release: 2.5.0

.github/workflows/ci.yml

@@ -1,12 +1,14 @@
 name: CI
 on:
   push:
-    branches-ignore:
-      - 'generated'
-      - 'codegen/**'
-      - 'integrated/**'
-      - 'stl-preview-head/**'
-      - 'stl-preview-base/**'
+    branches:
+      - '**'
+      - '!integrated/**'
+      - '!stl-preview-head/**'
+      - '!stl-preview-base/**'
+      - '!generated'
+      - '!codegen/**'
+      - 'codegen/stl/**'
   pull_request:
     branches-ignore:
       - 'stl-preview-head/**'

.release-please-manifest.json

@@ -1,3 +1,3 @@
 {
-  ".": "2.4.0"
+  ".": "2.5.0"
 }

.stats.yml

@@ -1,4 +1,4 @@
 configured_endpoints: 74
-openapi_spec_url: https://storage.googleapis.com/stainless-sdk-openapi-specs/togetherai%2Ftogetherai-452048e531c558b879791ccc9788c3f3c23c50c808c909a6d95f47af360566a4.yml
-openapi_spec_hash: 11835b47ff4c2d1b4dbeed74c49908e1
-config_hash: b66198d27b4d5c152688ff6cccfdeab5
+openapi_spec_url: https://storage.googleapis.com/stainless-sdk-openapi-specs/togetherai%2Ftogetherai-31893d157d3c85caa1d8615b73a5fa431ea2cc126bd2410e0f84f3defd5c7dec.yml
+openapi_spec_hash: b652a4d504b4a3dbf585ab803b0f59fc
+config_hash: 52d213100a0ca1a4b2cdcd2718936b51

CHANGELOG.md

@@ -1,5 +1,30 @@
 # Changelog
 
+## 2.5.0 (2026-03-18)
+
+Full Changelog: [v2.4.0...v2.5.0](https://github.com/togethercomputer/together-py/compare/v2.4.0...v2.5.0)
+
+### Features
+
+* **api:** manual updates ([9db2163](https://github.com/togethercomputer/together-py/commit/9db2163681520a95a54e03472ed6719abc40dc05))
+* **api:** manual updates ([a3436ea](https://github.com/togethercomputer/together-py/commit/a3436ea1f04273a0d91103cbd850136854389794))
+* **api:** manual updates ([ca7f97d](https://github.com/togethercomputer/together-py/commit/ca7f97de14718352cb990c281d3fa053e954fc42))
+* Update llama 3.1 8b w/ qwen 3.5 9b ([0680858](https://github.com/togethercomputer/together-py/commit/0680858c6f88e9e8ad72c76dc6d374e5074e12a2))
+
+
+### Bug Fixes
+
+* **deps:** bump minimum typing-extensions version ([94c9632](https://github.com/togethercomputer/together-py/commit/94c96328c13e6955f777e4c6a3517744b84a1d23))
+* **jig:** use Together.get instead of Together._client.get so that registry errors are handled correctly ([#302](https://github.com/togethercomputer/together-py/issues/302)) ([aefd483](https://github.com/togethercomputer/together-py/commit/aefd483f7c335232fe798be6f92f52a960f70e88))
+* **pydantic:** do not pass `by_alias` unless set ([dca687e](https://github.com/togethercomputer/together-py/commit/dca687e39cf86f80cfd051e4e1596e8f54753968))
+* **types:** remove model enum constraint in chat completions ([c30e2a0](https://github.com/togethercomputer/together-py/commit/c30e2a0929888955fe7bae94f38a024ff0a601ed))
+
+
+### Chores
+
+* Fix unit tests with recent model deprecations ([#305](https://github.com/togethercomputer/together-py/issues/305)) ([28902b4](https://github.com/togethercomputer/together-py/commit/28902b4f67f0f5158ea09589fe63de5e51efc9b7))
+* **internal:** tweak CI branches ([556c449](https://github.com/togethercomputer/together-py/commit/556c4491ed2c1ad39f4d587db2c25d94dd210192))
+
 ## 2.4.0 (2026-03-11)
 
 Full Changelog: [v2.3.2...v2.4.0](https://github.com/togethercomputer/together-py/compare/v2.3.2...v2.4.0)

