Add tests + cleanup

Bumps [pytest](https://github.com/pytest-dev/pytest) from 8.3.3 to 8.3.4.
- [Release notes](https://github.com/pytest-dev/pytest/releases)
- [Changelog](https://github.com/pytest-dev/pytest/blob/main/CHANGELOG.rst)
- [Commits](https://github.com/pytest-dev/pytest/compare/8.3.3...8.3.4)

---
updated-dependencies:
- dependency-name: pytest
  dependency-type: direct:development
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

README.md

@@ -37,9 +37,6 @@ See [_types.py](ollama/_types.py) for more information on the response types.
 
 Response streaming can be enabled by setting `stream=True`.
 
-> [!NOTE]
-> Streaming Tool/Function calling is not yet supported.
-
 ```python
 from ollama import chat
 

examples/README.md

@@ -30,6 +30,16 @@ python3 examples/<example>.py
 - [multimodal_generate.py](multimodal_generate.py)
 
 
+### Structured Outputs - Generate structured outputs with a model
+- [structured-outputs.py](structured-outputs.py)
+- [async-structured-outputs.py](async-structured-outputs.py)
+- [structured-outputs-image.py](structured-outputs-image.py)
+
+
+### Tokenization - Tokenize and detokenize text with a model
+- [tokenization.py](tokenization.py)
+
+
 ### Ollama List - List all downloaded models and their properties
 - [list.py](list.py)
 

examples/async-structured-outputs.py

@@ -0,0 +1,32 @@
+from pydantic import BaseModel
+from ollama import AsyncClient
+import asyncio
+
+
+# Define the schema for the response
+class FriendInfo(BaseModel):
+  name: str
+  age: int
+  is_available: bool
+
+
+class FriendList(BaseModel):
+  friends: list[FriendInfo]
+
+
+async def main():
+  client = AsyncClient()
+  response = await client.chat(
+    model='llama3.1:8b',
+    messages=[{'role': 'user', 'content': 'I have two friends. The first is Ollama 22 years old busy saving the world, and the second is Alonso 23 years old and wants to hang out. Return a list of friends in JSON format'}],
+    format=FriendList.model_json_schema(),  # Use Pydantic to generate the schema
+    options={'temperature': 0},  # Make responses more deterministic
+  )
+
+  # Use Pydantic to validate the response
+  friends_response = FriendList.model_validate_json(response.message.content)
+  print(friends_response)
+
+
+if __name__ == '__main__':
+  asyncio.run(main())

examples/async-tools.py

@@ -41,21 +41,21 @@ subtract_two_numbers_tool = {
   },
 }
 
+messages = [{'role': 'user', 'content': 'What is three plus one?'}]
+print('Prompt:', messages[0]['content'])
+
+available_functions = {
+  'add_two_numbers': add_two_numbers,
+  'subtract_two_numbers': subtract_two_numbers,
+}
+
 
 async def main():
   client = ollama.AsyncClient()
 
-  prompt = 'What is three plus one?'
