together 2.0.0a9__py3-none-any.whl → 2.0.0a11__py3-none-any.whl

together/resources/chat/completions.py

@@ -136,6 +136,14 @@ class CompletionsResource(SyncAPIResource):
 
           response_format: An object specifying the format that the model must output.
 
+              Setting to `{ "type": "json_schema", "json_schema": {...} }` enables Structured
+              Outputs which ensures the model will match your supplied JSON schema. Learn more
+              in the [Structured Outputs guide](https://docs.together.ai/docs/json-mode).
+
+              Setting to `{ "type": "json_object" }` enables the older JSON mode, which
+              ensures the message the model generates is valid JSON. Using `json_schema` is
+              preferred for models that support it.
+
           safety_model: The name of the moderation model used to validate tokens. Choose from the
               available moderation models found
               [here](https://docs.together.ai/docs/inference-models#moderation-models).
@@ -277,6 +285,14 @@ class CompletionsResource(SyncAPIResource):
 
           response_format: An object specifying the format that the model must output.
 
+              Setting to `{ "type": "json_schema", "json_schema": {...} }` enables Structured
+              Outputs which ensures the model will match your supplied JSON schema. Learn more
+              in the [Structured Outputs guide](https://docs.together.ai/docs/json-mode).
+
+              Setting to `{ "type": "json_object" }` enables the older JSON mode, which
+              ensures the message the model generates is valid JSON. Using `json_schema` is
+              preferred for models that support it.
+
           safety_model: The name of the moderation model used to validate tokens. Choose from the
               available moderation models found
               [here](https://docs.together.ai/docs/inference-models#moderation-models).
@@ -414,6 +430,14 @@ class CompletionsResource(SyncAPIResource):
 
           response_format: An object specifying the format that the model must output.
 
+              Setting to `{ "type": "json_schema", "json_schema": {...} }` enables Structured
+              Outputs which ensures the model will match your supplied JSON schema. Learn more
+              in the [Structured Outputs guide](https://docs.together.ai/docs/json-mode).
+
+              Setting to `{ "type": "json_object" }` enables the older JSON mode, which
+              ensures the message the model generates is valid JSON. Using `json_schema` is
+              preferred for models that support it.
+
           safety_model: The name of the moderation model used to validate tokens. Choose from the
               available moderation models found
               [here](https://docs.together.ai/docs/inference-models#moderation-models).
@@ -653,6 +677,14 @@ class AsyncCompletionsResource(AsyncAPIResource):
 
           response_format: An object specifying the format that the model must output.
 
+              Setting to `{ "type": "json_schema", "json_schema": {...} }` enables Structured
+              Outputs which ensures the model will match your supplied JSON schema. Learn more
+              in the [Structured Outputs guide](https://docs.together.ai/docs/json-mode).
+
+              Setting to `{ "type": "json_object" }` enables the older JSON mode, which
+              ensures the message the model generates is valid JSON. Using `json_schema` is
+              preferred for models that support it.
+
           safety_model: The name of the moderation model used to validate tokens. Choose from the
               available moderation models found
               [here](https://docs.together.ai/docs/inference-models#moderation-models).
@@ -794,6 +826,14 @@ class AsyncCompletionsResource(AsyncAPIResource):
 
           response_format: An object specifying the format that the model must output.
 
+              Setting to `{ "type": "json_schema", "json_schema": {...} }` enables Structured
+              Outputs which ensures the model will match your supplied JSON schema. Learn more
+              in the [Structured Outputs guide](https://docs.together.ai/docs/json-mode).
+
+              Setting to `{ "type": "json_object" }` enables the older JSON mode, which
+              ensures the message the model generates is valid JSON. Using `json_schema` is
+              preferred for models that support it.
+
           safety_model: The name of the moderation model used to validate tokens. Choose from the
               available moderation models found
               [here](https://docs.together.ai/docs/inference-models#moderation-models).
@@ -931,6 +971,14 @@ class AsyncCompletionsResource(AsyncAPIResource):
 
           response_format: An object specifying the format that the model must output.
 
