together 1.5.30__tar.gz → 1.5.32__tar.gz

{together-1.5.30 → together-1.5.32}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: together
-Version: 1.5.30
-Summary: Python client for Together's Cloud Platform!
+Version: 1.5.32
+Summary: Python client for Together's Cloud Platform! Note: SDK 2.0 is now available at https://github.com/togethercomputer/together-py
 License: Apache-2.0
 License-File: LICENSE
 Author: Together AI
@@ -42,6 +42,34 @@ Description-Content-Type: text/markdown
   </a>
 </div>
 
+> [!NOTE]
+> ## 🚀 Together Python SDK 2.0 is now available!
+>
+> Check out the new SDK: **[together-py](https://github.com/togethercomputer/together-py)**
+>
+> 📖 **Migration Guide:** [https://docs.together.ai/docs/pythonv2-migration-guide](https://docs.together.ai/docs/pythonv2-migration-guide)
+>
+> ### Install the Beta
+>
+> **Using uv (Recommended):**
+> ```bash
+> # Install uv if you haven't already
+> curl -LsSf https://astral.sh/uv/install.sh | sh
+>
+> # Install together python SDK
+> uv add together --prerelease allow
+>
+> # Or upgrade an existing installation
+> uv sync --upgrade-package together --prerelease allow
+> ```
+>
+> **Using pip:**
+> ```bash
+> pip install --pre together
+> ```
+>
+> This package will be maintained until January 2026.
+
 # Together Python API library
 
 [![PyPI version](https://img.shields.io/pypi/v/together.svg)](https://pypi.org/project/together/)

{together-1.5.30 → together-1.5.32}/README.md

@@ -4,6 +4,34 @@
   </a>
 </div>
 
+> [!NOTE]
+> ## 🚀 Together Python SDK 2.0 is now available!
+>
+> Check out the new SDK: **[together-py](https://github.com/togethercomputer/together-py)**
+>
+> 📖 **Migration Guide:** [https://docs.together.ai/docs/pythonv2-migration-guide](https://docs.together.ai/docs/pythonv2-migration-guide)
+>
+> ### Install the Beta
+>
+> **Using uv (Recommended):**
+> ```bash
+> # Install uv if you haven't already
+> curl -LsSf https://astral.sh/uv/install.sh | sh
+>
+> # Install together python SDK
+> uv add together --prerelease allow
+>
+> # Or upgrade an existing installation
+> uv sync --upgrade-package together --prerelease allow
+> ```
+>
+> **Using pip:**
+> ```bash
+> pip install --pre together
+> ```
+>
+> This package will be maintained until January 2026.
+
 # Together Python API library
 
 [![PyPI version](https://img.shields.io/pypi/v/together.svg)](https://pypi.org/project/together/)

{together-1.5.30 → together-1.5.32}/pyproject.toml

@@ -12,9 +12,9 @@ build-backend = "poetry.masonry.api"
 
 [tool.poetry]
 name = "together"
-version = "1.5.30"
+version = "1.5.32"
 authors = ["Together AI <support@together.ai>"]
-description = "Python client for Together's Cloud Platform!"
+description = "Python client for Together's Cloud Platform! Note: SDK 2.0 is now available at https://github.com/togethercomputer/together-py"
 readme = "README.md"
 license = "Apache-2.0"
 classifiers = [

