together 2.0.0a10__py3-none-any.whl → 2.0.0a12__py3-none-any.whl

together/lib/utils/files.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 import os
 import csv
 import json
-from typing import Any, Dict, List, cast
+from typing import Any, Dict, List, Union, cast
 from pathlib import Path
 from traceback import format_exc
 
@@ -13,8 +13,11 @@ from together.types import FilePurpose
 from together.lib.constants import (
     MIN_SAMPLES,
     DISABLE_TQDM,
+    MAX_IMAGE_BYTES,
     NUM_BYTES_IN_GB,
     MAX_FILE_SIZE_GB,
+    MAX_IMAGES_PER_EXAMPLE,
+    MAX_BASE64_IMAGE_LENGTH,
     PARQUET_EXPECTED_COLUMNS,
     REQUIRED_COLUMNS_MESSAGE,
     JSONL_REQUIRED_COLUMNS_MAP,
@@ -22,6 +25,15 @@ from together.lib.constants import (
     DatasetFormat,
 )
 
+# MessageContent is a string or a list of dicts with 'type': 'text' or 'image_url', and 'text' or 'image_url.url'
+# Example: "Hello" or [
+#   {"type": "text", "text": "Hello"},
+#   {"type": "image_url", "image_url": {
+#     "url": "data:image/jpeg;base64,..."
+#   }}
+# ]
+MessageContent = Union[str, List[Dict[str, Any]]]
+
 
 class InvalidFileFormatError(ValueError):
     """Exception raised for invalid file formats during file checks."""
@@ -103,7 +115,7 @@ def check_file(
     return report_dict
 
 
-def _check_conversation_type(messages: List[Dict[str, str | bool]], idx: int) -> None:
+def _check_conversation_type(messages: List[Dict[str, str | int | MessageContent]], idx: int) -> None:
     """Check that the conversation has correct type.
 
     Args:
@@ -144,12 +156,6 @@ def _check_conversation_type(messages: List[Dict[str, str | bool]], idx: int) ->
                     line_number=idx + 1,
                     error_source="key_value",
                 )
-            if not isinstance(message[column], str):
-                raise InvalidFileFormatError(
-                    message=f"Column `{column}` is not a string on line {idx + 1}. Found {type(message[column])}",
-                    line_number=idx + 1,
-                    error_source="text_field",
-                )
 
 
 def _check_conversation_roles(require_assistant_role: bool, assistant_role_exists: bool, idx: int) -> None:
@@ -172,7 +178,7 @@ def _check_conversation_roles(require_assistant_role: bool, assistant_role_exist
         )
 
 
-def _check_message_weight(message: Dict[str, str | bool], idx: int) -> None:
+def _check_message_weight(message: Dict[str, str | int | MessageContent], idx: int) -> int | None:
     """Check that the message has a weight with the correct type and value.
 
     Args:
@@ -196,9 +202,12 @@ def _check_message_weight(message: Dict[str, str | bool], idx: int) -> None:
                 line_number=idx + 1,
                 error_source="key_value",
             )
+        return weight
+
+    return None
 
 
-def _check_message_role(message: Dict[str, str | bool], previous_role: str | bool | None, idx: int) -> str | bool:
+def _check_message_role(message: Dict[str, str | int | MessageContent], previous_role: str | None, idx: int) -> str:
     """Check that the message has correct roles.
 
     Args:
@@ -212,6 +221,14 @@ def _check_message_role(message: Dict[str, str | bool], previous_role: str | boo
     Raises:
         InvalidFileFormatError: If the message role is invalid.
     """
+    if not isinstance(message["role"], str):
+        raise InvalidFileFormatError(
+            message=f"Invalid role `{message['role']}` in conversation on line {idx + 1}. "
+            f"Role must be a string. Found {type(message['role'])}",
+            line_number=idx + 1,
+            error_source="key_value",
+        )
+
     if message["role"] not in POSSIBLE_ROLES_CONVERSATION:
         raise InvalidFileFormatError(
             message=f"Invalid role `{message['role']}` in conversation on line {idx + 1}. "
@@ -229,7 +246,130 @@ def _check_message_role(message: Dict[str, str | bool], previous_role: str | boo
     return message["role"]
 