+              Setting to `{ "type": "json_schema", "json_schema": {...} }` enables Structured
+              Outputs which ensures the model will match your supplied JSON schema. Learn more
+              in the [Structured Outputs guide](https://docs.together.ai/docs/json-mode).
+
+              Setting to `{ "type": "json_object" }` enables the older JSON mode, which
+              ensures the message the model generates is valid JSON. Using `json_schema` is
+              preferred for models that support it.
+
           safety_model: The name of the moderation model used to validate tokens. Choose from the
               available moderation models found
               [here](https://docs.together.ai/docs/inference-models#moderation-models).

together/resources/fine_tuning.py

@@ -7,7 +7,7 @@ from typing_extensions import Literal
 import httpx
 from rich import print as rprint
 
-from ..types import fine_tuning_delete_params, fine_tuning_content_params
+from ..types import fine_tuning_delete_params, fine_tuning_content_params, fine_tuning_estimate_price_params
 from .._types import Body, Omit, Query, Headers, NotGiven, omit, not_given
 from .._utils import maybe_transform, async_maybe_transform
 from .._compat import cached_property
@@ -27,17 +27,31 @@ from .._response import (
     async_to_custom_streamed_response_wrapper,
 )
 from .._base_client import make_request_options
-from ..lib.types.fine_tuning import FinetuneResponse as FinetuneResponseLib, FinetuneTrainingLimits
+from ..lib.types.fine_tuning import (
+    FinetuneResponse as FinetuneResponseLib,
+    FinetuneTrainingLimits,
+)
 from ..types.finetune_response import FinetuneResponse
-from ..lib.resources.fine_tuning import get_model_limits, async_get_model_limits, create_finetune_request
+from ..lib.resources.fine_tuning import (
+    get_model_limits,
+    async_get_model_limits,
+    create_finetune_request,
+)
 from ..types.fine_tuning_list_response import FineTuningListResponse
 from ..types.fine_tuning_cancel_response import FineTuningCancelResponse
 from ..types.fine_tuning_delete_response import FineTuningDeleteResponse
 from ..types.fine_tuning_list_events_response import FineTuningListEventsResponse
+from ..types.fine_tuning_estimate_price_response import FineTuningEstimatePriceResponse
 from ..types.fine_tuning_list_checkpoints_response import FineTuningListCheckpointsResponse
 
 __all__ = ["FineTuningResource", "AsyncFineTuningResource"]
 
+_WARNING_MESSAGE_INSUFFICIENT_FUNDS = (
+    "The estimated price of the fine-tuning job is {} which is significantly "
+    "greater than your current credit limit and balance combined. "
+    "It will likely get cancelled due to insufficient funds. "
+    "Proceed at your own risk."
+)
 
 class FineTuningResource(SyncAPIResource):
     @cached_property
@@ -179,7 +193,7 @@ class FineTuningResource(SyncAPIResource):
                 pass
             model_limits = get_model_limits(self._client, str(model_name))
 
-        finetune_request = create_finetune_request(
+        finetune_request, training_type_cls, training_method_cls = create_finetune_request(
             model_limits=model_limits,
             training_file=training_file,
             model=model,
@@ -218,11 +232,32 @@ class FineTuningResource(SyncAPIResource):
             hf_output_repo_name=hf_output_repo_name,
         )
 