README.md

@@ -193,7 +193,7 @@ chat_completion = client.chat.completions.create(
             "role": "system",
         }
     ],
-    model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
+    model="model",
     reasoning={},
 )
 print(chat_completion.reasoning)

api.md

@@ -117,6 +117,7 @@ Types:
 from together.types.chat import (
     ChatCompletion,
     ChatCompletionChunk,
+    ChatCompletionPrompt,
     ChatCompletionStructuredMessageImageURL,
     ChatCompletionStructuredMessageText,
     ChatCompletionStructuredMessageVideoURL,

examples/fine_tuning.py

@@ -5,7 +5,7 @@
 client = Together()
 
 file_id = "file-bf72b951-fa1a-41af-a152-fe385dca0201"
-fine_tune_model = client.fine_tuning.create(model="meta-llama/Meta-Llama-3-8B", training_file=file_id)
+fine_tune_model = client.fine_tuning.create(model="openai/gpt-oss-20b", training_file=file_id)
 print(fine_tune_model)
 
 fine_tune_id = fine_tune_model.id

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "together"
-version = "2.4.0"
+version = "2.5.0"
 description = "The official Python library for the together API"
 dynamic = ["readme"]
 license = "Apache-2.0"
@@ -11,7 +11,7 @@ authors = [
 dependencies = [
   "httpx>=0.23.0, <1",
   "pydantic>=1.9.0, <3",
-  "typing-extensions>=4.10, <5",
+  "typing-extensions>=4.14, <5",
   "anyio>=3.5.0, <5",
   "distro>=1.7.0, <2",
   "sniffio",

requirements-dev.lock

@@ -42,7 +42,7 @@ execnet==2.1.2
     # via pytest-xdist
 filelock==3.19.1 ; python_full_version < '3.10'
     # via together
-filelock==3.20.3 ; python_full_version >= '3.10'
+filelock==3.25.2 ; python_full_version >= '3.10'
     # via together
 frozenlist==1.8.0
     # via
@@ -87,7 +87,7 @@ pathspec==1.0.3
     # via mypy
 pillow==11.3.0 ; python_full_version < '3.10'
     # via together
-pillow==12.1.0 ; python_full_version >= '3.10'
+pillow==12.1.1 ; python_full_version >= '3.10'
     # via together
 pluggy==1.6.0
     # via pytest
@@ -128,7 +128,9 @@ six==1.17.0 ; python_full_version < '3.10'
     # via python-dateutil
 sniffio==1.3.1
     # via together
-tabulate==0.9.0
+tabulate==0.9.0 ; python_full_version < '3.10'
+    # via together
+tabulate==0.10.0 ; python_full_version >= '3.10'
     # via together
 time-machine==2.19.0 ; python_full_version < '3.10'
 time-machine==3.2.0 ; python_full_version >= '3.10'
@@ -143,9 +145,13 @@ types-pyyaml==6.0.12.20250915
     # via together
 types-requests==2.32.4.20260107
     # via types-tqdm
-types-tabulate==0.9.0.20241207
+types-tabulate==0.9.0.20241207 ; python_full_version < '3.10'
+    # via together
+types-tabulate==0.10.0.20260308 ; python_full_version >= '3.10'
+    # via together
+types-tqdm==4.67.3.20260205 ; python_full_version < '3.10'
     # via together
-types-tqdm==4.67.3.20260205
+types-tqdm==4.67.3.20260303 ; python_full_version >= '3.10'
     # via together
 typing-extensions==4.15.0
     # via

src/together/_compat.py

@@ -2,7 +2,7 @@
 
 from typing import TYPE_CHECKING, Any, Union, Generic, TypeVar, Callable, cast, overload
 from datetime import date, datetime
-from typing_extensions import Self, Literal
+from typing_extensions import Self, Literal, TypedDict
 
 import pydantic
 from pydantic.fields import FieldInfo
@@ -131,6 +131,10 @@ def model_json(model: pydantic.BaseModel, *, indent: int | None = None) -> str:
     return model.model_dump_json(indent=indent)
 
