together 2.0.0a7__tar.gz → 2.0.0a9__tar.gz

together-2.0.0a9/.release-please-manifest.json

@@ -0,0 +1,3 @@
+{
+  ".": "2.0.0-alpha.9"
+}

{together-2.0.0a7 → together-2.0.0a9}/CHANGELOG.md

@@ -1,5 +1,48 @@
 # Changelog
 
+## 2.0.0-alpha.9 (2025-12-03)
+
+Full Changelog: [v2.0.0-alpha.8...v2.0.0-alpha.9](https://github.com/togethercomputer/together-py/compare/v2.0.0-alpha.8...v2.0.0-alpha.9)
+
+### Features
+
+* **api:** api update ([fa5e6f3](https://github.com/togethercomputer/together-py/commit/fa5e6f3eb27475ac2e377bbea9150d45bf4e141e))
+* **api:** api update ([236996f](https://github.com/togethercomputer/together-py/commit/236996f0eba5c0a33d2da59b438a830684e89192))
+
+
+### Bug Fixes
+
+* ensure streams are always closed ([db990c7](https://github.com/togethercomputer/together-py/commit/db990c744ebfffcfe48f52dc44b1ca7b47f1f79a))
+
+
+### Chores
+
+* bump required `uv` version ([1dfec56](https://github.com/togethercomputer/together-py/commit/1dfec5659c5a8e6c8abc7a1035d602a3e47ff67a))
+* **deps:** mypy 1.18.1 has a regression, pin to 1.17 ([2235b95](https://github.com/togethercomputer/together-py/commit/2235b95d3e8dc11c9edc308e2b4b69b1463d21cb))
+* **docs:** use environment variables for authentication in code snippets ([051c1b4](https://github.com/togethercomputer/together-py/commit/051c1b489cb80ded1ad60f6b8722512dd2efae3f))
+* fix internal type issues ([4a2b0f1](https://github.com/togethercomputer/together-py/commit/4a2b0f1cfebc013102e21d54318269a0fe037b7a))
+* update lockfile ([e93c953](https://github.com/togethercomputer/together-py/commit/e93c95338756fb37f279aec946d0c5f74cf22877))
+
+## 2.0.0-alpha.8 (2025-11-26)
+
+Full Changelog: [v2.0.0-alpha.7...v2.0.0-alpha.8](https://github.com/togethercomputer/together-py/compare/v2.0.0-alpha.7...v2.0.0-alpha.8)
+
+### Features
+
+* **api:** api update ([49bb5d4](https://github.com/togethercomputer/together-py/commit/49bb5d4ba69ca118ecc34be2d69c4253665e2e81))
+* **api:** Fix internal references for VideoJob spec ([fb5e7bb](https://github.com/togethercomputer/together-py/commit/fb5e7bb3dbaa9427d291de7440c201529b6cf528))
+
+
+### Bug Fixes
+
+* Address incorrect logic for `endpoint [command] --wait false` logic ([31236a9](https://github.com/togethercomputer/together-py/commit/31236a9df29c22fe7444c2dbb0d4bfc518bc79aa))
+
+
+### Chores
+
+* Remove incorrect file upload docs ([5bb847e](https://github.com/togethercomputer/together-py/commit/5bb847e33b55e5d0978c742e86cf931a2c08f919))
+* Remove incorrect file upload docs ([bb97093](https://github.com/togethercomputer/together-py/commit/bb970938650b6f9580538528979221d142f74b6a))
+
 ## 2.0.0-alpha.7 (2025-11-26)
 
 Full Changelog: [v2.0.0-alpha.6...v2.0.0-alpha.7](https://github.com/togethercomputer/together-py/compare/v2.0.0-alpha.6...v2.0.0-alpha.7)