+
+        price_estimation_result = self.estimate_price(
+            training_file=training_file,
+            from_checkpoint=from_checkpoint or Omit(),
+            validation_file=validation_file or Omit(),
+            model=model or "",
+            n_epochs=finetune_request.n_epochs,
+            n_evals=finetune_request.n_evals or 0,
+            training_type=training_type_cls,
+            training_method=training_method_cls,
+        )
+
+
         if verbose:
             rprint(
                 "Submitting a fine-tuning job with the following parameters:",
                 finetune_request,
             )
+            if not price_estimation_result.allowed_to_proceed:
+                rprint(
+                    "[red]"
+                    + _WARNING_MESSAGE_INSUFFICIENT_FUNDS.format(
+                        price_estimation_result.estimated_total_price # pyright: ignore[reportPossiblyUnboundVariable]
+                    )
+                    + "[/red]",
+                )
         parameter_payload = finetune_request.model_dump(exclude_none=True)
 
         return self._client.post(
@@ -413,6 +448,76 @@ class FineTuningResource(SyncAPIResource):
             cast_to=BinaryAPIResponse,
         )
 
+    def estimate_price(
+        self,
+        *,
+        training_file: str,
+        from_checkpoint: str | Omit = omit,
+        model: str | Omit = omit,
+        n_epochs: int | Omit = omit,
+        n_evals: int | Omit = omit,
+        training_method: fine_tuning_estimate_price_params.TrainingMethod | Omit = omit,
+        training_type: fine_tuning_estimate_price_params.TrainingType | Omit = omit,
+        validation_file: str | Omit = omit,
+        # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
+        # The extra values given here take precedence over values defined on the client or passed to this method.
+        extra_headers: Headers | None = None,
+        extra_query: Query | None = None,
+        extra_body: Body | None = None,
+        timeout: float | httpx.Timeout | None | NotGiven = not_given,
+    ) -> FineTuningEstimatePriceResponse:
+        """
+        Estimate the price of a fine-tuning job.
+
+        Args:
+          training_file: File-ID of a training file uploaded to the Together API
+
+          from_checkpoint: The checkpoint identifier to continue training from a previous fine-tuning job.
+              Format is `{$JOB_ID}` or `{$OUTPUT_MODEL_NAME}` or `{$JOB_ID}:{$STEP}` or
+              `{$OUTPUT_MODEL_NAME}:{$STEP}`. The step value is optional; without it, the
+              final checkpoint will be used.
+
+          model: Name of the base model to run fine-tune job on
+
+          n_epochs: Number of complete passes through the training dataset (higher values may
+              improve results but increase cost and risk of overfitting)
+
+          n_evals: Number of evaluations to be run on a given validation set during training
+
+          training_method: The training method to use. 'sft' for Supervised Fine-Tuning or 'dpo' for Direct
+              Preference Optimization.
+
+          validation_file: File-ID of a validation file uploaded to the Together API
+
+          extra_headers: Send extra headers
+
+          extra_query: Add additional query parameters to the request
+
+          extra_body: Add additional JSON properties to the request
+
+          timeout: Override the client-level default timeout for this request, in seconds
+        """
+        return self._post(
+            "/fine-tunes/estimate-price",
+            body=maybe_transform(
+                {
+                    "training_file": training_file,
+                    "from_checkpoint": from_checkpoint,
+                    "model": model,
+                    "n_epochs": n_epochs,
+                    "n_evals": n_evals,
+                    "training_method": training_method,
+                    "training_type": training_type,
+                    "validation_file": validation_file,
+                },
+                fine_tuning_estimate_price_params.FineTuningEstimatePriceParams,
+            ),
+            options=make_request_options(
+                extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
+            ),
+            cast_to=FineTuningEstimatePriceResponse,
+        )
+
     def list_checkpoints(
         self,
         id: str,
@@ -620,7 +725,7 @@ class AsyncFineTuningResource(AsyncAPIResource):
                 pass
             model_limits = await async_get_model_limits(self._client, str(model_name))
 
-        finetune_request = create_finetune_request(
+        finetune_request, training_type_cls, training_method_cls = create_finetune_request(
             model_limits=model_limits,
             training_file=training_file,
             model=model,
@@ -659,11 +764,32 @@ class AsyncFineTuningResource(AsyncAPIResource):
             hf_output_repo_name=hf_output_repo_name,
         )
 
