together 1.5.30__tar.gz → 1.5.31__tar.gz

{together-1.5.30 → together-1.5.31}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: together
-Version: 1.5.30
+Version: 1.5.31
 Summary: Python client for Together's Cloud Platform!
 License: Apache-2.0
 License-File: LICENSE

{together-1.5.30 → together-1.5.31}/pyproject.toml

@@ -12,7 +12,7 @@ build-backend = "poetry.masonry.api"
 
 [tool.poetry]
 name = "together"
-version = "1.5.30"
+version = "1.5.31"
 authors = ["Together AI <support@together.ai>"]
 description = "Python client for Together's Cloud Platform!"
 readme = "README.md"

{together-1.5.30 → together-1.5.31}/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-1.5.31}/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-1.5.31}/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-1.5.31}/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-1.5.31}/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-1.5.31}/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-1.5.31}/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-1.5.31}/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):