together-1.5.32/src/together/__init__.py

@@ -0,0 +1,123 @@
+from __future__ import annotations
+
+import os
+import sys
+
+# =============================================================================
+# SDK 2.0 ANNOUNCEMENT
+# =============================================================================
+_ANNOUNCEMENT_MESSAGE = """
+================================================================================
+Together Python SDK 2.0 is now available!
+
+Install: pip install --pre together
+New SDK: https://github.com/togethercomputer/together-py
+Migration guide: https://docs.together.ai/docs/pythonv2-migration-guide
+
+This package will be maintained until January 2026.
+================================================================================
+"""
+
+# Show info banner (unless suppressed)
+if not os.environ.get("TOGETHER_NO_BANNER"):
+    try:
+        from rich.console import Console
+        from rich.panel import Panel
+
+        console = Console(stderr=True)
+        console.print(
+            Panel(
+                "[bold cyan]Together Python SDK 2.0 is now available![/bold cyan]\n\n"
+                "Install the beta:\n"
+                "[green]pip install --pre together[/green]  or  "
+                "[green]uv add together --prerelease allow[/green]\n\n"
+                "New SDK: [link=https://github.com/togethercomputer/together-py]"
+                "https://github.com/togethercomputer/together-py[/link]\n"
+                "Migration guide: [link=https://docs.together.ai/docs/pythonv2-migration-guide]"
+                "https://docs.together.ai/docs/pythonv2-migration-guide[/link]\n\n"
+                "[dim]This package will be maintained until January 2026.\n"
+                "Set TOGETHER_NO_BANNER=1 to hide this message.[/dim]",
+                title="🚀 New SDK Available",
+                border_style="cyan",
+            )
+        )
+    except Exception:
+        # Fallback for any error (ImportError, OSError in daemons, rich errors, etc.)
+        # Banner display should never break module imports
+        try:
+            print(_ANNOUNCEMENT_MESSAGE, file=sys.stderr)
+        except Exception:
+            pass  # Silently ignore if even stderr is unavailable
+
+# =============================================================================
+
+from contextvars import ContextVar
+from typing import TYPE_CHECKING, Callable
+
+from together import (
+    abstract,
+    client,
+    constants,
+    error,
+    filemanager,
+    resources,
+    together_response,
+    types,
+    utils,
+)
+from together.version import VERSION
+
+from together.legacy.complete import AsyncComplete, Complete, Completion
+from together.legacy.embeddings import Embeddings
+from together.legacy.files import Files
+from together.legacy.finetune import Finetune
+from together.legacy.images import Image
+from together.legacy.models import Models
+
+version = VERSION
+
+log: str | None = None  # Set to either 'debug' or 'info', controls console logging
+
+if TYPE_CHECKING:
+    import requests
+    from aiohttp import ClientSession
+
+requestssession: "requests.Session" | Callable[[], "requests.Session"] | None = None
+
+aiosession: ContextVar["ClientSession" | None] = ContextVar(
+    "aiohttp-session", default=None
+)
+
+from together.client import AsyncClient, AsyncTogether, Client, Together
+
+
+api_key: str | None = None  # To be deprecated in the next major release
+
+# Legacy functions
+
+
+__all__ = [
+    "aiosession",
+    "constants",
+    "version",
+    "Together",
+    "AsyncTogether",
+    "Client",
+    "AsyncClient",
+    "resources",
+    "types",
+    "abstract",
+    "filemanager",
+    "error",
+    "together_response",
+    "client",
+    "utils",
+    "Complete",
+    "AsyncComplete",
+    "Completion",
+    "Embeddings",
+    "Files",
+    "Finetune",
+    "Image",
+    "Models",
+]

{together-1.5.30 → together-1.5.32}/src/together/cli/api/chat.py

@@ -2,7 +2,7 @@ from __future__ import annotations
 
 import cmd
 import json
-from typing import List, Tuple
+from typing import Any, Dict, List, Tuple
 
 import click
 
@@ -181,6 +181,12 @@ def interactive(
     "--frequency-penalty", type=float, help="Frequency penalty sampling method"
 )
 @click.option("--min-p", type=float, help="Min p sampling")
+@click.option(
+    "--audio-url",
+    type=str,
+    multiple=True,
+    help="Audio URL to attach to the last user message",
+)
 @click.option("--no-stream", is_flag=True, help="Disable streaming")
 @click.option("--logprobs", type=int, help="Return logprobs. Only works with --raw.")
 @click.option("--echo", is_flag=True, help="Echo prompt. Only works with --raw.")
@@ -200,6 +206,7 @@ def chat(
     presence_penalty: float | None = None,
     frequency_penalty: float | None = None,
     min_p: float | None = None,
+    audio_url: List[str] | None = None,
     no_stream: bool = False,
     logprobs: int | None = None,
     echo: bool | None = None,
@@ -210,7 +217,22 @@ def chat(
     """Generate chat completions from messages"""
     client: Together = ctx.obj
 