+
+        price_estimation_result = await self.estimate_price(
+            training_file=training_file,
+            from_checkpoint=from_checkpoint or Omit(),
+            validation_file=validation_file or Omit(),
+            model=model or "",
+            n_epochs=finetune_request.n_epochs,
+            n_evals=finetune_request.n_evals or 0,
+            training_type=training_type_cls,
+            training_method=training_method_cls,
+        )
+
+
         if verbose:
             rprint(
                 "Submitting a fine-tuning job with the following parameters:",
                 finetune_request,
             )
+            if not price_estimation_result.allowed_to_proceed:
+                rprint(
+                    "[red]"
+                    + _WARNING_MESSAGE_INSUFFICIENT_FUNDS.format(
+                        price_estimation_result.estimated_total_price # pyright: ignore[reportPossiblyUnboundVariable]
+                    )
+                    + "[/red]",
+                )
         parameter_payload = finetune_request.model_dump(exclude_none=True)
 
         return await self._client.post(
@@ -854,6 +980,76 @@ class AsyncFineTuningResource(AsyncAPIResource):
             cast_to=AsyncBinaryAPIResponse,
         )
 
+    async def estimate_price(
+        self,
+        *,
+        training_file: str,
+        from_checkpoint: str | Omit = omit,
+        model: str | Omit = omit,
+        n_epochs: int | Omit = omit,
+        n_evals: int | Omit = omit,
+        training_method: fine_tuning_estimate_price_params.TrainingMethod | Omit = omit,
+        training_type: fine_tuning_estimate_price_params.TrainingType | Omit = omit,
+        validation_file: str | Omit = omit,
+        # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
+        # The extra values given here take precedence over values defined on the client or passed to this method.
+        extra_headers: Headers | None = None,
+        extra_query: Query | None = None,
+        extra_body: Body | None = None,
+        timeout: float | httpx.Timeout | None | NotGiven = not_given,
+    ) -> FineTuningEstimatePriceResponse:
+        """
+        Estimate the price of a fine-tuning job.
+
+        Args:
+          training_file: File-ID of a training file uploaded to the Together API
+
+          from_checkpoint: The checkpoint identifier to continue training from a previous fine-tuning job.
+              Format is `{$JOB_ID}` or `{$OUTPUT_MODEL_NAME}` or `{$JOB_ID}:{$STEP}` or
+              `{$OUTPUT_MODEL_NAME}:{$STEP}`. The step value is optional; without it, the
+              final checkpoint will be used.
+
+          model: Name of the base model to run fine-tune job on
+
+          n_epochs: Number of complete passes through the training dataset (higher values may
+              improve results but increase cost and risk of overfitting)
+
+          n_evals: Number of evaluations to be run on a given validation set during training
+
+          training_method: The training method to use. 'sft' for Supervised Fine-Tuning or 'dpo' for Direct
+              Preference Optimization.
+
+          validation_file: File-ID of a validation file uploaded to the Together API
+
+          extra_headers: Send extra headers
+
+          extra_query: Add additional query parameters to the request
+
+          extra_body: Add additional JSON properties to the request
+
+          timeout: Override the client-level default timeout for this request, in seconds
+        """
+        return await self._post(
+            "/fine-tunes/estimate-price",
+            body=await async_maybe_transform(
+                {
+                    "training_file": training_file,
+                    "from_checkpoint": from_checkpoint,
+                    "model": model,
+                    "n_epochs": n_epochs,
+                    "n_evals": n_evals,
+                    "training_method": training_method,
+                    "training_type": training_type,
+                    "validation_file": validation_file,
+                },
+                fine_tuning_estimate_price_params.FineTuningEstimatePriceParams,
+            ),
+            options=make_request_options(
+                extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
+            ),
+            cast_to=FineTuningEstimatePriceResponse,
+        )
+
     async def list_checkpoints(
         self,
         id: str,
@@ -941,6 +1137,9 @@ class FineTuningResourceWithRawResponse:
             fine_tuning.content,
             BinaryAPIResponse,
         )
+        self.estimate_price = to_raw_response_wrapper(
+            fine_tuning.estimate_price,
+        )
         self.list_checkpoints = to_raw_response_wrapper(
             fine_tuning.list_checkpoints,
         )
@@ -969,6 +1168,9 @@ class AsyncFineTuningResourceWithRawResponse:
             fine_tuning.content,
             AsyncBinaryAPIResponse,
         )
+        self.estimate_price = async_to_raw_response_wrapper(
+            fine_tuning.estimate_price,
+        )
         self.list_checkpoints = async_to_raw_response_wrapper(
             fine_tuning.list_checkpoints,
         )
@@ -997,6 +1199,9 @@ class FineTuningResourceWithStreamingResponse:
             fine_tuning.content,
             StreamedBinaryAPIResponse,
         )
+        self.estimate_price = to_streamed_response_wrapper(
+            fine_tuning.estimate_price,
+        )
         self.list_checkpoints = to_streamed_response_wrapper(
             fine_tuning.list_checkpoints,
         )
@@ -1025,6 +1230,9 @@ class AsyncFineTuningResourceWithStreamingResponse:
             fine_tuning.content,
             AsyncStreamedBinaryAPIResponse,
         )
+        self.estimate_price = async_to_streamed_response_wrapper(
+            fine_tuning.estimate_price,
+        )
         self.list_checkpoints = async_to_streamed_response_wrapper(
             fine_tuning.list_checkpoints,
         )

