together 1.5.25__py3-none-any.whl → 1.5.27__py3-none-any.whl

together/cli/api/evaluation.py

@@ -24,10 +24,28 @@ def evaluation(ctx: click.Context) -> None:
     help="Type of evaluation to create.",
 )
 @click.option(
-    "--judge-model-name",
+    "--judge-model",
     type=str,
     required=True,
-    help="Name of the judge model to use for evaluation.",
+    help="Name or URL of the judge model to use for evaluation.",
+)
+@click.option(
+    "--judge-model-source",
+    type=click.Choice(["serverless", "dedicated", "external"]),
+    required=True,
+    help="Source of the judge model.",
+)
+@click.option(
+    "--judge-external-api-token",
+    type=str,
+    required=False,
+    help="Optional external API token for the judge model.",
+)
+@click.option(
+    "--judge-external-base-url",
+    type=str,
+    required=False,
+    help="Optional external base URLs for the judge model.",
 )
 @click.option(
     "--judge-system-template",
@@ -48,10 +66,25 @@ def evaluation(ctx: click.Context) -> None:
     "Can not be used when model-a-name and other model config parameters are specified",
 )
 @click.option(
-    "--model-to-evaluate-name",
+    "--model-to-evaluate",
     type=str,
     help="Model name when using the detailed config",
 )
+@click.option(
+    "--model-to-evaluate-source",
+    type=click.Choice(["serverless", "dedicated", "external"]),
+    help="Source of the model to evaluate.",
+)
+@click.option(
+    "--model-to-evaluate-external-api-token",
+    type=str,
+    help="Optional external API token for the model to evaluate.",
+)
+@click.option(
+    "--model-to-evaluate-external-base-url",
+    type=str,
+    help="Optional external base URL for the model to evaluate.",
+)
 @click.option(
     "--model-to-evaluate-max-tokens",
     type=int,
@@ -104,9 +137,24 @@ def evaluation(ctx: click.Context) -> None:
         Can not be used when model-a-name and other model config parameters are specified",
 )
 @click.option(
-    "--model-a-name",
+    "--model-a",
     type=str,
-    help="Model name for model A when using detailed config.",
+    help="Model name or URL for model A when using detailed config.",
+)
+@click.option(
+    "--model-a-source",
+    type=click.Choice(["serverless", "dedicated", "external"]),
+    help="Source of model A.",
+)
+@click.option(
+    "--model-a-external-api-token",
+    type=str,
+    help="Optional external API token for model A.",
+)
+@click.option(
+    "--model-a-external-base-url",
+    type=str,
+    help="Optional external base URL for model A.",
 )
 @click.option(
     "--model-a-max-tokens",
@@ -135,9 +183,24 @@ def evaluation(ctx: click.Context) -> None:
           Can not be used when model-b-name and other model config parameters are specified",
 )
 @click.option(
-    "--model-b-name",
+    "--model-b",
     type=str,
-    help="Model name for model B when using detailed config.",
+    help="Model name or URL for model B when using detailed config.",
+)
+@click.option(
+    "--model-b-source",
+    type=click.Choice(["serverless", "dedicated", "external"]),
+    help="Source of model B.",
+)
+@click.option(
+    "--model-b-external-api-token",
+    type=str,
+    help="Optional external API token for model B.",
+)
+@click.option(
+    "--model-b-external-base-url",
+    type=str,
+    help="Optional external base URL for model B.",
 )
 @click.option(
     "--model-b-max-tokens",
@@ -162,11 +225,17 @@ def evaluation(ctx: click.Context) -> None:
 def create(
     ctx: click.Context,
     type: str,
-    judge_model_name: str,
+    judge_model: str,
+    judge_model_source: str,
     judge_system_template: str,
+    judge_external_api_token: Optional[str],
+    judge_external_base_url: Optional[str],
     input_data_file_path: str,
     model_field: Optional[str],
-    model_to_evaluate_name: Optional[str],
+    model_to_evaluate: Optional[str],
+    model_to_evaluate_source: Optional[str],
+    model_to_evaluate_external_api_token: Optional[str],
+    model_to_evaluate_external_base_url: Optional[str],
     model_to_evaluate_max_tokens: Optional[int],
     model_to_evaluate_temperature: Optional[float],
     model_to_evaluate_system_template: Optional[str],
@@ -177,13 +246,19 @@ def create(
     max_score: Optional[float],
     pass_threshold: Optional[float],
     model_a_field: Optional[str],
-    model_a_name: Optional[str],
+    model_a: Optional[str],
+    model_a_source: Optional[str],
+    model_a_external_api_token: Optional[str],
+    model_a_external_base_url: Optional[str],
     model_a_max_tokens: Optional[int],
     model_a_temperature: Optional[float],
     model_a_system_template: Optional[str],
     model_a_input_template: Optional[str],
     model_b_field: Optional[str],
-    model_b_name: Optional[str],
+    model_b: Optional[str],
+    model_b_source: Optional[str],
+    model_b_external_api_token: Optional[str],
+    model_b_external_base_url: Optional[str],
     model_b_max_tokens: Optional[int],
     model_b_temperature: Optional[float],
     model_b_system_template: Optional[str],
@@ -203,7 +278,8 @@ def create(
     # Check if any config parameters are provided
     config_params_provided = any(
         [
-            model_to_evaluate_name,
+            model_to_evaluate,
+            model_to_evaluate_source,
             model_to_evaluate_max_tokens,
             model_to_evaluate_temperature,
             model_to_evaluate_system_template,
@@ -223,17 +299,27 @@ def create(
     elif config_params_provided:
         # Config mode: config parameters are provided
         model_to_evaluate_final = {
-            "model_name": model_to_evaluate_name,
+            "model": model_to_evaluate,
+            "model_source": model_to_evaluate_source,
             "max_tokens": model_to_evaluate_max_tokens,
             "temperature": model_to_evaluate_temperature,
             "system_template": model_to_evaluate_system_template,
             "input_template": model_to_evaluate_input_template,
         }
+        if model_to_evaluate_external_api_token:
+            model_to_evaluate_final["external_api_token"] = (
+                model_to_evaluate_external_api_token
+            )
+        if model_to_evaluate_external_base_url:
+            model_to_evaluate_final["external_base_url"] = (
+                model_to_evaluate_external_base_url
+            )
 