 
-def validate_messages(messages: List[Dict[str, str | bool]], idx: int, require_assistant_role: bool = True) -> None:
+def _check_message_content(message_content: str | int | MessageContent, role: str, idx: int) -> tuple[bool, int]:
+    """Check that the message content has the correct type.
+    Message content can be either a) a string or b) an OpenAI-style multimodal list of content items
+    Example:
+        a) "Hello", or
+        b) [
+             {"type": "text", "text": "Hello"},
+             {"type": "image_url", "image_url": {
+                "url": "data:image/jpeg;base64,..."
+             }}
+           ]
+
+    Args:
+        message_content: The message content to check.
+        role: The role of the message.
+        idx: Line number in the file.
+
+    Returns:
+        tuple[bool, int]: A tuple with message is multimodal and the number of images in the message content.
+    """
+    # Text-only message content
+    if isinstance(message_content, str):
+        return False, 0
+
+    # Multimodal message content
+    if isinstance(message_content, list):
+        num_images = 0
+        for item in message_content:
+            if not isinstance(cast(Any, item), dict):
+                raise InvalidFileFormatError(
+                    "The dataset is malformed, the `content` field must be a list of dicts.",
+                    line_number=idx + 1,
+                    error_source="key_value",
+                )
+            if "type" not in item:
+                raise InvalidFileFormatError(
+                    "The dataset is malformed, the `content` field must be a list of dicts with a `type` field.",
+                    line_number=idx + 1,
+                    error_source="key_value",
+                )
+
+            if item["type"] == "text":
+                if "text" not in item or not isinstance(item["text"], str):
+                    raise InvalidFileFormatError(
+                        "The dataset is malformed, the `text` field must be present in the `content` item field and be"
+                        f" a string. Got '{item.get('text')!r}' instead.",
+                        line_number=idx + 1,
+                        error_source="key_value",
+                    )
+            elif item["type"] == "image_url":
+                if role != "user":
+                    raise InvalidFileFormatError(
+                        "The dataset is malformed, only user messages can contain images.",
+                        line_number=idx + 1,
+                        error_source="key_value",
+                    )
+
+                if "image_url" not in item or not isinstance(item["image_url"], dict):
+                    raise InvalidFileFormatError(
+                        "The dataset is malformed, the `image_url` field must be present in the `content` field and "
+                        f"be a dictionary. Got {item.get('image_url')!r} instead.",
+                        line_number=idx + 1,
+                        error_source="key_value",
+                    )
+
+                image_data = cast(Any, item["image_url"]).get("url")
+                if not image_data or not isinstance(image_data, str):
+                    raise InvalidFileFormatError(
+                        "The dataset is malformed, the `url` field must be present in the `image_url` field and be "
+                        f"a string. Got {image_data!r} instead.",
+                        line_number=idx + 1,
+                        error_source="key_value",
+                    )
+
+                if not any(image_data.startswith(f"data:image/{fmt};base64,") for fmt in ["jpeg", "png", "webp"]):
+                    raise InvalidFileFormatError(
+                        "The dataset is malformed, the `url` field must be either a JPEG, PNG or WEBP base64-encoded "
+                        "image in 'data:image/<format>;base64,<base64_encoded_image>' format. "
+                        f"Got '{image_data[:100]}...' instead.",
+                        line_number=idx + 1,
+                    )
+
+                if len(image_data) > MAX_BASE64_IMAGE_LENGTH:
+                    raise InvalidFileFormatError(
+                        "The dataset is malformed, the `url` field must contain base64-encoded image "
+                        f"that is less than {MAX_IMAGE_BYTES // (1024**2)}MB, found ~{len(image_data) * 3 // 4} bytes.",
+                        line_number=idx + 1,
+                        error_source="key_value",
+                    )
+
+                num_images += 1
+            else:
+                raise InvalidFileFormatError(
+                    "The dataset is malformed, the `type` field must be either 'text' or 'image_url'. "
+                    f"Got {item['type']!r}.",
+                    line_number=idx + 1,
+                    error_source="key_value",
+                )
+
+        if num_images > MAX_IMAGES_PER_EXAMPLE:
+            raise InvalidFileFormatError(
+                f"The dataset is malformed, the `content` field must contain at most "
+                f"{MAX_IMAGES_PER_EXAMPLE} images, found {num_images}.",
+                line_number=idx + 1,
+                error_source="key_value",
+            )
+
+        # We still consider text-only messages in such format as multimodal, even if they don't have any images
+        # included - so we can process datasets with rather sparse images (i.e. not in each sample) consistently.
+        return True, num_images
+
+    raise InvalidFileFormatError(
+        f"Invalid content type on line {idx + 1} of the input file. Expected string or multimodal list of dicts, "
+        f"found {type(message_content)}",
+        line_number=idx + 1,
+        error_source="key_value",
+    )
+
+
+def validate_messages(
+    messages: List[Dict[str, str | int | MessageContent]],
+    idx: int,
+    require_assistant_role: bool = True,
+) -> None:
     """Validate the messages column.
 