together/resources/models.py

@@ -6,7 +6,7 @@ from typing_extensions import Literal
 
 import httpx
 
-from ..types import model_upload_params
+from ..types import model_list_params, model_upload_params
 from .._types import Body, Omit, Query, Headers, NotGiven, omit, not_given
 from .._utils import maybe_transform, async_maybe_transform
 from .._compat import cached_property
@@ -47,6 +47,7 @@ class ModelsResource(SyncAPIResource):
     def list(
         self,
         *,
+        dedicated: bool | Omit = omit,
         # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
         # The extra values given here take precedence over values defined on the client or passed to this method.
         extra_headers: Headers | None = None,
@@ -54,11 +55,28 @@ class ModelsResource(SyncAPIResource):
         extra_body: Body | None = None,
         timeout: float | httpx.Timeout | None | NotGiven = not_given,
     ) -> ModelListResponse:
-        """Lists all of Together's open-source models"""
+        """
+        Lists all of Together's open-source models
+
+        Args:
+          dedicated: Filter models to only return dedicated models
+
+          extra_headers: Send extra headers
+
+          extra_query: Add additional query parameters to the request
+
+          extra_body: Add additional JSON properties to the request
+
+          timeout: Override the client-level default timeout for this request, in seconds
+        """
         return self._get(
             "/models",
             options=make_request_options(
-                extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
+                extra_headers=extra_headers,
+                extra_query=extra_query,
+                extra_body=extra_body,
+                timeout=timeout,
+                query=maybe_transform({"dedicated": dedicated}, model_list_params.ModelListParams),
             ),
             cast_to=ModelListResponse,
         )
@@ -152,6 +170,7 @@ class AsyncModelsResource(AsyncAPIResource):
     async def list(
         self,
         *,
+        dedicated: bool | Omit = omit,
         # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
         # The extra values given here take precedence over values defined on the client or passed to this method.
         extra_headers: Headers | None = None,
@@ -159,11 +178,28 @@ class AsyncModelsResource(AsyncAPIResource):
         extra_body: Body | None = None,
         timeout: float | httpx.Timeout | None | NotGiven = not_given,
     ) -> ModelListResponse:
-        """Lists all of Together's open-source models"""
+        """
+        Lists all of Together's open-source models
+
+        Args:
+          dedicated: Filter models to only return dedicated models
+
+          extra_headers: Send extra headers
+
+          extra_query: Add additional query parameters to the request
+
+          extra_body: Add additional JSON properties to the request
+
+          timeout: Override the client-level default timeout for this request, in seconds
+        """
         return await self._get(
             "/models",
             options=make_request_options(
-                extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
+                extra_headers=extra_headers,
+                extra_query=extra_query,
+                extra_body=extra_body,
+                timeout=timeout,
+                query=await async_maybe_transform({"dedicated": dedicated}, model_list_params.ModelListParams),
             ),
             cast_to=ModelListResponse,
         )