     # Build model-a configuration
     model_a_final: Union[Dict[str, Any], None, str] = None
     model_a_config_params = [
-        model_a_name,
+        model_a,
+        model_a_source,
         model_a_max_tokens,
         model_a_temperature,
         model_a_system_template,
@@ -252,17 +338,23 @@ def create(
     elif any(model_a_config_params):
         # Config mode: config parameters are provided
         model_a_final = {
-            "model_name": model_a_name,
+            "model": model_a,
+            "model_source": model_a_source,
             "max_tokens": model_a_max_tokens,
             "temperature": model_a_temperature,
             "system_template": model_a_system_template,
             "input_template": model_a_input_template,
         }
+        if model_a_external_api_token:
+            model_a_final["external_api_token"] = model_a_external_api_token
+        if model_a_external_base_url:
+            model_a_final["external_base_url"] = model_a_external_base_url
 
     # Build model-b configuration
     model_b_final: Union[Dict[str, Any], None, str] = None
     model_b_config_params = [
-        model_b_name,
+        model_b,
+        model_b_source,
         model_b_max_tokens,
         model_b_temperature,
         model_b_system_template,
@@ -281,18 +373,26 @@ def create(
     elif any(model_b_config_params):
         # Config mode: config parameters are provided
         model_b_final = {
-            "model_name": model_b_name,
+            "model": model_b,
+            "model_source": model_b_source,
             "max_tokens": model_b_max_tokens,
             "temperature": model_b_temperature,
             "system_template": model_b_system_template,
             "input_template": model_b_input_template,
         }
+        if model_b_external_api_token:
+            model_b_final["external_api_token"] = model_b_external_api_token
+        if model_b_external_base_url:
+            model_b_final["external_base_url"] = model_b_external_base_url
 
     try:
         response = client.evaluation.create(
             type=type,
-            judge_model_name=judge_model_name,
+            judge_model=judge_model,
+            judge_model_source=judge_model_source,
             judge_system_template=judge_system_template,
+            judge_external_api_token=judge_external_api_token,
+            judge_external_base_url=judge_external_base_url,
             input_data_file_path=input_data_file_path,
             model_to_evaluate=model_to_evaluate_final,
             labels=labels_list,

together/cli/api/finetune.py

@@ -543,3 +543,30 @@ def download(
     )
 
     click.echo(json.dumps(response.model_dump(exclude_none=True), indent=4))
+
+
+@fine_tuning.command()
+@click.pass_context
+@click.argument("fine_tune_id", type=str, required=True)
+@click.option("--force", is_flag=True, help="Force deletion without confirmation")
+@click.option(
+    "--quiet", is_flag=True, help="Do not prompt for confirmation before deleting job"
+)
+def delete(
+    ctx: click.Context, fine_tune_id: str, force: bool = False, quiet: bool = False
+) -> None:
+    """Delete fine-tuning job"""
+    client: Together = ctx.obj
+
+    if not quiet:
+        confirm_response = input(
+            f"Are you sure you want to delete fine-tuning job {fine_tune_id}? "
+            "This action cannot be undone. [y/N] "
+        )
+        if confirm_response.lower() != "y":
+            click.echo("Deletion cancelled")
+            return
+
+    response = client.fine_tuning.delete(fine_tune_id, force=force)
+
+    click.echo(json.dumps(response.model_dump(exclude_none=True), indent=4))

together/cli/api/models.py

@@ -4,7 +4,7 @@ import click
 from tabulate import tabulate
 
 from together import Together
-from together.types.models import ModelObject
+from together.types.models import ModelObject, ModelUploadResponse
 