-    messages = [{"role": msg[0], "content": msg[1]} for msg in message]
+    messages: List[Dict[str, Any]] = [
+        {"role": msg[0], "content": msg[1]} for msg in message
+    ]
+
+    if audio_url and messages:
+        last_msg = messages[-1]
+        if last_msg["role"] == "user":
+            # Convert content to list if it is string
+            if isinstance(last_msg["content"], str):
+                last_msg["content"] = [{"type": "text", "text": last_msg["content"]}]
+
+            # Append audio URLs
+            for url in audio_url:
+                last_msg["content"].append(
+                    {"type": "audio_url", "audio_url": {"url": url}}
+                )
 
     response = client.chat.completions.create(
         model=model,

{together-1.5.30 → together-1.5.32}/src/together/cli/api/endpoints.py

@@ -137,8 +137,7 @@ def endpoints(ctx: click.Context) -> None:
     help="Start endpoint in specified availability zone (e.g., us-central-4b)",
 )
 @click.option(
-    "--wait",
-    is_flag=True,
+    "--wait/--no-wait",
     default=True,
     help="Wait for the endpoint to be ready after creation",
 )
@@ -284,7 +283,9 @@ def fetch_and_print_hardware_options(
 @endpoints.command()
 @click.argument("endpoint-id", required=True)
 @click.option(
-    "--wait", is_flag=True, default=True, help="Wait for the endpoint to stop"
+    "--wait/--no-wait",
+    default=True,
+    help="Wait for the endpoint to stop",
 )
 @click.pass_obj
 @handle_api_errors
@@ -307,7 +308,9 @@ def stop(client: Together, endpoint_id: str, wait: bool) -> None:
 @endpoints.command()
 @click.argument("endpoint-id", required=True)
 @click.option(
-    "--wait", is_flag=True, default=True, help="Wait for the endpoint to start"
+    "--wait/--no-wait",
+    default=True,
+    help="Wait for the endpoint to start",
 )
 @click.pass_obj
 @handle_api_errors

{together-1.5.30 → together-1.5.32}/src/together/cli/api/finetune.py

@@ -17,6 +17,8 @@ from together.types.finetune import (
     DownloadCheckpointType,
     FinetuneEventType,
     FinetuneTrainingLimits,
+    FullTrainingType,
+    LoRATrainingType,
 )
 from together.utils import (
     finetune_price_to_dollars,
@@ -29,13 +31,21 @@ from together.utils import (
 
 _CONFIRMATION_MESSAGE = (
     "You are about to create a fine-tuning job. "
-    "The cost of your job will be determined by the model size, the number of tokens "
+    "The estimated price of this job is {price}. "
+    "The actual cost of your job will be determined by the model size, the number of tokens "
     "in the training file, the number of tokens in the validation file, the number of epochs, and "
-    "the number of evaluations. Visit https://www.together.ai/pricing to get a price estimate.\n"
+    "the number of evaluations. Visit https://www.together.ai/pricing to learn more about fine-tuning pricing.\n"
+    "{warning}"
     "You can pass `-y` or `--confirm` to your command to skip this message.\n\n"
     "Do you want to proceed?"
 )
 
+_WARNING_MESSAGE_INSUFFICIENT_FUNDS = (
+    "The estimated price of this job is significantly greater than your current credit limit and balance combined. "
+    "It will likely get cancelled due to insufficient funds. "
+    "Consider increasing your credit limit at https://api.together.xyz/settings/profile\n"
+)
+
 
 class DownloadCheckpointTypeChoice(click.Choice):
     def __init__(self) -> None:
@@ -357,12 +367,36 @@ def create(
             "You have specified a number of evaluation loops but no validation file."
         )
 