{together-2.0.0a7 → together-2.0.0a9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: together
-Version: 2.0.0a7
+Version: 2.0.0a9
 Summary: The official Python library for the together API
 Project-URL: Homepage, https://github.com/togethercomputer/together-py
 Project-URL: Repository, https://github.com/togethercomputer/together-py
@@ -144,6 +144,7 @@ pip install 'together[aiohttp] @ git+ssh://git@github.com/togethercomputer/toget
 Then you can enable it by instantiating the client with `http_client=DefaultAioHttpClient()`:
 
 ```python
+import os
 import asyncio
 from together import DefaultAioHttpClient
 from together import AsyncTogether
@@ -151,7 +152,7 @@ from together import AsyncTogether
 
 async def main() -> None:
     async with AsyncTogether(
-        api_key="My API Key",
+        api_key=os.environ.get("TOGETHER_API_KEY"),  # This is the default and can be omitted
         http_client=DefaultAioHttpClient(),
     ) as client:
         chat_completion = await client.chat.completions.create(
@@ -244,22 +245,6 @@ chat_completion = client.chat.completions.create(
 print(chat_completion.response_format)
 ```
 
-## File uploads
-
-Request parameters that correspond to file uploads can be passed as `bytes`, or a [`PathLike`](https://docs.python.org/3/library/os.html#os.PathLike) instance or a tuple of `(filename, contents, media type)`.
-
-```python
-from pathlib import Path
-from together import Together
-
-client = Together()
-
-client.files.upload(
-    file=Path("/path/to/file"),
-    purpose="fine-tune",
-)
-```
-
 The async client uses the exact same interface. If you pass a [`PathLike`](https://docs.python.org/3/library/os.html#os.PathLike) instance, the file contents will be read asynchronously automatically.
 
 ## Handling errors
@@ -528,40 +513,6 @@ with Together() as client:
 
 ## Usage – CLI
 
-### Chat Completions
-
-```bash
-together chat.completions \
-  --message "system" "You are a helpful assistant named Together" \
-  --message "user" "What is your name?" \
-  --model mistralai/Mixtral-8x7B-Instruct-v0.1
-```
-
-The Chat Completions CLI enables streaming tokens to stdout by default. To disable streaming, use `--no-stream`.
-
-### Completions
-
-```bash
-together completions \
-  "Large language models are " \
-  --model mistralai/Mixtral-8x7B-v0.1 \
-  --max-tokens 512 \
-  --stop "."
-```
-
-The Completions CLI enables streaming tokens to stdout by default. To disable streaming, use `--no-stream`.
-
-### Image Generations
-
-```bash
-together images generate \
-  "space robots" \
-  --model stabilityai/stable-diffusion-xl-base-1.0 \
-  --n 4
-```
-
-The image is opened in the default image viewer by default. To disable this, use `--no-show`.
-
 ### Files
 
 ```bash

{together-2.0.0a7 → together-2.0.0a9}/README.md

@@ -98,6 +98,7 @@ pip install 'together[aiohttp] @ git+ssh://git@github.com/togethercomputer/toget
 Then you can enable it by instantiating the client with `http_client=DefaultAioHttpClient()`:
 
 ```python
+import os
 import asyncio
 from together import DefaultAioHttpClient
 from together import AsyncTogether
@@ -105,7 +106,7 @@ from together import AsyncTogether
 
 async def main() -> None:
     async with AsyncTogether(
-        api_key="My API Key",
+        api_key=os.environ.get("TOGETHER_API_KEY"),  # This is the default and can be omitted
         http_client=DefaultAioHttpClient(),
     ) as client:
         chat_completion = await client.chat.completions.create(
@@ -198,22 +199,6 @@ chat_completion = client.chat.completions.create(
 print(chat_completion.response_format)
 ```
 
-## File uploads
-
-Request parameters that correspond to file uploads can be passed as `bytes`, or a [`PathLike`](https://docs.python.org/3/library/os.html#os.PathLike) instance or a tuple of `(filename, contents, media type)`.
-
-```python
-from pathlib import Path
-from together import Together
-
-client = Together()
-
-client.files.upload(
-    file=Path("/path/to/file"),
-    purpose="fine-tune",
-)
-```
-
 The async client uses the exact same interface. If you pass a [`PathLike`](https://docs.python.org/3/library/os.html#os.PathLike) instance, the file contents will be read asynchronously automatically.
 
 ## Handling errors
@@ -482,40 +467,6 @@ with Together() as client:
 
 ## Usage – CLI
 
-### Chat Completions
-
-```bash
-together chat.completions \
-  --message "system" "You are a helpful assistant named Together" \
-  --message "user" "What is your name?" \
-  --model mistralai/Mixtral-8x7B-Instruct-v0.1
-```
-
-The Chat Completions CLI enables streaming tokens to stdout by default. To disable streaming, use `--no-stream`.
-
-### Completions
-
-```bash
-together completions \
-  "Large language models are " \
-  --model mistralai/Mixtral-8x7B-v0.1 \
-  --max-tokens 512 \
-  --stop "."
-```
-
-The Completions CLI enables streaming tokens to stdout by default. To disable streaming, use `--no-stream`.
-
-### Image Generations
-
-```bash
-together images generate \
-  "space robots" \
-  --model stabilityai/stable-diffusion-xl-base-1.0 \
-  --n 4
-```
-
-The image is opened in the default image viewer by default. To disable this, use `--no-show`.
-
 ### Files
 