     Args:
@@ -242,15 +382,43 @@ def validate_messages(messages: List[Dict[str, str | bool]], idx: int, require_a
     """
     _check_conversation_type(messages, idx)
 
-    has_weights = any("weight" in message for message in messages)
     previous_role = None
     assistant_role_exists = False
 
+    messages_are_multimodal: bool | None = None
+    total_number_of_images = 0
+
     for message in messages:
-        if has_weights:
-            _check_message_weight(message, idx)
+        message_weight = _check_message_weight(message, idx)
         previous_role = _check_message_role(message, previous_role, idx)
         assistant_role_exists |= previous_role == "assistant"
+        is_multimodal, number_of_images = _check_message_content(message["content"], role=previous_role, idx=idx)
+        # Multimodal validation
+        if number_of_images > 0 and message_weight is not None and message_weight != 0:
+            raise InvalidFileFormatError(
+                "Messages with images cannot have non-zero weights.",
+                line_number=idx + 1,
+                error_source="key_value",
+            )
+        if messages_are_multimodal is None:
+            # Detect the format of the messages in the conversation.
+            messages_are_multimodal = is_multimodal
+        elif messages_are_multimodal != is_multimodal:
+            # Due to the format limitation, we cannot mix multimodal and text only messages in the same sample.
+            raise InvalidFileFormatError(
+                "Messages in the conversation must be either all in multimodal or all in text-only format.",
+                line_number=idx + 1,
+                error_source="key_value",
+            )
+        total_number_of_images += number_of_images
+
+    if total_number_of_images > MAX_IMAGES_PER_EXAMPLE:
+        raise InvalidFileFormatError(
+            f"The dataset is malformed, the `messages` must contain at most {MAX_IMAGES_PER_EXAMPLE} images. "
+            f"Found {total_number_of_images} images.",
+            line_number=idx + 1,
+            error_source="key_value",
+        )
 
     _check_conversation_roles(require_assistant_role, assistant_role_exists, idx)
 
@@ -279,7 +447,7 @@ def validate_preference_openai(example: Dict[str, Any], idx: int = 0) -> None:
             error_source="key_value",
         )
 
-    messages: List[Dict[str, str | bool]] = cast(Any, example["input"]["messages"])
+    messages: List[Dict[str, str | int | MessageContent]] = cast(Any, example["input"]["messages"])
     validate_messages(messages, idx, require_assistant_role=False)
 
     if example["input"]["messages"][-1]["role"] == "assistant":
@@ -341,12 +509,7 @@ def validate_preference_openai(example: Dict[str, Any], idx: int = 0) -> None:
                 error_source="key_value",
             )
 
