Add from_hf_model argument

src/together/cli/api/finetune.py

@@ -200,6 +200,19 @@ def fine_tuning(ctx: click.Context) -> None:
     "The format: {$JOB_ID/$OUTPUT_MODEL_NAME}:{$STEP}. "
     "The step value is optional, without it the final checkpoint will be used.",
 )
+@click.option(
+    "--from-hf-model",
+    type=str,
+    default=None,
+    help="The Hugging Face repo to start training from. "
+    "Should be paired with the base model, specified in `model` argument.",
+)
+@click.option(
+    "--hf-model-revision",
+    type=str,
+    default=None,
+    help="The revision of the Hugging Face model to continue training from.",
+)
 @click.option(
     "--hf-api-token",
     type=str,
@@ -246,6 +259,8 @@ def create(
     rpo_alpha: float | None,
     simpo_gamma: float | None,
     from_checkpoint: str,
+    from_hf_model: str,
+    hf_model_revision: str,
     hf_api_token: str | None,
     hf_output_repo_name: str | None,
 ) -> None:
@@ -284,6 +299,8 @@ def create(
         rpo_alpha=rpo_alpha,
         simpo_gamma=simpo_gamma,
         from_checkpoint=from_checkpoint,
+        from_hf_model=from_hf_model,
+        hf_model_revision=hf_model_revision,
         hf_api_token=hf_api_token,
         hf_output_repo_name=hf_output_repo_name,
     )

src/together/resources/finetune.py

@@ -76,6 +76,8 @@ def create_finetune_request(
     rpo_alpha: float | None = None,
     simpo_gamma: float | None = None,
     from_checkpoint: str | None = None,
+    from_hf_model: str | None = None,
+    hf_model_revision: str | None = None,
     hf_api_token: str | None = None,
     hf_output_repo_name: str | None = None,
 ) -> FinetuneRequest:
@@ -87,6 +89,16 @@ def create_finetune_request(
     if model is None and from_checkpoint is None:
         raise ValueError("You must specify either a model or a checkpoint")
 
+    if from_checkpoint is not None and from_hf_model is not None:
+        raise ValueError(
+            "You must specify either a Hugging Face model or a checkpoint to start a job from, not both"
+        )
+
+    if from_hf_model is not None and model is None:
+        raise ValueError(
+            "You must specify base model to run a fine-tuning job with a Hugging Face model"
+        )
+
     model_or_checkpoint = model or from_checkpoint
 
     if warmup_ratio is None:
@@ -251,6 +263,8 @@ def create_finetune_request(
         wandb_name=wandb_name,
         training_method=training_method_cls,
         from_checkpoint=from_checkpoint,
+        from_hf_model=from_hf_model,
+        hf_model_revision=hf_model_revision,
         hf_api_token=hf_api_token,
         hf_output_repo_name=hf_output_repo_name,
     )
@@ -332,6 +346,8 @@ def create(
         rpo_alpha: float | None = None,
         simpo_gamma: float | None = None,
         from_checkpoint: str | None = None,
+        from_hf_model: str | None = None,
+        hf_model_revision: str | None = None,
         hf_api_token: str | None = None,
         hf_output_repo_name: str | None = None,
     ) -> FinetuneResponse:
@@ -390,6 +406,9 @@ def create(
             from_checkpoint (str, optional): The checkpoint identifier to continue training from a previous fine-tuning job.
                 The format: {$JOB_ID/$OUTPUT_MODEL_NAME}:{$STEP}.
                 The step value is optional, without it the final checkpoint will be used.
+            from_hf_model (str, optional): The Hugging Face repo to start training from.
+                Should be paired with the base model, specified in `model` argument.
+            hf_model_revision (str, optional): The revision of the Hugging Face model to continue training from. Defaults to None.
             hf_api_token (str, optional): API key for the Hugging Face Hub. Defaults to None.
             hf_output_repo_name (str, optional): HF repo to upload the fine-tuned model to. Defaults to None.
 
@@ -445,6 +464,8 @@ def create(
             rpo_alpha=rpo_alpha,
             simpo_gamma=simpo_gamma,
             from_checkpoint=from_checkpoint,
+            from_hf_model=from_hf_model,
+            hf_model_revision=hf_model_revision,
             hf_api_token=hf_api_token,
             hf_output_repo_name=hf_output_repo_name,
         )
@@ -759,6 +780,8 @@ async def create(
         rpo_alpha: float | None = None,
         simpo_gamma: float | None = None,
         from_checkpoint: str | None = None,
+        from_hf_model: str | None = None,
+        hf_model_revision: str | None = None,
         hf_api_token: str | None = None,
         hf_output_repo_name: str | None = None,
     ) -> FinetuneResponse:
@@ -817,6 +840,9 @@ async def create(
             from_checkpoint (str, optional): The checkpoint identifier to continue training from a previous fine-tuning job.
                 The format: {$JOB_ID/$OUTPUT_MODEL_NAME}:{$STEP}.
                 The step value is optional, without it the final checkpoint will be used.
+            from_hf_model (str, optional): The Hugging Face repo to start training from.
+                Should be paired with the base model, specified in `model` argument.
+            hf_model_revision (str, optional): The revision of the Hugging Face model to continue training from. Defaults to None.
             hf_api_token (str, optional): API key for the Huggging Face Hub. Defaults to None.
             hf_output_repo_name (str, optional): HF repo to upload the fine-tuned model to. Defaults to None.
 
@@ -872,6 +898,8 @@ async def create(
             rpo_alpha=rpo_alpha,
             simpo_gamma=simpo_gamma,
             from_checkpoint=from_checkpoint,
+            from_hf_model=from_hf_model,
+            hf_model_revision=hf_model_revision,
             hf_api_token=hf_api_token,
             hf_output_repo_name=hf_output_repo_name,
         )