-  print('Prompt:', prompt)
-
-  available_functions = {
-    'add_two_numbers': add_two_numbers,
-    'subtract_two_numbers': subtract_two_numbers,
-  }
-
   response: ChatResponse = await client.chat(
     'llama3.1',
-    messages=[{'role': 'user', 'content': prompt}],
+    messages=messages,
     tools=[add_two_numbers, subtract_two_numbers_tool],
   )
 
@@ -66,10 +66,24 @@ async def main():
       if function_to_call := available_functions.get(tool.function.name):
         print('Calling function:', tool.function.name)
         print('Arguments:', tool.function.arguments)
-        print('Function output:', function_to_call(**tool.function.arguments))
+        output = function_to_call(**tool.function.arguments)
+        print('Function output:', output)
       else:
         print('Function', tool.function.name, 'not found')
 
+  # Only needed to chat with the model using the tool call results
+  if response.message.tool_calls:
+    # Add the function response to messages for the model to use
+    messages.append(response.message)
+    messages.append({'role': 'tool', 'content': str(output), 'name': tool.function.name})
+
+    # Get final response from model with function outputs
+    final_response = await client.chat('llama3.1', messages=messages)
+    print('Final response:', final_response.message.content)
+
+  else:
+    print('No tool calls returned from model')
+
 
 if __name__ == '__main__':
   try:

examples/structured-outputs-image.py

@@ -0,0 +1,49 @@
+from pathlib import Path
+from pydantic import BaseModel
+from typing import List, Optional, Literal
+from ollama import chat
+
+
+# Define the schema for image objects
+class Object(BaseModel):
+  name: str
+  confidence: float
+  attributes: str
+
+
+class ImageDescription(BaseModel):
+  summary: str
+  objects: List[Object]
+  scene: str
+  colors: List[str]
+  time_of_day: Literal['Morning', 'Afternoon', 'Evening', 'Night']
+  setting: Literal['Indoor', 'Outdoor', 'Unknown']
+  text_content: Optional[str] = None
+
+
+# Get path from user input
+path = input('Enter the path to your image: ')
+path = Path(path)
+
+# Verify the file exists
+if not path.exists():
+  raise FileNotFoundError(f'Image not found at: {path}')
+
+# Set up chat as usual
+response = chat(
+  model='llama3.2-vision',
+  format=ImageDescription.model_json_schema(),  # Pass in the schema for the response
+  messages=[
+    {
+      'role': 'user',
+      'content': 'Analyze this image and return a detailed JSON description including objects, scene, colors and any text detected. If you cannot determine certain details, leave those fields empty.',
+      'images': [path],
+    },
+  ],
+  options={'temperature': 0},  # Set temperature to 0 for more deterministic output
+)
+
+
+# Convert received content to the schema
+image_analysis = ImageDescription.model_validate_json(response.message.content)
+print(image_analysis)

examples/structured-outputs.py

@@ -0,0 +1,26 @@
+from ollama import chat
+from pydantic import BaseModel
+
+
+# Define the schema for the response
+class FriendInfo(BaseModel):
+  name: str
+  age: int
+  is_available: bool
+
+
+class FriendList(BaseModel):
+  friends: list[FriendInfo]
+
+
+# schema = {'type': 'object', 'properties': {'friends': {'type': 'array', 'items': {'type': 'object', 'properties': {'name': {'type': 'string'}, 'age': {'type': 'integer'}, 'is_available': {'type': 'boolean'}}, 'required': ['name', 'age', 'is_available']}}}, 'required': ['friends']}
+response = chat(
+  model='llama3.1:8b',
+  messages=[{'role': 'user', 'content': 'I have two friends. The first is Ollama 22 years old busy saving the world, and the second is Alonso 23 years old and wants to hang out. Return a list of friends in JSON format'}],
+  format=FriendList.model_json_schema(),  # Use Pydantic to generate the schema or format=schema
+  options={'temperature': 0},  # Make responses more deterministic
+)
+
+# Use Pydantic to validate the response
+friends_response = FriendList.model_validate_json(response.message.content)
+print(friends_response)

examples/tokenization.py

@@ -0,0 +1,10 @@
+import ollama
+
+# Get tokens from a model
+response = ollama.tokenize(model='llama3.2', text='Why the sky is blue?')
+tokens = response.tokens
+print('Tokens from model', tokens)
+
+# Convert tokens back to text
+response = ollama.detokenize(model='llama3.2', tokens=tokens)
+print('Text from tokens', response.text)  # Prints: Why the sky is blue?

examples/tools.py

@@ -40,8 +40,8 @@ subtract_two_numbers_tool = {
   },
 }
 
-prompt = 'What is three plus one?'