 ```bash

{together-2.0.0a7 → together-2.0.0a9}/api.md

@@ -134,12 +134,12 @@ Methods:
 Types:
 
 ```python
-from together.types import VideoJob, VideoCreateResponse
+from together.types import VideoJob
 ```
 
 Methods:
 
-- <code title="post /videos">client.videos.<a href="./src/together/resources/videos.py">create</a>(\*\*<a href="src/together/types/video_create_params.py">params</a>) -> <a href="./src/together/types/video_create_response.py">VideoCreateResponse</a></code>
+- <code title="post /videos">client.videos.<a href="./src/together/resources/videos.py">create</a>(\*\*<a href="src/together/types/video_create_params.py">params</a>) -> <a href="./src/together/types/video_job.py">VideoJob</a></code>
 - <code title="get /videos/{id}">client.videos.<a href="./src/together/resources/videos.py">retrieve</a>(id) -> <a href="./src/together/types/video_job.py">VideoJob</a></code>
 
 # Audio

{together-2.0.0a7 → together-2.0.0a9}/pyproject.toml

@@ -1,19 +1,20 @@
 [project]
 name = "together"
-version = "2.0.0-alpha.7"
+version = "2.0.0-alpha.9"
 description = "The official Python library for the together API"
 dynamic = ["readme"]
 license = "Apache-2.0"
 authors = [
 { name = "Together", email = "dev-feedback@TogetherAI.com" },
 ]
+
 dependencies = [
-    "httpx>=0.23.0, <1",
-    "pydantic>=1.9.0, <3",
-    "typing-extensions>=4.10, <5",
-    "anyio>=3.5.0, <5",
-    "distro>=1.7.0, <2",
-    "sniffio",
+  "httpx>=0.23.0, <1",
+  "pydantic>=1.9.0, <3",
+  "typing-extensions>=4.10, <5",
+  "anyio>=3.5.0, <5",
+  "distro>=1.7.0, <2",
+  "sniffio",
     "click>=8.1.7",
     "rich>=13.7.1",
     "tabulate>=0.9.0",
@@ -23,6 +24,7 @@ dependencies = [
     "types-tqdm>=4.67.0.20250516",
     "filelock>=3.13.1",
 ]
+
 requires-python = ">= 3.9"
 classifiers = [
   "Typing :: Typed",
@@ -55,7 +57,7 @@ together = "together.lib.cli.cli:main"
 
 [tool.uv]
 managed = true
-required-version = ">=0.5.0"
+required-version = ">=0.9"
 conflicts = [
     [
       { group = "pydantic-v1" },
@@ -67,7 +69,7 @@ conflicts = [
 # version pins are in uv.lock
 dev = [
     "pyright==1.1.399",
-    "mypy",
+    "mypy==1.17",
     "respx",
     "pytest",
     "pytest-asyncio",

