Add from_step

src/together/cli/api/finetune.py

@@ -11,7 +11,7 @@
 from tabulate import tabulate
 
 from together import Together
-from together.cli.api.utils import BOOL_WITH_AUTO, INT_WITH_MAX
+from together.cli.api.utils import BOOL_WITH_AUTO, INT_WITH_MAX, FROM_STEP_TYPE
 from together.utils import (
     finetune_price_to_dollars,
     log_warn,
@@ -126,6 +126,12 @@ def fine_tuning(ctx: click.Context) -> None:
     help="Whether to mask the user messages in conversational data or prompts in instruction data. "
     "`auto` will automatically determine whether to mask the inputs based on the data format.",
 )
+@click.option(
+    "--from-step",
+    type=FROM_STEP_TYPE,
+    default="final",
+    help="From which checkpoint start a fine-tuning job"
+)
 def create(
     ctx: click.Context,
     training_file: str,
@@ -152,6 +158,7 @@ def create(
     wandb_name: str,
     confirm: bool,
     train_on_inputs: bool | Literal["auto"],
+    from_step: int | Literal["final"],
 ) -> None:
     """Start fine-tuning"""
     client: Together = ctx.obj
@@ -180,6 +187,7 @@ def create(
         wandb_project_name=wandb_project_name,
         wandb_name=wandb_name,
         train_on_inputs=train_on_inputs,
+        from_step=from_step,
     )
 
     model_limits: FinetuneTrainingLimits = client.fine_tuning.get_model_limits(

src/together/cli/api/utils.py

@@ -47,5 +47,22 @@ def convert(
             )
 
 
+class FromStepParamType(click.ParamType):
+    name = "from_step"
+
+    def convert(self, value: str, param: click.Parameter | None, ctx: click.Context | None) -> int | Literal["final"] | None:
+        if value == "final":
+            return "final"
+        try:
+            return int(value)
+        except ValueError:
+            self.fail(
+                _("{value!r} is not a valid {type}.").format(
+                    value=value, type=self.name
+                ),
+            )
+
+
 INT_WITH_MAX = AutoIntParamType()
 BOOL_WITH_AUTO = BooleanWithAutoParamType()
+FROM_STEP_TYPE = FromStepParamType()

src/together/resources/finetune.py

@@ -52,6 +52,7 @@ def createFinetuneRequest(
     wandb_project_name: str | None = None,
     wandb_name: str | None = None,
     train_on_inputs: bool | Literal["auto"] = "auto",
+    from_step: int | Literal["final"] = "final"
 ) -> FinetuneRequest:
     if batch_size == "max":
         log_warn_once(
@@ -100,6 +101,9 @@ def createFinetuneRequest(
     if weight_decay is not None and (weight_decay < 0):
         raise ValueError("Weight decay should be non-negative")
 
+    if from_step == "final":
+        from_step = -1
+
     lrScheduler = FinetuneLRScheduler(
         lr_scheduler_type="linear",
         lr_scheduler_args=FinetuneLinearLRSchedulerArgs(min_lr_ratio=min_lr_ratio),
@@ -125,6 +129,7 @@ def createFinetuneRequest(
         wandb_project_name=wandb_project_name,
         wandb_name=wandb_name,
         train_on_inputs=train_on_inputs,
+        from_step=from_step,
     )
 
     return finetune_request
@@ -162,6 +167,7 @@ def create(
         verbose: bool = False,
         model_limits: FinetuneTrainingLimits | None = None,
         train_on_inputs: bool | Literal["auto"] = "auto",
+        from_step: int | Literal["final"] = "final",
     ) -> FinetuneResponse:
         """
         Method to initiate a fine-tuning job
@@ -207,6 +213,7 @@ def create(
                 For datasets with the "messages" field (conversational format) or "prompt" and "completion" fields
                 (Instruction format), inputs will be masked.
                 Defaults to "auto".
+            from_step (int or "final"): From which checkpoint start a fine-tuning job
 
         Returns:
             FinetuneResponse: Object containing information about fine-tuning job.
@@ -244,6 +251,7 @@ def create(
             wandb_project_name=wandb_project_name,
             wandb_name=wandb_name,
             train_on_inputs=train_on_inputs,
+            from_step=from_step,
         )
 
         if verbose:

src/together/types/finetune.py

@@ -178,6 +178,8 @@ class FinetuneRequest(BaseModel):
     training_type: FullTrainingType | LoRATrainingType | None = None
     # train on inputs
     train_on_inputs: StrictBool | Literal["auto"] = "auto"
+    # from step
+    from_step: int | None = -1
 
 
 class FinetuneResponse(BaseModel):
@@ -256,6 +258,7 @@ class FinetuneResponse(BaseModel):
     training_file_num_lines: int | None = Field(None, alias="TrainingFileNumLines")
     training_file_size: int | None = Field(None, alias="TrainingFileSize")
     train_on_inputs: StrictBool | Literal["auto"] | None = "auto"
+    from_step: int | None = "-1"
 
     @field_validator("training_type")
     @classmethod