together/types/__init__.py

@@ -26,6 +26,7 @@ from .execute_response import ExecuteResponse as ExecuteResponse
 from .autoscaling_param import AutoscalingParam as AutoscalingParam
 from .finetune_response import FinetuneResponse as FinetuneResponse
 from .job_list_response import JobListResponse as JobListResponse
+from .model_list_params import ModelListParams as ModelListParams
 from .tool_choice_param import ToolChoiceParam as ToolChoiceParam
 from .dedicated_endpoint import DedicatedEndpoint as DedicatedEndpoint
 from .eval_create_params import EvalCreateParams as EvalCreateParams
@@ -62,6 +63,8 @@ from .fine_tuning_delete_response import FineTuningDeleteResponse as FineTuningD
 from .endpoint_list_avzones_response import EndpointListAvzonesResponse as EndpointListAvzonesResponse
 from .code_interpreter_execute_params import CodeInterpreterExecuteParams as CodeInterpreterExecuteParams
 from .fine_tuning_list_events_response import FineTuningListEventsResponse as FineTuningListEventsResponse
+from .fine_tuning_estimate_price_params import FineTuningEstimatePriceParams as FineTuningEstimatePriceParams
+from .fine_tuning_estimate_price_response import FineTuningEstimatePriceResponse as FineTuningEstimatePriceResponse
 from .fine_tuning_list_checkpoints_response import (
     FineTuningListCheckpointsResponse as FineTuningListCheckpointsResponse,
 )

together/types/audio/transcription_create_params.py

@@ -11,8 +11,11 @@ __all__ = ["TranscriptionCreateParams"]
 
 
 class TranscriptionCreateParams(TypedDict, total=False):
-    file: Required[FileTypes]
-    """Audio file to transcribe"""
+    file: Required[Union[FileTypes, str]]
+    """Audio file upload or public HTTP/HTTPS URL.
+
+    Supported formats .wav, .mp3, .m4a, .webm, .flac.
+    """
 
     diarize: bool
     """Whether to enable speaker diarization.

together/types/audio/translation_create_params.py

@@ -11,8 +11,11 @@ __all__ = ["TranslationCreateParams"]
 
 
 class TranslationCreateParams(TypedDict, total=False):
-    file: Required[FileTypes]
-    """Audio file to translate"""
+    file: Required[Union[FileTypes, str]]
+    """Audio file upload or public HTTP/HTTPS URL.
+
+    Supported formats .wav, .mp3, .m4a, .webm, .flac.
+    """
 
     language: str
     """Target output language.

together/types/audio/voice_list_response.py

@@ -14,10 +14,14 @@ class DataVoice(BaseModel):
 
 
 class Data(BaseModel):
+    """Represents a model with its available voices."""
+
     model: str
 
     voices: List[DataVoice]
 
 
 class VoiceListResponse(BaseModel):
+    """Response containing a list of models and their available voices."""
+
     data: List[Data]

together/types/autoscaling.py

@@ -6,6 +6,8 @@ __all__ = ["Autoscaling"]
 
 
 class Autoscaling(BaseModel):
+    """Configuration for automatic scaling of replicas based on demand."""
+
     max_replicas: int
     """The maximum number of replicas to scale up to under load"""
 

together/types/autoscaling_param.py

@@ -8,6 +8,8 @@ __all__ = ["AutoscalingParam"]
 
 
 class AutoscalingParam(TypedDict, total=False):
+    """Configuration for automatic scaling of replicas based on demand."""
+
     max_replicas: Required[int]
     """The maximum number of replicas to scale up to under load"""