-        if not isinstance(example[key][0]["content"], str):
-            raise InvalidFileFormatError(
-                message=f"The dataset is malformed, the 'content' field in `{key}` must be a string on line {idx + 1}.",
-                line_number=idx + 1,
-                error_source="key_value",
-            )
+        _check_message_content(example[key][0]["content"], role="assistant", idx=idx)
 
 
 def _check_utf8(file: Path) -> Dict[str, Any]:
@@ -514,7 +677,7 @@ def _check_jsonl(file: Path, purpose: FilePurpose | str) -> Dict[str, Any]:
                     elif current_format == DatasetFormat.CONVERSATION:
                         message_column = JSONL_REQUIRED_COLUMNS_MAP[DatasetFormat.CONVERSATION][0]
                         require_assistant = purpose != "eval"
-                        message: List[Dict[str, str | bool]] = cast(Any, json_line[message_column])
+                        message: List[Dict[str, str | int | MessageContent]] = cast(Any, json_line[message_column])
                         validate_messages(
                             message,
                             idx,
@@ -522,13 +685,8 @@ def _check_jsonl(file: Path, purpose: FilePurpose | str) -> Dict[str, Any]:
                         )
                     else:
                         for column in JSONL_REQUIRED_COLUMNS_MAP[current_format]:
-                            if not isinstance(json_line[column], str):
-                                raise InvalidFileFormatError(
-                                    message=f'Invalid value type for "{column}" key on line {idx + 1}. '
-                                    f"Expected string. Found {type(cast(Any, json_line[column]))}.",
-                                    line_number=idx + 1,
-                                    error_source="key_value",
-                                )
+                            role = "assistant" if column in {"completion"} else "user"
+                            _check_message_content(cast(Any, json_line[column]), role=role, idx=idx)
 
                     if dataset_format is None:
                         dataset_format = current_format

together/resources/audio/transcriptions.py

@@ -47,7 +47,7 @@ class TranscriptionsResource(SyncAPIResource):
     def create(
         self,
         *,
-        file: FileTypes,
+        file: Union[FileTypes, str],
         diarize: bool | Omit = omit,
         language: str | Omit = omit,
         max_speakers: int | Omit = omit,
@@ -68,7 +68,8 @@ class TranscriptionsResource(SyncAPIResource):
         Transcribes audio into text
 
         Args:
-          file: Audio file to transcribe
+          file: Audio file upload or public HTTP/HTTPS URL. Supported formats .wav, .mp3, .m4a,
+              .webm, .flac.
 
           diarize: Whether to enable speaker diarization. When enabled, you will get the speaker id
               for each word in the transcription. In the response, in the words array, you
@@ -168,7 +169,7 @@ class AsyncTranscriptionsResource(AsyncAPIResource):
     async def create(
         self,
         *,
-        file: FileTypes,
+        file: Union[FileTypes, str],
         diarize: bool | Omit = omit,
         language: str | Omit = omit,
         max_speakers: int | Omit = omit,
@@ -189,7 +190,8 @@ class AsyncTranscriptionsResource(AsyncAPIResource):
         Transcribes audio into text
 
         Args:
-          file: Audio file to transcribe
+          file: Audio file upload or public HTTP/HTTPS URL. Supported formats .wav, .mp3, .m4a,
+              .webm, .flac.
 
           diarize: Whether to enable speaker diarization. When enabled, you will get the speaker id
               for each word in the transcription. In the response, in the words array, you

together/resources/audio/translations.py

@@ -47,7 +47,7 @@ class TranslationsResource(SyncAPIResource):
     def create(
         self,
         *,
-        file: FileTypes,
+        file: Union[FileTypes, str],
         language: str | Omit = omit,
         model: Literal["openai/whisper-large-v3"] | Omit = omit,
         prompt: str | Omit = omit,
@@ -65,7 +65,8 @@ class TranslationsResource(SyncAPIResource):
         Translates audio into English
 
         Args:
-          file: Audio file to translate
+          file: Audio file upload or public HTTP/HTTPS URL. Supported formats .wav, .mp3, .m4a,
+              .webm, .flac.
 
           language: Target output language. Optional ISO 639-1 language code. If omitted, language
               is set to English.
@@ -145,7 +146,7 @@ class AsyncTranslationsResource(AsyncAPIResource):
     async def create(
         self,
         *,
-        file: FileTypes,
+        file: Union[FileTypes, str],
         language: str | Omit = omit,
         model: Literal["openai/whisper-large-v3"] | Omit = omit,
         prompt: str | Omit = omit,
@@ -163,7 +164,8 @@ class AsyncTranslationsResource(AsyncAPIResource):
         Translates audio into English
 