 
 @click.group()
@@ -53,3 +53,81 @@ def list(ctx: click.Context, type: str | None, json: bool) -> None:
         click.echo(json_lib.dumps(display_list, indent=2))
     else:
         click.echo(tabulate(display_list, headers="keys", tablefmt="plain"))
+
+
+@models.command()
+@click.option(
+    "--model-name",
+    required=True,
+    help="The name to give to your uploaded model",
+)
+@click.option(
+    "--model-source",
+    required=True,
+    help="The source location of the model (Hugging Face repo or S3 path)",
+)
+@click.option(
+    "--model-type",
+    type=click.Choice(["model", "adapter"]),
+    default="model",
+    help="Whether the model is a full model or an adapter",
+)
+@click.option(
+    "--hf-token",
+    help="Hugging Face token (if uploading from Hugging Face)",
+)
+@click.option(
+    "--description",
+    help="A description of your model",
+)
+@click.option(
+    "--base-model",
+    help="The base model to use for an adapter if setting it to run against a serverless pool. Only used for model_type 'adapter'.",
+)
+@click.option(
+    "--lora-model",
+    help="The lora pool to use for an adapter if setting it to run against, say, a dedicated pool. Only used for model_type 'adapter'.",
+)
+@click.option(
+    "--json",
+    is_flag=True,
+    help="Output in JSON format",
+)
+@click.pass_context
+def upload(
+    ctx: click.Context,
+    model_name: str,
+    model_source: str,
+    model_type: str,
+    hf_token: str | None,
+    description: str | None,
+    base_model: str | None,
+    lora_model: str | None,
+    json: bool,
+) -> None:
+    """Upload a custom model or adapter from Hugging Face or S3"""
+    client: Together = ctx.obj
+
+    response: ModelUploadResponse = client.models.upload(
+        model_name=model_name,
+        model_source=model_source,
+        model_type=model_type,
+        hf_token=hf_token,
+        description=description,
+        base_model=base_model,
+        lora_model=lora_model,
+    )
+
+    if json:
+        click.echo(json_lib.dumps(response.model_dump(), indent=2))
+    else:
+        click.echo(f"Model upload job created successfully!")
+        if response.job_id:
+            click.echo(f"Job ID: {response.job_id}")
+        if response.model_name:
+            click.echo(f"Model Name: {response.model_name}")
+        if response.model_id:
+            click.echo(f"Model ID: {response.model_id}")
+        if response.model_source:
+            click.echo(f"Model Source: {response.model_source}")
+        click.echo(f"Message: {response.message}")

together/client.py

@@ -26,6 +26,7 @@ class Together:
     batches: resources.Batches
     code_interpreter: CodeInterpreter
     evaluation: resources.Evaluation
+    videos: resources.Videos
 
     # client options
     client: TogetherClient
@@ -94,6 +95,7 @@ class Together:
         self.code_interpreter = CodeInterpreter(self.client)
         self.batches = resources.Batches(self.client)
         self.evaluation = resources.Evaluation(self.client)
+        self.videos = resources.Videos(self.client)
 
 
 class AsyncTogether:
@@ -109,6 +111,7 @@ class AsyncTogether:
     code_interpreter: CodeInterpreter
     batches: resources.AsyncBatches
     evaluation: resources.AsyncEvaluation
+    videos: resources.AsyncVideos
     # client options
     client: TogetherClient
 
@@ -175,6 +178,7 @@ class AsyncTogether:
         self.code_interpreter = CodeInterpreter(self.client)
         self.batches = resources.AsyncBatches(self.client)
         self.evaluation = resources.AsyncEvaluation(self.client)
+        self.videos = resources.AsyncVideos(self.client)
 
 
 Client = Together

together/constants.py

@@ -15,6 +15,20 @@ BASE_URL = "https://api.together.xyz/v1"
 DOWNLOAD_BLOCK_SIZE = 10 * 1024 * 1024  # 10 MB
 DISABLE_TQDM = False
 
+# Upload defaults
+MAX_CONCURRENT_PARTS = 4  # Maximum concurrent parts for multipart upload
+
+# Multipart upload constants
+MIN_PART_SIZE_MB = 5  # Minimum part size (S3 requirement)
+TARGET_PART_SIZE_MB = 100  # Target part size for optimal performance
+MAX_MULTIPART_PARTS = 250  # Maximum parts per upload (S3 limit)
+MULTIPART_UPLOAD_TIMEOUT = 300  # Timeout in seconds for uploading each part
+MULTIPART_THRESHOLD_GB = 5.0  # threshold for switching to multipart upload
+
+# maximum number of GB sized files we support finetuning for
+MAX_FILE_SIZE_GB = 25.0
+
+
 # Messages
 MISSING_API_KEY_MESSAGE = """TOGETHER_API_KEY not found.
 Please set it as an environment variable or set it as together.api_key
@@ -26,8 +40,6 @@ MIN_SAMPLES = 1
 # the number of bytes in a gigabyte, used to convert bytes to GB for readable comparison
 NUM_BYTES_IN_GB = 2**30
 
-# maximum number of GB sized files we support finetuning for
-MAX_FILE_SIZE_GB = 4.9
 
 # expected columns for Parquet files
 PARQUET_EXPECTED_COLUMNS = ["input_ids", "attention_mask", "labels"]