-print('Prompt:', prompt)
+messages = [{'role': 'user', 'content': 'What is three plus one?'}]
+print('Prompt:', messages[0]['content'])
 
 available_functions = {
   'add_two_numbers': add_two_numbers,
@@ -50,7 +50,7 @@ available_functions = {
 
 response: ChatResponse = chat(
   'llama3.1',
-  messages=[{'role': 'user', 'content': prompt}],
+  messages=messages,
   tools=[add_two_numbers, subtract_two_numbers_tool],
 )
 
@@ -61,6 +61,20 @@ if response.message.tool_calls:
     if function_to_call := available_functions.get(tool.function.name):
       print('Calling function:', tool.function.name)
       print('Arguments:', tool.function.arguments)
-      print('Function output:', function_to_call(**tool.function.arguments))
+      output = function_to_call(**tool.function.arguments)
+      print('Function output:', output)
     else:
       print('Function', tool.function.name, 'not found')
+
+# Only needed to chat with the model using the tool call results
+if response.message.tool_calls:
+  # Add the function response to messages for the model to use
+  messages.append(response.message)
+  messages.append({'role': 'tool', 'content': str(output), 'name': tool.function.name})
+
+  # Get final response from model with function outputs
+  final_response = chat('llama3.1', messages=messages)
+  print('Final response:', final_response.message.content)
+
+else:
+  print('No tool calls returned from model')

ollama/__init__.py

@@ -12,6 +12,8 @@ from ollama._types import (
   ListResponse,
   ShowResponse,
   ProcessResponse,
+  TokenizeResponse,
+  DetokenizeResponse,
   RequestError,
   ResponseError,
 )
@@ -31,6 +33,8 @@ __all__ = [
   'ListResponse',
   'ShowResponse',
   'ProcessResponse',
+  'TokenizeResponse',
+  'DetokenizeResponse',
   'RequestError',
   'ResponseError',
 ]
@@ -49,3 +53,5 @@ list = _client.list
 copy = _client.copy
 show = _client.show
 ps = _client.ps
+tokenize = _client.tokenize
+detokenize = _client.detokenize

ollama/_client.py

@@ -23,6 +23,8 @@ from typing import (
 
 import sys
 
+from pydantic.json_schema import JsonSchemaValue
+
 
 from ollama._utils import convert_function_to_tool
 
@@ -46,6 +48,8 @@ from ollama._types import (
   CreateRequest,
   CopyRequest,
   DeleteRequest,
+  DetokenizeRequest,
+  DetokenizeResponse,
   EmbedRequest,
   EmbedResponse,
   EmbeddingsRequest,
@@ -65,6 +69,8 @@ from ollama._types import (
   ShowRequest,
   ShowResponse,
   StatusResponse,
+  TokenizeRequest,
+  TokenizeResponse,
   Tool,
 )
 
@@ -186,7 +192,7 @@ class Client(BaseClient):
     context: Optional[Sequence[int]] = None,
     stream: Literal[False] = False,
     raw: bool = False,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     images: Optional[Sequence[Union[str, bytes]]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
@@ -204,7 +210,7 @@ class Client(BaseClient):
     context: Optional[Sequence[int]] = None,
     stream: Literal[True] = True,
     raw: bool = False,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     images: Optional[Sequence[Union[str, bytes]]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
@@ -221,7 +227,7 @@ class Client(BaseClient):
     context: Optional[Sequence[int]] = None,
     stream: bool = False,
     raw: Optional[bool] = None,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     images: Optional[Sequence[Union[str, bytes]]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
@@ -265,7 +271,7 @@ class Client(BaseClient):
     *,
     tools: Optional[Sequence[Union[Mapping[str, Any], Tool, Callable]]] = None,
     stream: Literal[False] = False,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
   ) -> ChatResponse: ...
@@ -278,7 +284,7 @@ class Client(BaseClient):
     *,
     tools: Optional[Sequence[Union[Mapping[str, Any], Tool, Callable]]] = None,
     stream: Literal[True] = True,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
   ) -> Iterator[ChatResponse]: ...
@@ -290,7 +296,7 @@ class Client(BaseClient):
     *,
     tools: Optional[Sequence[Union[Mapping[str, Any], Tool, Callable]]] = None,
     stream: bool = False,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
   ) -> Union[ChatResponse, Iterator[ChatResponse]]:
@@ -327,7 +333,6 @@ class Client(BaseClient):
 
     Returns `ChatResponse` if `stream` is `False`, otherwise returns a `ChatResponse` generator.