         Args:
-          file: Audio file to translate
+          file: Audio file upload or public HTTP/HTTPS URL. Supported formats .wav, .mp3, .m4a,
+              .webm, .flac.
 
           language: Target output language. Optional ISO 639-1 language code. If omitted, language
               is set to English.

together/resources/fine_tuning.py

@@ -53,6 +53,7 @@ _WARNING_MESSAGE_INSUFFICIENT_FUNDS = (
     "Proceed at your own risk."
 )
 
+
 class FineTuningResource(SyncAPIResource):
     @cached_property
     def with_raw_response(self) -> FineTuningResourceWithRawResponse:
@@ -95,6 +96,7 @@ class FineTuningResource(SyncAPIResource):
         lora_dropout: float | None = 0,
         lora_alpha: float | None = None,
         lora_trainable_modules: str | None = "all-linear",
+        train_vision: bool = False,
         suffix: str | None = None,
         wandb_api_key: str | None = None,
         wandb_base_url: str | None = None,
@@ -140,6 +142,7 @@ class FineTuningResource(SyncAPIResource):
             lora_dropout (float, optional): Dropout rate for LoRA adapters. Defaults to 0.
             lora_alpha (float, optional): Alpha for LoRA adapters. Defaults to 8.
             lora_trainable_modules (str, optional): Trainable modules for LoRA adapters. Defaults to "all-linear".
+            train_vision (bool, optional): Whether to train the vision encoder (Only for multimodal models). Defaults to False.
             suffix (str, optional): Up to 40 character suffix that will be added to your fine-tuned model name.
                 Defaults to None.
             wandb_api_key (str, optional): API key for Weights & Biases integration.
@@ -214,6 +217,7 @@ class FineTuningResource(SyncAPIResource):
             lora_dropout=lora_dropout,
             lora_alpha=lora_alpha,
             lora_trainable_modules=lora_trainable_modules,
+            train_vision=train_vision,
             suffix=suffix,
             wandb_api_key=wandb_api_key,
             wandb_base_url=wandb_base_url,
@@ -232,29 +236,32 @@ 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 not model_limits.supports_vision:
+            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,
+            )
+            price_limit_passed = price_estimation_result.allowed_to_proceed
+        else:
+            # unsupported case
+            price_limit_passed = True
 
         if verbose:
             rprint(
                 "Submitting a fine-tuning job with the following parameters:",
                 finetune_request,
             )
-            if not price_estimation_result.allowed_to_proceed:
+            if not price_limit_passed:
                 rprint(
                     "[red]"
                     + _WARNING_MESSAGE_INSUFFICIENT_FUNDS.format(
-                        price_estimation_result.estimated_total_price # pyright: ignore[reportPossiblyUnboundVariable]
+                        price_estimation_result.estimated_total_price  # pyright: ignore[reportPossiblyUnboundVariable]
                     )
                     + "[/red]",
                 )
@@ -627,6 +634,7 @@ class AsyncFineTuningResource(AsyncAPIResource):
         lora_dropout: float | None = 0,
         lora_alpha: float | None = None,
         lora_trainable_modules: str | None = "all-linear",
+        train_vision: bool = False,
         suffix: str | None = None,
         wandb_api_key: str | None = None,
         wandb_base_url: str | None = None,
@@ -672,6 +680,7 @@ class AsyncFineTuningResource(AsyncAPIResource):
             lora_dropout (float, optional): Dropout rate for LoRA adapters. Defaults to 0.
             lora_alpha (float, optional): Alpha for LoRA adapters. Defaults to 8.
             lora_trainable_modules (str, optional): Trainable modules for LoRA adapters. Defaults to "all-linear".
+            train_vision (bool, optional): Whether to train the vision encoder (Only for multimodal models). Defaults to False.
             suffix (str, optional): Up to 40 character suffix that will be added to your fine-tuned model name.
                 Defaults to None.
             wandb_api_key (str, optional): API key for Weights & Biases integration.
@@ -746,6 +755,7 @@ class AsyncFineTuningResource(AsyncAPIResource):
             lora_dropout=lora_dropout,
             lora_alpha=lora_alpha,
             lora_trainable_modules=lora_trainable_modules,
+            train_vision=train_vision,
             suffix=suffix,
             wandb_api_key=wandb_api_key,
             wandb_base_url=wandb_base_url,
@@ -764,7 +774,6 @@ 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(),
@@ -776,7 +785,6 @@ class AsyncFineTuningResource(AsyncAPIResource):
             training_method=training_method_cls,
         )
 