-    if confirm or click.confirm(_CONFIRMATION_MESSAGE, default=True, show_default=True):
+    finetune_price_estimation_result = client.fine_tuning.estimate_price(
+        training_file=training_file,
+        validation_file=validation_file,
+        model=model,
+        n_epochs=n_epochs,
+        n_evals=n_evals,
+        training_type="lora" if lora else "full",
+        training_method=training_method,
+    )
+
+    price = click.style(
+        f"${finetune_price_estimation_result.estimated_total_price:.2f}",
+        bold=True,
+    )
+
+    if not finetune_price_estimation_result.allowed_to_proceed:
+        warning = click.style(_WARNING_MESSAGE_INSUFFICIENT_FUNDS, fg="red", bold=True)
+    else:
+        warning = ""
+
+    confirmation_message = _CONFIRMATION_MESSAGE.format(
+        price=price,
+        warning=warning,
+    )
+
+    if confirm or click.confirm(confirmation_message, default=True, show_default=True):
         response = client.fine_tuning.create(
             **training_args,
             verbose=True,
         )
-
         report_string = f"Successfully submitted a fine-tuning job {response.id}"
         if response.created_at is not None:
             created_time = datetime.strptime(

{together-1.5.30 → together-1.5.32}/src/together/constants.py

@@ -20,13 +20,13 @@ 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)
+TARGET_PART_SIZE_MB = 250  # Target part size
+MAX_MULTIPART_PARTS = 250  # Maximum parts per upload
 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
+MAX_FILE_SIZE_GB = 50.1
 
 
 # Messages

{together-1.5.30 → together-1.5.32}/src/together/filemanager.py

@@ -6,10 +6,10 @@ import shutil
 import stat
 import tempfile
 import uuid
-from concurrent.futures import ThreadPoolExecutor, as_completed
+from concurrent.futures import Future, ThreadPoolExecutor, as_completed
 from functools import partial
 from pathlib import Path
-from typing import Any, Dict, List, Tuple
+from typing import Any, BinaryIO, Dict, List, Tuple
 
 import requests
 from filelock import FileLock
@@ -212,6 +212,7 @@ class DownloadManager:
                     ),
                     remaining_retries=MAX_RETRIES,
                     stream=True,
+                    request_timeout=3600,
                 )
 
                 try:
@@ -512,6 +513,18 @@ class MultipartUploadManager:
 
         return response.data
 
+    def _submit_part(
+        self,
+        executor: ThreadPoolExecutor,
+        f: BinaryIO,
+        part_info: Dict[str, Any],
+        part_size: int,
+    ) -> Future[str]:
+        """Submit a single part for upload and return the future"""
+        f.seek((part_info["PartNumber"] - 1) * part_size)
+        part_data = f.read(part_size)
+        return executor.submit(self._upload_single_part, part_info, part_data)
+
     def _upload_parts_concurrent(
         self, file: Path, upload_info: Dict[str, Any], part_size: int
     ) -> List[Dict[str, Any]]:
@@ -522,29 +535,39 @@ class MultipartUploadManager:
 
         with ThreadPoolExecutor(max_workers=self.max_concurrent_parts) as executor:
             with tqdm(total=len(parts), desc="Uploading parts", unit="part") as pbar:
-                future_to_part = {}
-
                 with open(file, "rb") as f:
-                    for part_info in parts:
-                        f.seek((part_info["PartNumber"] - 1) * part_size)
-                        part_data = f.read(part_size)
+                    future_to_part = {}
+                    part_index = 0
 