     """
-
     return self._request(
       ChatResponse,
       'POST',
@@ -610,6 +615,28 @@ class Client(BaseClient):
       '/api/ps',
     )
 
+  def tokenize(self, model: str, text: str) -> TokenizeResponse:
+    return self._request(
+      TokenizeResponse,
+      'POST',
+      '/api/tokenize',
+      json=TokenizeRequest(
+        model=model,
+        text=text,
+      ).model_dump(exclude_none=True),
+    )
+
+  def detokenize(self, model: str, tokens: Sequence[int]) -> DetokenizeResponse:
+    return self._request(
+      DetokenizeResponse,
+      'POST',
+      '/api/detokenize',
+      json=DetokenizeRequest(
+        model=model,
+        tokens=tokens,
+      ).model_dump(exclude_none=True),
+    )
+
 
 class AsyncClient(BaseClient):
   def __init__(self, host: Optional[str] = None, **kwargs) -> None:
@@ -689,7 +716,7 @@ class AsyncClient(BaseClient):
     context: Optional[Sequence[int]] = None,
     stream: Literal[False] = False,
     raw: bool = False,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     images: Optional[Sequence[Union[str, bytes]]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
@@ -707,7 +734,7 @@ class AsyncClient(BaseClient):
     context: Optional[Sequence[int]] = None,
     stream: Literal[True] = True,
     raw: bool = False,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     images: Optional[Sequence[Union[str, bytes]]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
@@ -724,7 +751,7 @@ class AsyncClient(BaseClient):
     context: Optional[Sequence[int]] = None,
     stream: bool = False,
     raw: Optional[bool] = None,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     images: Optional[Sequence[Union[str, bytes]]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
@@ -767,7 +794,7 @@ class AsyncClient(BaseClient):
     *,
     tools: Optional[Sequence[Union[Mapping[str, Any], Tool, Callable]]] = None,
     stream: Literal[False] = False,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
   ) -> ChatResponse: ...
@@ -780,7 +807,7 @@ class AsyncClient(BaseClient):
     *,
     tools: Optional[Sequence[Union[Mapping[str, Any], Tool, Callable]]] = None,
     stream: Literal[True] = True,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
   ) -> AsyncIterator[ChatResponse]: ...
@@ -792,7 +819,7 @@ class AsyncClient(BaseClient):
     *,
     tools: Optional[Sequence[Union[Mapping[str, Any], Tool, Callable]]] = None,
     stream: bool = False,
-    format: Optional[Literal['', 'json']] = None,
+    format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None,
     options: Optional[Union[Mapping[str, Any], Options]] = None,
     keep_alive: Optional[Union[float, str]] = None,
   ) -> Union[ChatResponse, AsyncIterator[ChatResponse]]:
@@ -1119,6 +1146,28 @@ class AsyncClient(BaseClient):
       '/api/ps',
     )
 
+  async def tokenize(self, model: str, text: str) -> TokenizeResponse:
+    return await self._request(
+      TokenizeResponse,
+      'POST',
+      '/api/tokenize',
+      json=TokenizeRequest(
+        model=model,
+        text=text,
+      ).model_dump(exclude_none=True),
+    )
+
+  async def detokenize(self, model: str, tokens: Sequence[int]) -> DetokenizeResponse:
+    return await self._request(
+      DetokenizeResponse,
+      'POST',
+      '/api/detokenize',
+      json=DetokenizeRequest(
+        model=model,
+        tokens=tokens,
+      ).model_dump(exclude_none=True),
+    )
+
 
 def _copy_messages(messages: Optional[Sequence[Union[Mapping[str, Any], Message]]]) -> Iterator[Message]:
   for message in messages or []:

ollama/_types.py

@@ -4,6 +4,7 @@ from pathlib import Path
 from datetime import datetime
 from typing import Any, Mapping, Optional, Union, Sequence
 
+from pydantic.json_schema import JsonSchemaValue
 from typing_extensions import Annotated, Literal
 
 from pydantic import (
@@ -150,7 +151,7 @@ class BaseGenerateRequest(BaseStreamableRequest):
   options: Optional[Union[Mapping[str, Any], Options]] = None
   'Options to use for the request.'