{together-2.0.0a7 → together-2.0.0a9}/requirements-dev.lock

@@ -8,7 +8,7 @@ aiosignal==1.4.0
     # via aiohttp
 annotated-types==0.7.0
     # via pydantic
-anyio==4.11.0
+anyio==4.12.0
     # via
     #   httpx
     #   together
@@ -76,7 +76,7 @@ multidict==6.7.0
     # via
     #   aiohttp
     #   yarl
-mypy==1.18.2
+mypy==1.17.0
 mypy-extensions==1.1.0
     # via mypy
 nodeenv==1.9.1
@@ -95,7 +95,7 @@ propcache==0.4.1
     # via
     #   aiohttp
     #   yarl
-pydantic==2.12.4
+pydantic==2.12.5
     # via together
 pydantic-core==2.41.5
     # via pydantic
@@ -123,13 +123,11 @@ python-dateutil==2.9.0.post0 ; python_full_version < '3.10'
 respx==0.22.0
 rich==14.2.0
     # via together
-ruff==0.14.6
+ruff==0.14.7
 six==1.17.0 ; python_full_version < '3.10'
     # via python-dateutil
 sniffio==1.3.1
-    # via
-    #   anyio
-    #   together
+    # via together
 tabulate==0.9.0
     # via together
 time-machine==2.19.0 ; python_full_version < '3.10'

{together-2.0.0a7 → together-2.0.0a9}/src/together/_streaming.py

@@ -55,30 +55,31 @@ class Stream(Generic[_T]):
         process_data = self._client._process_response_data
         iterator = self._iter_events()
 
-        for sse in iterator:
-            if sse.data.startswith("[DONE]"):
-                break
-
-            if sse.event is None:
-                data = sse.json()
-                if is_mapping(data) and data.get("error"):
-                    message = None
-                    error = data.get("error")
-                    if is_mapping(error):
-                        message = error.get("message")
-                    if not message or not isinstance(message, str):
-                        message = "An error occurred during streaming"
-
-                    raise APIError(
-                        message=message,
-                        request=self.response.request,
-                        body=data["error"],
-                    )
-
-                yield process_data(data=data, cast_to=cast_to, response=response)
-
-        # As we might not fully consume the response stream, we need to close it explicitly
-        response.close()
+        try:
+            for sse in iterator:
+                if sse.data.startswith("[DONE]"):
+                    break
+
+                if sse.event is None:
+                    data = sse.json()
+                    if is_mapping(data) and data.get("error"):
+                        message = None
+                        error = data.get("error")
+                        if is_mapping(error):
+                            message = error.get("message")
+                        if not message or not isinstance(message, str):
+                            message = "An error occurred during streaming"
+
+                        raise APIError(
+                            message=message,
+                            request=self.response.request,
+                            body=data["error"],
+                        )
+
+                    yield process_data(data=data, cast_to=cast_to, response=response)
+        finally:
+            # Ensure the response is closed even if the consumer doesn't read all data
+            response.close()
 
     def __enter__(self) -> Self:
         return self
@@ -137,30 +138,31 @@ class AsyncStream(Generic[_T]):
         process_data = self._client._process_response_data
         iterator = self._iter_events()
 
-        async for sse in iterator:
-            if sse.data.startswith("[DONE]"):
-                break
-
-            if sse.event is None:
-                data = sse.json()
-                if is_mapping(data) and data.get("error"):
-                    message = None
-                    error = data.get("error")
-                    if is_mapping(error):
-                        message = error.get("message")
-                    if not message or not isinstance(message, str):
-                        message = "An error occurred during streaming"
-
-                    raise APIError(
-                        message=message,
-                        request=self.response.request,
-                        body=data["error"],
-                    )
-
-                yield process_data(data=data, cast_to=cast_to, response=response)
-
-        # As we might not fully consume the response stream, we need to close it explicitly
-        await response.aclose()
+        try:
+            async for sse in iterator:
+                if sse.data.startswith("[DONE]"):
+                    break
+
+                if sse.event is None:
+                    data = sse.json()
+                    if is_mapping(data) and data.get("error"):
+                        message = None
+                        error = data.get("error")
+                        if is_mapping(error):
+                            message = error.get("message")
+                        if not message or not isinstance(message, str):
+                            message = "An error occurred during streaming"
+
+                        raise APIError(
+                            message=message,
+                            request=self.response.request,
+                            body=data["error"],
+                        )
+
+                    yield process_data(data=data, cast_to=cast_to, response=response)
+        finally:
+            # Ensure the response is closed even if the consumer doesn't read all data
+            await response.aclose()
 
