integrate tool calls

examples/tools/README.md

@@ -0,0 +1,3 @@
+# tools
+
+This example demonstrates how to utilize tool calls with an asynchronous Ollama client and the chat endpoint. 

examples/tools/main.py

@@ -0,0 +1,87 @@
+import json
+import ollama
+import asyncio
+
+
+# Simulates an API call to get flight times
+# In a real application, this would fetch data from a live database or API
+def get_flight_times(departure: str, arrival: str) -> str:
+  flights = {
+    'NYC-LAX': {'departure': '08:00 AM', 'arrival': '11:30 AM', 'duration': '5h 30m'},
+    'LAX-NYC': {'departure': '02:00 PM', 'arrival': '10:30 PM', 'duration': '5h 30m'},
+    'LHR-JFK': {'departure': '10:00 AM', 'arrival': '01:00 PM', 'duration': '8h 00m'},
+    'JFK-LHR': {'departure': '09:00 PM', 'arrival': '09:00 AM', 'duration': '7h 00m'},
+    'CDG-DXB': {'departure': '11:00 AM', 'arrival': '08:00 PM', 'duration': '6h 00m'},
+    'DXB-CDG': {'departure': '03:00 AM', 'arrival': '07:30 AM', 'duration': '7h 30m'},
+  }
+
+  key = f'{departure}-{arrival}'.upper()
+  return json.dumps(flights.get(key, {'error': 'Flight not found'}))
+
+
+async def run(model: str):
+  client = ollama.AsyncClient()
+  # Initialize conversation with a user query
+  messages = [{'role': 'user', 'content': 'What is the flight time from New York (NYC) to Los Angeles (LAX)?'}]
+
+  # First API call: Send the query and function description to the model
+  response = await client.chat(
+    model=model,
+    messages=messages,
+    tools=[
+      {
+        'type': 'function',
+        'function': {
+          'name': 'get_flight_times',
+          'description': 'Get the flight times between two cities',
+          'parameters': {
+            'type': 'object',
+            'properties': {
+              'departure': {
+                'type': 'string',
+                'description': 'The departure city (airport code)',
+              },
+              'arrival': {
+                'type': 'string',
+                'description': 'The arrival city (airport code)',
+              },
+            },
+            'required': ['departure', 'arrival'],
+          },
+        },
+      },
+    ],
+  )
+
+  # Add the model's response to the conversation history
+  messages.append(response['message'])
+
+  # Check if the model decided to use the provided function
+  if not response['message'].get('tool_calls'):
+    print("The model didn't use the function. Its response was:")
+    print(response['message']['content'])
+    return
+
+  # Process function calls made by the model
+  if response['message'].get('tool_calls'):
+    available_functions = {
+      'get_flight_times': get_flight_times,
+    }
+    for tool in response['message']['tool_calls']:
+      function_to_call = available_functions[tool['function']['name']]
+      function_response = function_to_call(tool['function']['arguments']['departure'], tool['function']['arguments']['arrival'])
+      # Add function response to the conversation
+      messages.append(
+        {
+          'role': 'tool',
+          'content': function_response,
+        }
+      )
+
+  # Second API call: Get final response from the model
+  final_response = await client.chat(model=model, messages=messages)
+  print(final_response['message']['content'])
+
+
+# Run the async function
+asyncio.run(run('mistral'))

ollama/_client.py

@@ -27,7 +27,7 @@
 except metadata.PackageNotFoundError:
   __version__ = '0.0.0'
 
-from ollama._types import Message, Options, RequestError, ResponseError
+from ollama._types import Message, Options, RequestError, ResponseError, Tool
 
 
 class BaseClient:
@@ -180,6 +180,7 @@ def chat(
     self,
     model: str = '',
     messages: Optional[Sequence[Message]] = None,
+    tools: Optional[Sequence[Tool]] = None,
     stream: Literal[False] = False,
     format: Literal['', 'json'] = '',
     options: Optional[Options] = None,
@@ -191,6 +192,7 @@ def chat(
     self,
     model: str = '',
     messages: Optional[Sequence[Message]] = None,
+    tools: Optional[Sequence[Tool]] = None,
     stream: Literal[True] = True,
     format: Literal['', 'json'] = '',
     options: Optional[Options] = None,
@@ -201,6 +203,7 @@ def chat(
     self,
     model: str = '',
     messages: Optional[Sequence[Message]] = None,
+    tools: Optional[Sequence[Tool]] = None,
     stream: bool = False,
     format: Literal['', 'json'] = '',
     options: Optional[Options] = None,
@@ -222,12 +225,6 @@ def chat(
     messages = deepcopy(messages)
 
     for message in messages or []:
-      if not isinstance(message, dict):
-        raise TypeError('messages must be a list of Message or dict-like objects')
-      if not (role := message.get('role')) or role not in ['system', 'user', 'assistant']:
-        raise RequestError('messages must contain a role and it must be one of "system", "user", or "assistant"')
-      if 'content' not in message:
-        raise RequestError('messages must contain content')
       if images := message.get('images'):
         message['images'] = [_encode_image(image) for image in images]
 
