.get_data_transform(self._data_format)
-        print("data_transform", data_transform.__name__)
         ds = SFTDataset(
             rows,
             message_transform=data_transform(
@@ -450,10 +449,6 @@ class LoraFinetuningSingleDevice:
 
     async def _loss_step(self, batch: Dict[str, torch.Tensor]) -> torch.Tensor:
         # Shape [b, s], needed for the loss not the model
-        # print("tokens", batch["tokens"])
-        torch.save(batch["tokens"], "/home/markchen1015/new_alpaca_tokens.pth")
-        # print("labels", batch["labels"])
-        torch.save(batch["labels"], "/home/markchen1015/new_alpaca_labels.pth")
         labels = batch.pop("labels")
         # run model
         with self.activations_handling_ctx:
@@ -595,7 +590,7 @@ class LoraFinetuningSingleDevice:
         log.info("Starting validation...")
         pbar = tqdm(total=len(self._validation_dataloader))
         for idx, batch in enumerate(self._validation_dataloader):
-            if idx == 10:
+            if idx == self.max_validation_steps:
                 break
             torchtune_utils.batch_to_device(batch, self._device)