-                        future = executor.submit(
-                            self._upload_single_part, part_info, part_data
-                        )
+                    # Submit initial batch limited by max_concurrent_parts
+                    for _ in range(min(self.max_concurrent_parts, len(parts))):
+                        part_info = parts[part_index]
+                        future = self._submit_part(executor, f, part_info, part_size)
                         future_to_part[future] = part_info["PartNumber"]
-
-                # Collect results
-                for future in as_completed(future_to_part):
-                    part_number = future_to_part[future]
-                    try:
-                        etag = future.result()
-                        completed_parts.append(
-                            {"part_number": part_number, "etag": etag}
-                        )
-                        pbar.update(1)
-                    except Exception as e:
-                        raise Exception(f"Failed to upload part {part_number}: {e}")
+                        part_index += 1
+
+                    # Process completions and submit new parts (sliding window)
+                    while future_to_part:
+                        done_future = next(as_completed(future_to_part))
+                        part_number = future_to_part.pop(done_future)
+
+                        try:
+                            etag = done_future.result()
+                            completed_parts.append(
+                                {"part_number": part_number, "etag": etag}
+                            )
+                            pbar.update(1)
+                        except Exception as e:
+                            raise Exception(f"Failed to upload part {part_number}: {e}")
+
+                        # Submit next part if available
+                        if part_index < len(parts):
+                            part_info = parts[part_index]
+                            future = self._submit_part(
+                                executor, f, part_info, part_size
+                            )
+                            future_to_part[future] = part_info["PartNumber"]
+                            part_index += 1
 
         completed_parts.sort(key=lambda x: x["part_number"])
         return completed_parts

{together-1.5.30 → together-1.5.32}/src/together/resources/finetune.py

@@ -20,6 +20,8 @@ from together.types import (
     FinetuneLRScheduler,
     FinetuneRequest,
     FinetuneResponse,
+    FinetunePriceEstimationRequest,
+    FinetunePriceEstimationResponse,
     FinetuneTrainingLimits,
     FullTrainingType,
     LinearLRScheduler,
@@ -31,7 +33,7 @@ from together.types import (
     TrainingMethodSFT,
     TrainingType,
 )
-from together.types.finetune import DownloadCheckpointType
+from together.types.finetune import DownloadCheckpointType, TrainingMethod
 from together.utils import log_warn_once, normalize_key
 
 
@@ -42,6 +44,12 @@ AVAILABLE_TRAINING_METHODS = {
     TrainingMethodSFT().method,
     TrainingMethodDPO().method,
 }
+_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."
+)
 
 
 def create_finetune_request(
@@ -473,12 +481,34 @@ class FineTuning:
             hf_api_token=hf_api_token,
             hf_output_repo_name=hf_output_repo_name,
         )
+        if from_checkpoint is None and from_hf_model is None:
+            price_estimation_result = self.estimate_price(
+                training_file=training_file,
+                validation_file=validation_file,
+                model=model_name,
+                n_epochs=finetune_request.n_epochs,
+                n_evals=finetune_request.n_evals,
+                training_type="lora" if lora else "full",
+                training_method=training_method,
+            )
+            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_limit_passed:
+                rprint(
+                    "[red]"
+                    + _WARNING_MESSAGE_INSUFFICIENT_FUNDS.format(
+                        price_estimation_result.estimated_total_price
+                    )
+                    + "[/red]",
+                )
         parameter_payload = finetune_request.model_dump(exclude_none=True)
 
         response, _, _ = requestor.request(
@@ -493,6 +523,81 @@ class FineTuning:
 
         return FinetuneResponse(**response.data)
 