 
-  format: Optional[Literal['', 'json']] = None
+  format: Optional[Union[Literal['', 'json'], JsonSchemaValue]] = None
   'Format of the response.'
 
   keep_alive: Optional[Union[float, str]] = None
@@ -493,6 +494,24 @@ class ProcessResponse(SubscriptableBaseModel):
   models: Sequence[Model]
 
 
+class TokenizeRequest(BaseRequest):
+  model: str
+  text: str
+
+
+class TokenizeResponse(BaseGenerateResponse):
+  tokens: Sequence[int]
+
+
+class DetokenizeRequest(BaseRequest):
+  model: str
+  tokens: Sequence[int]
+
+
+class DetokenizeResponse(BaseGenerateResponse):
+  text: str
+
+
 class RequestError(Exception):
   """
   Common class for request errors.

poetry.lock

@@ -559,13 +559,13 @@ typing-extensions = ">=4.6.0,<4.7.0 || >4.7.0"
 
 [[package]]
 name = "pytest"
-version = "8.3.3"
+version = "8.3.4"
 description = "pytest: simple powerful testing with Python"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "pytest-8.3.3-py3-none-any.whl", hash = "sha256:a6853c7375b2663155079443d2e45de913a911a11d669df02a50814944db57b2"},
-    {file = "pytest-8.3.3.tar.gz", hash = "sha256:70b98107bd648308a7952b06e6ca9a50bc660be218d53c257cc1fc94fda10181"},
+    {file = "pytest-8.3.4-py3-none-any.whl", hash = "sha256:50e16d954148559c9a74109af1eaf0c945ba2d8f30f0a3d3335edde19788b6f6"},
+    {file = "pytest-8.3.4.tar.gz", hash = "sha256:965370d062bce11e73868e0335abac31b4d3de0e82f4007408d242b4f8610761"},
 ]
 
 [package.dependencies]

tests/test_client.py

@@ -1,7 +1,7 @@
 import os
 import io
 import json
-from pydantic import ValidationError
+from pydantic import ValidationError, BaseModel
 import pytest
 import tempfile
 from pathlib import Path
@@ -122,6 +122,128 @@ def test_client_chat_images(httpserver: HTTPServer):
     assert response['message']['content'] == "I don't know."
 
 
+def test_client_chat_format_json(httpserver: HTTPServer):
+  httpserver.expect_ordered_request(
+    '/api/chat',
+    method='POST',
+    json={
+      'model': 'dummy',
+      'messages': [{'role': 'user', 'content': 'Why is the sky blue?'}],
+      'tools': [],
+      'format': 'json',
+      'stream': False,
+    },
+  ).respond_with_json(
+    {
+      'model': 'dummy',
+      'message': {
+        'role': 'assistant',
+        'content': '{"answer": "Because of Rayleigh scattering"}',
+      },
+    }
+  )
+
+  client = Client(httpserver.url_for('/'))
+  response = client.chat('dummy', messages=[{'role': 'user', 'content': 'Why is the sky blue?'}], format='json')
+  assert response['model'] == 'dummy'
+  assert response['message']['role'] == 'assistant'
+  assert response['message']['content'] == '{"answer": "Because of Rayleigh scattering"}'
+
+
+def test_client_chat_format_pydantic(httpserver: HTTPServer):
+  class ResponseFormat(BaseModel):
+    answer: str
+    confidence: float
+
+  httpserver.expect_ordered_request(
+    '/api/chat',
+    method='POST',
+    json={
+      'model': 'dummy',
+      'messages': [{'role': 'user', 'content': 'Why is the sky blue?'}],
+      'tools': [],
+      'format': {'title': 'ResponseFormat', 'type': 'object', 'properties': {'answer': {'title': 'Answer', 'type': 'string'}, 'confidence': {'title': 'Confidence', 'type': 'number'}}, 'required': ['answer', 'confidence']},