     async def __aenter__(self) -> Self:
         return self

{together-2.0.0a7 → together-2.0.0a9}/src/together/_utils/_utils.py

@@ -373,7 +373,7 @@ def get_required_header(headers: HeadersLike, header: str) -> str:
     lower_header = header.lower()
     if is_mapping_t(headers):
         # mypy doesn't understand the type narrowing here
-        for k, v in headers.items():  # type: ignore[misc, has-type]
+        for k, v in headers.items():  # type: ignore[misc, has-type, attr-defined]
             if k.lower() == lower_header and isinstance(v, str):  # type: ignore[has-type]
                 return v  # type: ignore[has-type]
 

{together-2.0.0a7 → together-2.0.0a9}/src/together/_version.py

@@ -1,4 +1,4 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
 __title__ = "together"
-__version__ = "2.0.0-alpha.7"  # x-release-please-version
+__version__ = "2.0.0-alpha.9"  # x-release-please-version

{together-2.0.0a7 → together-2.0.0a9}/src/together/lib/cli/api/endpoints.py

@@ -128,8 +128,7 @@ 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",
 )
@@ -272,7 +271,7 @@ def fetch_and_print_hardware_options(client: Together, model: str | None, print_
 
 @endpoints.command()
 @click.argument("endpoint-id", required=True)
-@click.option("--wait", is_flag=True, default=True, help="Wait for the endpoint to stop")
+@click.option("--wait/--no-wait", default=True, help="Wait for the endpoint to stop")
 @click.pass_obj
 @handle_api_errors
 def stop(client: Together, endpoint_id: str, wait: bool) -> None:
@@ -293,7 +292,7 @@ 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")
+@click.option("--wait/--no-wait", default=True, help="Wait for the endpoint to start")
 @click.pass_obj
 @handle_api_errors
 def start(client: Together, endpoint_id: str, wait: bool) -> None:

{together-2.0.0a7 → together-2.0.0a9}/src/together/lib/types/fine_tuning.py

@@ -70,9 +70,6 @@ class FinetuneEventLevels(str, Enum):
     INFO = "Info"
     WARNING = "Warning"
     ERROR = "Error"
-    LEGACY_INFO = "info"
-    LEGACY_IWARNING = "warning"
-    LEGACY_IERROR = "error"
 
 
 class FinetuneEvent(BaseModel):
@@ -85,7 +82,7 @@ class FinetuneEvent(BaseModel):
     # created at datetime stamp
     created_at: Union[str, None] = None
     # event log level
-    level: Union[FinetuneEventLevels, None] = None
+    level: Union[FinetuneEventLevels, str, None] = None
     # event message string
     message: Union[str, None] = None
     # event type
@@ -120,7 +117,20 @@ class LoRATrainingType(BaseModel):
     type: Literal["Lora"] = "Lora"
 
 
-TrainingType: TypeAlias = Union[FullTrainingType, LoRATrainingType]
+class UnknownTrainingType(BaseModel):
+    """
+    Catch-all for unknown training types (forward compatibility).
+    Accepts any training type not explicitly defined.
+    """
+
+    type: str
+
+
+TrainingType: TypeAlias = Union[
+    FullTrainingType,
+    LoRATrainingType,
+    UnknownTrainingType,
+]
 
 
 class FinetuneFullTrainingLimits(BaseModel):
@@ -163,7 +173,19 @@ class TrainingMethodDPO(BaseModel):
     simpo_gamma: Union[float, None] = None
 