 
+class _ModelDumpKwargs(TypedDict, total=False):
+    by_alias: bool
+
+
 def model_dump(
     model: pydantic.BaseModel,
     *,
@@ -142,14 +146,17 @@ def model_dump(
     by_alias: bool | None = None,
 ) -> dict[str, Any]:
     if (not PYDANTIC_V1) or hasattr(model, "model_dump"):
+        kwargs: _ModelDumpKwargs = {}
+        if by_alias is not None:
+            kwargs["by_alias"] = by_alias
         return model.model_dump(
             mode=mode,
             exclude=exclude,
             exclude_unset=exclude_unset,
             exclude_defaults=exclude_defaults,
             # warnings are not supported in Pydantic v1
             warnings=True if PYDANTIC_V1 else warnings,
-            by_alias=by_alias,
+            **kwargs,
         )
     return cast(
         "dict[str, Any]",

src/together/_version.py

@@ -1,4 +1,4 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
 __title__ = "together"
-__version__ = "2.4.0"  # x-release-please-version
+__version__ = "2.5.0"  # x-release-please-version

src/together/lib/cli/api/beta/jig/jig.py

@@ -505,11 +505,9 @@ def state(self) -> State:
     def registry(self) -> str:
         """Get registry and namespace for current user"""
         if not self.state.registry_base_path:
-            response = self.together._client.get("/image-repositories/base-path", headers=self.together.auth_headers)
-            if not response.is_success:
-                raise JigError(f"Failed to get registry path (HTTP {response.status_code})")
+            response = self.together.get("/image-repositories/base-path", cast_to=dict[str, str])
             # strip protocol for docker image format
-            self.state.registry_base_path = response.json()["base-path"].split("://", 1)[-1]
+            self.state.registry_base_path = response["base-path"].split("://", 1)[-1]
             self.state.save()
         return self.state.registry_base_path + "/"
 

src/together/lib/cli/api/endpoints/create.py

@@ -15,7 +15,7 @@
 @click.option(
     "--model",
     required=True,
-    help="The model to deploy (e.g. meta-llama/Llama-4-Scout-17B-16E-Instruct)",
+    help="The model to deploy",
 )
 @click.option(
     "--min-replicas",

src/together/lib/cli/api/fine_tuning/create.py

@@ -148,6 +148,12 @@ def get_confirmation_message(price: str, warning: str) -> str:
 @click.option("--wandb-project-name", type=str, default=None, help="Wandb project name")
 @click.option("--wandb-name", type=str, default=None, help="Wandb run name")
 @click.option("--wandb-entity", type=str, default=None, help="Wandb entity name")
+@click.option(
+    "--random-seed",
+    type=int,
+    default=None,
+    help="Random seed for reproducible training (e.g. 42). If not set, server default is used.",
+)
 @click.option(
     "--confirm",
     "-y",
@@ -232,6 +238,7 @@ def create(
     wandb_project_name: str | None,
     wandb_name: str | None,
     wandb_entity: str | None,
+    random_seed: int | None,
     confirm: bool | None,
     train_on_inputs: bool | Literal["auto"] | None,
     training_method: str | None,
@@ -275,6 +282,7 @@ def create(
         wandb_project_name=wandb_project_name,
         wandb_name=wandb_name,
         wandb_entity=wandb_entity,
+        random_seed=random_seed,
         train_on_inputs=train_on_inputs,
         training_method=training_method,
         dpo_beta=dpo_beta,

src/together/lib/resources/fine_tuning.py

@@ -59,6 +59,7 @@ def create_finetune_request(
     wandb_project_name: str | None = None,
     wandb_name: str | None = None,
     wandb_entity: str | None = None,
+    random_seed: int | None = None,
     train_on_inputs: bool | Literal["auto"] | None = None,
     training_method: str = "sft",
     dpo_beta: float | None = None,
@@ -249,6 +250,7 @@ def create_finetune_request(
         wandb_project_name=wandb_project_name,
         wandb_name=wandb_name,
         wandb_entity=wandb_entity,
+        random_seed=random_seed,
         training_method=training_method_cls,  # pyright: ignore[reportPossiblyUnboundVariable]
         multimodal_params=multimodal_params,
         from_checkpoint=from_checkpoint,

src/together/lib/types/fine_tuning.py

@@ -483,6 +483,7 @@ class FinetuneRequest(BaseModel):
     wandb_name: Union[str, None] = None
     # wandb entity
     wandb_entity: Union[str, None] = None
+    random_seed: Union[int, None] = None
     # training type
     training_type: Union[TrainingType, None] = None
     # training method