+      'stream': False,
+    },
+  ).respond_with_json(
+    {
+      'model': 'dummy',
+      'message': {
+        'role': 'assistant',
+        'content': '{"answer": "Because of Rayleigh scattering", "confidence": 0.95}',
+      },
+    }
+  )
+
+  client = Client(httpserver.url_for('/'))
+  response = client.chat('dummy', messages=[{'role': 'user', 'content': 'Why is the sky blue?'}], format=ResponseFormat.model_json_schema())
+  assert response['model'] == 'dummy'
+  assert response['message']['role'] == 'assistant'
+  assert response['message']['content'] == '{"answer": "Because of Rayleigh scattering", "confidence": 0.95}'
+
+
+@pytest.mark.asyncio
+async def test_async_client_chat_format_json(httpserver: HTTPServer):
+  httpserver.expect_ordered_request(
+    '/api/chat',
+    method='POST',
+    json={
+      'model': 'dummy',
+      'messages': [{'role': 'user', 'content': 'Why is the sky blue?'}],
+      'tools': [],
+      'format': 'json',
+      'stream': False,
+    },
+  ).respond_with_json(
+    {
+      'model': 'dummy',
+      'message': {
+        'role': 'assistant',
+        'content': '{"answer": "Because of Rayleigh scattering"}',
+      },
+    }
+  )
+
+  client = AsyncClient(httpserver.url_for('/'))
+  response = await client.chat('dummy', messages=[{'role': 'user', 'content': 'Why is the sky blue?'}], format='json')
+  assert response['model'] == 'dummy'
+  assert response['message']['role'] == 'assistant'
+  assert response['message']['content'] == '{"answer": "Because of Rayleigh scattering"}'
+
+
+@pytest.mark.asyncio
+async def test_async_client_chat_format_pydantic(httpserver: HTTPServer):
+  class ResponseFormat(BaseModel):
+    answer: str
+    confidence: float
+
+  httpserver.expect_ordered_request(
+    '/api/chat',
+    method='POST',
+    json={
+      'model': 'dummy',
+      'messages': [{'role': 'user', 'content': 'Why is the sky blue?'}],
+      'tools': [],
+      'format': {'title': 'ResponseFormat', 'type': 'object', 'properties': {'answer': {'title': 'Answer', 'type': 'string'}, 'confidence': {'title': 'Confidence', 'type': 'number'}}, 'required': ['answer', 'confidence']},
+      'stream': False,
+    },
+  ).respond_with_json(
+    {
+      'model': 'dummy',
+      'message': {
+        'role': 'assistant',
+        'content': '{"answer": "Because of Rayleigh scattering", "confidence": 0.95}',
+      },
+    }
+  )
+
+  client = AsyncClient(httpserver.url_for('/'))
+  response = await client.chat('dummy', messages=[{'role': 'user', 'content': 'Why is the sky blue?'}], format=ResponseFormat.model_json_schema())
+  assert response['model'] == 'dummy'
+  assert response['message']['role'] == 'assistant'
+  assert response['message']['content'] == '{"answer": "Because of Rayleigh scattering", "confidence": 0.95}'
+
+
 def test_client_generate(httpserver: HTTPServer):
   httpserver.expect_ordered_request(
     '/api/generate',
@@ -205,6 +327,108 @@ def test_client_generate_images(httpserver: HTTPServer):
     assert response['response'] == 'Because it is.'