+    def estimate_price(
+        self,
+        *,
+        training_file: str,
+        model: str,
+        validation_file: str | None = None,
+        n_epochs: int | None = 1,
+        n_evals: int | None = 0,
+        training_type: str = "lora",
+        training_method: str = "sft",
+    ) -> FinetunePriceEstimationResponse:
+        """
+        Estimates the price of a fine-tuning job
+
+        Args:
+            training_file (str): File-ID of a file uploaded to the Together API
+            model (str): Name of the base model to run fine-tune job on
+            validation_file (str, optional): File ID of a file uploaded to the Together API for validation.
+            n_epochs (int, optional): Number of epochs for fine-tuning. Defaults to 1.
+            n_evals (int, optional): Number of evaluation loops to run. Defaults to 0.
+            training_type (str, optional): Training type. Defaults to "lora".
+            training_method (str, optional): Training method. Defaults to "sft".
+
+        Returns:
+            FinetunePriceEstimationResponse: Object containing the price estimation result.
+        """
+        training_type_cls: TrainingType
+        training_method_cls: TrainingMethod
+
+        if training_method == "sft":
+            training_method_cls = TrainingMethodSFT(method="sft")
+        elif training_method == "dpo":
+            training_method_cls = TrainingMethodDPO(method="dpo")
+        else:
+            raise ValueError(f"Unknown training method: {training_method}")
+
+        if training_type.lower() == "lora":
+            # parameters of lora are unused in price estimation
+            # but we need to set them to valid values
+            training_type_cls = LoRATrainingType(
+                type="Lora",
+                lora_r=16,
+                lora_alpha=16,
+                lora_dropout=0.0,
+                lora_trainable_modules="all-linear",
+            )
+        elif training_type.lower() == "full":
+            training_type_cls = FullTrainingType(type="Full")
+        else:
+            raise ValueError(f"Unknown training type: {training_type}")
+
+        request = FinetunePriceEstimationRequest(
+            training_file=training_file,
+            validation_file=validation_file,
+            model=model,
+            n_epochs=n_epochs,
+            n_evals=n_evals,
+            training_type=training_type_cls,
+            training_method=training_method_cls,
+        )
+        parameter_payload = request.model_dump(exclude_none=True)
+        requestor = api_requestor.APIRequestor(
+            client=self._client,
+        )
+
+        response, _, _ = requestor.request(
+            options=TogetherRequest(
+                method="POST", url="fine-tunes/estimate-price", params=parameter_payload
+            ),
+            stream=False,
+        )
+        assert isinstance(response, TogetherResponse)
+
+        return FinetunePriceEstimationResponse(**response.data)
+
     def list(self) -> FinetuneList:
         """
         Lists fine-tune job history
@@ -941,11 +1046,34 @@ class AsyncFineTuning:
             hf_output_repo_name=hf_output_repo_name,
         )
 
+        if from_checkpoint is None and from_hf_model is None:
+            price_estimation_result = await self.estimate_price(
+                training_file=training_file,
+                validation_file=validation_file,
+                model=model_name,
+                n_epochs=finetune_request.n_epochs,
+                n_evals=finetune_request.n_evals,
+                training_type="lora" if lora else "full",
+                training_method=training_method,
+            )
+            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_limit_passed:
+                rprint(
+                    "[red]"
+                    + _WARNING_MESSAGE_INSUFFICIENT_FUNDS.format(
+                        price_estimation_result.estimated_total_price
+                    )
+                    + "[/red]",
+                )
         parameter_payload = finetune_request.model_dump(exclude_none=True)
 
         response, _, _ = await requestor.arequest(
@@ -961,6 +1089,81 @@ class AsyncFineTuning:
 
         return FinetuneResponse(**response.data)
 