 
-TrainingMethod: TypeAlias = Union[TrainingMethodSFT, TrainingMethodDPO]
+class TrainingMethodUnknown(BaseModel):
+    """
+    Catch-all for unknown training methods (forward compatibility).
+    Accepts any training method not explicitly defined.
+    """
+
+    method: str
+
+TrainingMethod: TypeAlias = Union[
+    TrainingMethodSFT,
+    TrainingMethodDPO,
+    TrainingMethodUnknown,
+]
 
 
 class FinetuneTrainingLimits(BaseModel):
@@ -175,31 +197,67 @@ class FinetuneTrainingLimits(BaseModel):
 
 
 class LinearLRSchedulerArgs(BaseModel):
+    """
+    Linear learning rate scheduler arguments
+    """
+
     min_lr_ratio: Union[float, None] = 0.0
 
 
 class CosineLRSchedulerArgs(BaseModel):
+    """
+    Cosine learning rate scheduler arguments
+    """
+
     min_lr_ratio: Union[float, None] = 0.0
     num_cycles: Union[float, None] = 0.5
 
 
 class LinearLRScheduler(BaseModel):
+    """
+    Linear learning rate scheduler
+    """
+
     lr_scheduler_type: Literal["linear"] = "linear"
     lr_scheduler_args: Union[LinearLRSchedulerArgs, None] = None
 
 
 class CosineLRScheduler(BaseModel):
+    """
+    Cosine learning rate scheduler
+    """
+
     lr_scheduler_type: Literal["cosine"] = "cosine"
     lr_scheduler_args: Union[CosineLRSchedulerArgs, None] = None
 
 
-# placeholder for old fine-tuning jobs with no lr_scheduler_type specified
 class EmptyLRScheduler(BaseModel):
+    """
+    Empty learning rate scheduler
+
+    Placeholder for old fine-tuning jobs with no lr_scheduler_type specified
+    """
+
     lr_scheduler_type: Literal[""]
     lr_scheduler_args: None = None
 
+class UnknownLRScheduler(BaseModel):
+    """
+    Unknown learning rate scheduler
+
+    Catch-all for unknown LR scheduler types (forward compatibility)
+    """
+
+    lr_scheduler_type: str
+    lr_scheduler_args: Optional[Any] = None
+
 
-FinetuneLRScheduler: TypeAlias = Union[LinearLRScheduler, CosineLRScheduler, EmptyLRScheduler]
+FinetuneLRScheduler: TypeAlias = Union[
+    LinearLRScheduler,
+    CosineLRScheduler,
+    EmptyLRScheduler,
+    UnknownLRScheduler,
+]
 
 
 class FinetuneResponse(BaseModel):
@@ -213,8 +271,8 @@ class FinetuneResponse(BaseModel):
     created_at: datetime
     """Creation timestamp of the fine-tune job"""
 
-    status: Optional[FinetuneJobStatus] = None
-    """Status of the fine-tune job"""
+    status: Optional[Union[FinetuneJobStatus, str]] = None
+    """Status of the fine-tune job (accepts known enum values or string for forward compatibility)"""
 
     updated_at: datetime
     """Last update timestamp of the fine-tune job"""
@@ -222,8 +280,8 @@ class FinetuneResponse(BaseModel):
     batch_size: Optional[int] = None
     """Batch size used for training"""
 
-    events: Optional[List[FinetuneEvent]] = None
-    """Events related to this fine-tune job"""
+    events: Optional[List[Union[FinetuneEvent, str]]] = None
+    """Events related to this fine-tune job (accepts known enum values or string for forward compatibility)"""
 
     from_checkpoint: Optional[str] = None
     """Checkpoint used to continue training"""
@@ -361,7 +419,7 @@ class FinetuneRequest(BaseModel):
     # training learning rate
     learning_rate: float
     # learning rate scheduler type and args
-    lr_scheduler: Union[LinearLRScheduler, CosineLRScheduler, None] = None
+    lr_scheduler: Union[FinetuneLRScheduler, None] = None
     # learning rate warmup ratio
     warmup_ratio: float
     # max gradient norm
@@ -387,7 +445,7 @@ class FinetuneRequest(BaseModel):
     # training type
     training_type: Union[TrainingType, None] = None
     # training method
-    training_method: Union[TrainingMethodSFT, TrainingMethodDPO] = Field(default_factory=TrainingMethodSFT)
+    training_method: TrainingMethod = Field(default_factory=TrainingMethodSFT)
     # from step
     from_checkpoint: Union[str, None] = None
     from_hf_model: Union[str, None] = None