 
 
+def test_client_generate_format_json(httpserver: HTTPServer):
+  httpserver.expect_ordered_request(
+    '/api/generate',
+    method='POST',
+    json={
+      'model': 'dummy',
+      'prompt': 'Why is the sky blue?',
+      'format': 'json',
+      'stream': False,
+    },
+  ).respond_with_json(
+    {
+      'model': 'dummy',
+      'response': '{"answer": "Because of Rayleigh scattering"}',
+    }
+  )
+
+  client = Client(httpserver.url_for('/'))
+  response = client.generate('dummy', 'Why is the sky blue?', format='json')
+  assert response['model'] == 'dummy'
+  assert response['response'] == '{"answer": "Because of Rayleigh scattering"}'
+
+
+def test_client_generate_format_pydantic(httpserver: HTTPServer):
+  class ResponseFormat(BaseModel):
+    answer: str
+    confidence: float
+
+  httpserver.expect_ordered_request(
+    '/api/generate',
+    method='POST',
+    json={
+      'model': 'dummy',
+      'prompt': 'Why is the sky blue?',
+      'format': {'title': 'ResponseFormat', 'type': 'object', 'properties': {'answer': {'title': 'Answer', 'type': 'string'}, 'confidence': {'title': 'Confidence', 'type': 'number'}}, 'required': ['answer', 'confidence']},
+      'stream': False,
+    },
+  ).respond_with_json(
+    {
+      'model': 'dummy',
+      'response': '{"answer": "Because of Rayleigh scattering", "confidence": 0.95}',
+    }
+  )
+
+  client = Client(httpserver.url_for('/'))
+  response = client.generate('dummy', 'Why is the sky blue?', format=ResponseFormat.model_json_schema())
+  assert response['model'] == 'dummy'
+  assert response['response'] == '{"answer": "Because of Rayleigh scattering", "confidence": 0.95}'
+
+
+@pytest.mark.asyncio
+async def test_async_client_generate_format_json(httpserver: HTTPServer):
+  httpserver.expect_ordered_request(
+    '/api/generate',
+    method='POST',
+    json={
+      'model': 'dummy',
+      'prompt': 'Why is the sky blue?',
+      'format': 'json',
+      'stream': False,
+    },
+  ).respond_with_json(
+    {
+      'model': 'dummy',
+      'response': '{"answer": "Because of Rayleigh scattering"}',
+    }
+  )
+
+  client = AsyncClient(httpserver.url_for('/'))
+  response = await client.generate('dummy', 'Why is the sky blue?', format='json')
+  assert response['model'] == 'dummy'
+  assert response['response'] == '{"answer": "Because of Rayleigh scattering"}'
+
+
+@pytest.mark.asyncio
+async def test_async_client_generate_format_pydantic(httpserver: HTTPServer):
+  class ResponseFormat(BaseModel):
+    answer: str
+    confidence: float
+
+  httpserver.expect_ordered_request(
+    '/api/generate',
+    method='POST',
+    json={
+      'model': 'dummy',
+      'prompt': 'Why is the sky blue?',
+      'format': {'title': 'ResponseFormat', 'type': 'object', 'properties': {'answer': {'title': 'Answer', 'type': 'string'}, 'confidence': {'title': 'Confidence', 'type': 'number'}}, 'required': ['answer', 'confidence']},
+      'stream': False,
+    },
+  ).respond_with_json(
+    {
+      'model': 'dummy',
+      'response': '{"answer": "Because of Rayleigh scattering", "confidence": 0.95}',
+    }
+  )
+
+  client = AsyncClient(httpserver.url_for('/'))
+  response = await client.generate('dummy', 'Why is the sky blue?', format=ResponseFormat.model_json_schema())
+  assert response['model'] == 'dummy'
+  assert response['response'] == '{"answer": "Because of Rayleigh scattering", "confidence": 0.95}'
+
+
 def test_client_pull(httpserver: HTTPServer):
   httpserver.expect_ordered_request(
     '/api/pull',
@@ -1036,3 +1260,27 @@ def test_tool_validation():
   with pytest.raises(ValidationError):
     invalid_tool = {'type': 'invalid_type', 'function': {'name': 'test'}}
     list(_copy_tools([invalid_tool]))
+
+
+def test_client_tokenize(httpserver: HTTPServer):
+  httpserver.expect_ordered_request(
+    '/api/tokenize',
+    method='POST',
+    json={'model': 'dummy', 'text': 'Hello world!'},
+  ).respond_with_json({'tokens': [1, 2, 3]})
+
+  client = Client(httpserver.url_for('/'))
+  response = client.tokenize('dummy', 'Hello world!')
+  assert response.tokens == [1, 2, 3]
+
+
+def test_client_detokenize(httpserver: HTTPServer):
+  httpserver.expect_ordered_request(
+    '/api/detokenize',
+    method='POST',
+    json={'model': 'dummy', 'tokens': [1, 2, 3]},
+  ).respond_with_json({'text': 'Hello world!'})
+
+  client = Client(httpserver.url_for('/'))
+  response = client.detokenize('dummy', [1, 2, 3])
+  assert response.text == 'Hello world!'