+    async def estimate_price(
+        self,
+        *,
+        training_file: str,
+        model: str,
+        validation_file: str | None = None,
+        n_epochs: int | None = 1,
+        n_evals: int | None = 0,
+        training_type: str = "lora",
+        training_method: str = "sft",
+    ) -> FinetunePriceEstimationResponse:
+        """
+        Estimates the price of a fine-tuning job
+
+        Args:
+            training_file (str): File-ID of a file uploaded to the Together API
+            model (str): Name of the base model to run fine-tune job on
+            validation_file (str, optional): File ID of a file uploaded to the Together API for validation.
+            n_epochs (int, optional): Number of epochs for fine-tuning. Defaults to 1.
+            n_evals (int, optional): Number of evaluation loops to run. Defaults to 0.
+            training_type (str, optional): Training type. Defaults to "lora".
+            training_method (str, optional): Training method. Defaults to "sft".
+
+        Returns:
+            FinetunePriceEstimationResponse: Object containing the price estimation result.
+        """
+        training_type_cls: TrainingType
+        training_method_cls: TrainingMethod
+
+        if training_method == "sft":
+            training_method_cls = TrainingMethodSFT(method="sft")
+        elif training_method == "dpo":
+            training_method_cls = TrainingMethodDPO(method="dpo")
+        else:
+            raise ValueError(f"Unknown training method: {training_method}")
+
+        if training_type.lower() == "lora":
+            # parameters of lora are unused in price estimation
+            # but we need to set them to valid values
+            training_type_cls = LoRATrainingType(
+                type="Lora",
+                lora_r=16,
+                lora_alpha=16,
+                lora_dropout=0.0,
+                lora_trainable_modules="all-linear",
+            )
+        elif training_type.lower() == "full":
+            training_type_cls = FullTrainingType(type="Full")
+        else:
+            raise ValueError(f"Unknown training type: {training_type}")
+
+        request = FinetunePriceEstimationRequest(
+            training_file=training_file,
+            validation_file=validation_file,
+            model=model,
+            n_epochs=n_epochs,
+            n_evals=n_evals,
+            training_type=training_type_cls,
+            training_method=training_method_cls,
+        )
+        parameter_payload = request.model_dump(exclude_none=True)
+        requestor = api_requestor.APIRequestor(
+            client=self._client,
+        )
+
+        response, _, _ = await requestor.arequest(
+            options=TogetherRequest(
+                method="POST", url="fine-tunes/estimate-price", params=parameter_payload
+            ),
+            stream=False,
+        )
+        assert isinstance(response, TogetherResponse)
+
+        return FinetunePriceEstimationResponse(**response.data)
+
     async def list(self) -> FinetuneList:
         """
         Async method to list fine-tune job history

{together-1.5.30 → together-1.5.32}/src/together/types/__init__.py

@@ -54,6 +54,8 @@ from together.types.finetune import (
     FinetuneListEvents,
     FinetuneRequest,
     FinetuneResponse,
+    FinetunePriceEstimationRequest,
+    FinetunePriceEstimationResponse,
     FinetuneDeleteResponse,
     FinetuneTrainingLimits,
     FullTrainingType,
@@ -103,6 +105,8 @@ __all__ = [
     "FinetuneDeleteResponse",
     "FinetuneDownloadResult",
     "FinetuneLRScheduler",
+    "FinetunePriceEstimationRequest",
+    "FinetunePriceEstimationResponse",
     "LinearLRScheduler",
     "LinearLRSchedulerArgs",
     "CosineLRScheduler",

{together-1.5.30 → together-1.5.32}/src/together/types/chat_completions.py

@@ -46,6 +46,7 @@ class ChatCompletionMessageContentType(str, Enum):
     TEXT = "text"
     IMAGE_URL = "image_url"
     VIDEO_URL = "video_url"
+    AUDIO_URL = "audio_url"
 
 
 class ChatCompletionMessageContentImageURL(BaseModel):
@@ -56,11 +57,16 @@ class ChatCompletionMessageContentVideoURL(BaseModel):
     url: str
 
 
+class ChatCompletionMessageContentAudioURL(BaseModel):
+    url: str
+
+
 class ChatCompletionMessageContent(BaseModel):
     type: ChatCompletionMessageContentType
     text: str | None = None
     image_url: ChatCompletionMessageContentImageURL | None = None
     video_url: ChatCompletionMessageContentVideoURL | None = None
+    audio_url: ChatCompletionMessageContentAudioURL | None = None
 
 
 class ChatCompletionMessage(BaseModel):

{together-1.5.30 → together-1.5.32}/src/together/types/common.py

@@ -26,6 +26,7 @@ class UsageData(BaseModel):
 
 
 class ObjectType(str, Enum):
+    TextCompletion = "text_completion"
     Completion = "text.completion"
     CompletionChunk = "completion.chunk"
     ChatCompletion = "chat.completion"

{together-1.5.30 → together-1.5.32}/src/together/types/files.py

@@ -15,6 +15,7 @@ class FilePurpose(str, Enum):
     FineTune = "fine-tune"
     BatchAPI = "batch-api"
     Eval = "eval"
+    EvalOutput = "eval-output"
 
 
 class FileType(str, Enum):