-
         if verbose:
             rprint(
                 "Submitting a fine-tuning job with the following parameters:",
@@ -786,7 +794,7 @@ class AsyncFineTuningResource(AsyncAPIResource):
                 rprint(
                     "[red]"
                     + _WARNING_MESSAGE_INSUFFICIENT_FUNDS.format(
-                        price_estimation_result.estimated_total_price # pyright: ignore[reportPossiblyUnboundVariable]
+                        price_estimation_result.estimated_total_price  # pyright: ignore[reportPossiblyUnboundVariable]
                     )
                     + "[/red]",
                 )

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/fine_tuning_cancel_response.py

@@ -15,6 +15,7 @@ __all__ = [
     "LrSchedulerLrSchedulerArgs",
     "LrSchedulerLrSchedulerArgsLinearLrSchedulerArgs",
     "LrSchedulerLrSchedulerArgsCosineLrSchedulerArgs",
+    "Progress",
     "TrainingMethod",
     "TrainingMethodTrainingMethodSft",
     "TrainingMethodTrainingMethodDpo",
@@ -50,6 +51,16 @@ class LrScheduler(BaseModel):
     lr_scheduler_args: Optional[LrSchedulerLrSchedulerArgs] = None
 
 
+class Progress(BaseModel):
+    """Progress information for the fine-tuning job"""
+
+    estimate_available: bool
+    """Whether time estimate is available"""
+
+    seconds_remaining: int
+    """Estimated time remaining in seconds for the fine-tuning job to next state"""
+
+
 class TrainingMethodTrainingMethodSft(BaseModel):
     method: Literal["sft"]
 
@@ -163,6 +174,9 @@ class FineTuningCancelResponse(BaseModel):
     owner_address: Optional[str] = None
     """Owner address information"""
 
+    progress: Optional[Progress] = None
+    """Progress information for the fine-tuning job"""
+
     suffix: Optional[str] = None
     """Suffix added to the fine-tuned model name"""
 

together/types/fine_tuning_list_response.py

@@ -16,6 +16,7 @@ __all__ = [
     "DataLrSchedulerLrSchedulerArgs",
     "DataLrSchedulerLrSchedulerArgsLinearLrSchedulerArgs",
     "DataLrSchedulerLrSchedulerArgsCosineLrSchedulerArgs",
+    "DataProgress",
     "DataTrainingMethod",
     "DataTrainingMethodTrainingMethodSft",
     "DataTrainingMethodTrainingMethodDpo",
@@ -51,6 +52,16 @@ class DataLrScheduler(BaseModel):
     lr_scheduler_args: Optional[DataLrSchedulerLrSchedulerArgs] = None
 
 
+class DataProgress(BaseModel):
+    """Progress information for the fine-tuning job"""
+
+    estimate_available: bool
+    """Whether time estimate is available"""
+
+    seconds_remaining: int
+    """Estimated time remaining in seconds for the fine-tuning job to next state"""
+
+
 class DataTrainingMethodTrainingMethodSft(BaseModel):
     method: Literal["sft"]
 
@@ -164,6 +175,9 @@ class Data(BaseModel):
     owner_address: Optional[str] = None
     """Owner address information"""
 
+    progress: Optional[DataProgress] = None
+    """Progress information for the fine-tuning job"""
+
     suffix: Optional[str] = None
     """Suffix added to the fine-tuned model name"""