{together-1.5.30 → together-1.5.32}/src/together/types/finetune.py

@@ -308,6 +308,32 @@ class FinetuneResponse(BaseModel):
             raise ValueError("Unknown training type")
 
 
+class FinetunePriceEstimationRequest(BaseModel):
+    """
+    Fine-tune price estimation request type
+    """
+
+    training_file: str
+    validation_file: str | None = None
+    model: str
+    n_epochs: int
+    n_evals: int
+    training_type: TrainingType
+    training_method: TrainingMethod
+
+
+class FinetunePriceEstimationResponse(BaseModel):
+    """
+    Fine-tune price estimation response type
+    """
+
+    estimated_total_price: float
+    user_limit: float
+    estimated_train_token_count: int
+    estimated_eval_token_count: int
+    allowed_to_proceed: bool
+
+
 class FinetuneList(BaseModel):
     # object type
     object: Literal["list"] | None = None

{together-1.5.30 → together-1.5.32}/src/together/utils/files.py

@@ -7,6 +7,7 @@ from pathlib import Path
 from traceback import format_exc
 from typing import Any, Dict, List
 
+from tqdm import tqdm
 
 from together.constants import (
     MAX_FILE_SIZE_GB,
@@ -363,14 +364,19 @@ def _check_utf8(file: Path) -> Dict[str, Any]:
         Dict[str, Any]: A dictionary with the results of the check.
     """
     report_dict: Dict[str, Any] = {}
+
     try:
+        # Dry-run UTF-8 decode: iterate through file to validate encoding
         with file.open(encoding="utf-8") as f:
-            f.read()
+            for _ in f:
+                pass
+
         report_dict["utf8"] = True
     except UnicodeDecodeError as e:
         report_dict["utf8"] = False
         report_dict["message"] = f"File is not UTF-8 encoded. Error raised: {e}."
         report_dict["is_check_passed"] = False
+
     return report_dict
 
 
@@ -470,7 +476,7 @@ def _check_jsonl(file: Path, purpose: FilePurpose | str) -> Dict[str, Any]:
     with file.open() as f:
         idx = -1
         try:
-            for idx, line in enumerate(f):
+            for idx, line in tqdm(enumerate(f), desc="Validating file", unit=" lines"):
                 json_line = json.loads(line)
 
                 if not isinstance(json_line, dict):

together-1.5.30/src/together/__init__.py

@@ -1,72 +0,0 @@
-from __future__ import annotations
-
-from contextvars import ContextVar
-from typing import TYPE_CHECKING, Callable
-
-from together import (
-    abstract,
-    client,
-    constants,
-    error,
-    filemanager,
-    resources,
-    together_response,
-    types,
-    utils,
-)
-from together.version import VERSION
-
-from together.legacy.complete import AsyncComplete, Complete, Completion
-from together.legacy.embeddings import Embeddings
-from together.legacy.files import Files
-from together.legacy.finetune import Finetune
-from together.legacy.images import Image
-from together.legacy.models import Models
-
-version = VERSION
-
-log: str | None = None  # Set to either 'debug' or 'info', controls console logging
-
-if TYPE_CHECKING:
-    import requests
-    from aiohttp import ClientSession
-
-requestssession: "requests.Session" | Callable[[], "requests.Session"] | None = None
-
-aiosession: ContextVar["ClientSession" | None] = ContextVar(
-    "aiohttp-session", default=None
-)
-
-from together.client import AsyncClient, AsyncTogether, Client, Together
-
-
-api_key: str | None = None  # To be deprecated in the next major release
-
-# Legacy functions
-
-
-__all__ = [
-    "aiosession",
-    "constants",
-    "version",
-    "Together",
-    "AsyncTogether",
-    "Client",
-    "AsyncClient",
-    "resources",
-    "types",
-    "abstract",
-    "filemanager",
-    "error",
-    "together_response",
-    "client",
-    "utils",
-    "Complete",
-    "AsyncComplete",
-    "Completion",
-    "Embeddings",
-    "Files",
-    "Finetune",
-    "Image",
-    "Models",
-]