@@ -237,6 +234,7 @@ def chat(
       json={
         'model': model,
         'messages': messages,
+        'tools': tools or [],
         'stream': stream,
         'format': format,
         'options': options or {},
@@ -574,6 +572,7 @@ async def chat(
     self,
     model: str = '',
     messages: Optional[Sequence[Message]] = None,
+    tools: Optional[Sequence[Tool]] = None,
     stream: Literal[False] = False,
     format: Literal['', 'json'] = '',
     options: Optional[Options] = None,
@@ -585,6 +584,7 @@ async def chat(
     self,
     model: str = '',
     messages: Optional[Sequence[Message]] = None,
+    tools: Optional[Sequence[Tool]] = None,
     stream: Literal[True] = True,
     format: Literal['', 'json'] = '',
     options: Optional[Options] = None,
@@ -595,6 +595,7 @@ async def chat(
     self,
     model: str = '',
     messages: Optional[Sequence[Message]] = None,
+    tools: Optional[Sequence[Tool]] = None,
     stream: bool = False,
     format: Literal['', 'json'] = '',
     options: Optional[Options] = None,
@@ -615,12 +616,6 @@ async def chat(
     messages = deepcopy(messages)
 
     for message in messages or []:
-      if not isinstance(message, dict):
-        raise TypeError('messages must be a list of strings')
-      if not (role := message.get('role')) or role not in ['system', 'user', 'assistant']:
-        raise RequestError('messages must contain a role and it must be one of "system", "user", or "assistant"')
-      if 'content' not in message:
-        raise RequestError('messages must contain content')
       if images := message.get('images'):
         message['images'] = [_encode_image(image) for image in images]
 
@@ -630,6 +625,7 @@ async def chat(
       json={
         'model': model,
         'messages': messages,
+        'tools': tools or [],
         'stream': stream,
         'format': format,
         'options': options or {},

ollama/_types.py

@@ -1,5 +1,5 @@
 import json
-from typing import Any, TypedDict, Sequence, Literal
+from typing import Any, TypedDict, Sequence, Literal, Mapping
 
 import sys
 
@@ -53,6 +53,27 @@ class GenerateResponse(BaseGenerateResponse):
   'Tokenized history up to the point of the response.'
 
 
+class ToolCallFunction(TypedDict):
+  """
+  Tool call function.
+  """
+
+  name: str
+  'Name of the function.'
+
+  args: NotRequired[Mapping[str, Any]]
+  'Arguments of the function.'
+
+
+class ToolCall(TypedDict):
+  """
+  Model tool calls.
+  """
+
+  function: ToolCallFunction
+  'Function to be called.'
+
+
 class Message(TypedDict):
   """
   Chat message.
@@ -76,6 +97,34 @@ class Message(TypedDict):
   Valid image formats depend on the model. See the model card for more information.
   """
 
+  tool_calls: NotRequired[Sequence[ToolCall]]
+  """
+  Tools calls to be made by the model.
+  """
+
+
+class Property(TypedDict):
+  type: str
+  description: str
+  enum: NotRequired[Sequence[str]]  # `enum` is optional and can be a list of strings
+
+
+class Parameters(TypedDict):
+  type: str
+  required: Sequence[str]
+  properties: Mapping[str, Property]
+
+
+class ToolFunction(TypedDict):
+  name: str
+  description: str
+  parameters: Parameters
+
+
+class Tool(TypedDict):
+  type: str
+  function: ToolFunction
+
 
 class ChatResponse(BaseGenerateResponse):
   """

tests/test_client.py

@@ -26,6 +26,7 @@ def test_client_chat(httpserver: HTTPServer):
     json={
       'model': 'dummy',
       'messages': [{'role': 'user', 'content': 'Why is the sky blue?'}],
+      'tools': [],
       'stream': False,
       'format': '',
       'options': {},
@@ -73,6 +74,7 @@ def generate():
     json={
       'model': 'dummy',
       'messages': [{'role': 'user', 'content': 'Why is the sky blue?'}],
+      'tools': [],
       'stream': True,
       'format': '',
       'options': {},
@@ -102,6 +104,7 @@ def test_client_chat_images(httpserver: HTTPServer):
           'images': ['iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAIAAACQd1PeAAAADElEQVR4nGNgYGAAAAAEAAH2FzhVAAAAAElFTkSuQmCC'],
         },
       ],
+      'tools': [],
       'stream': False,
       'format': '',
       'options': {},
@@ -522,6 +525,7 @@ async def test_async_client_chat(httpserver: HTTPServer):
     json={
       'model': 'dummy',
       'messages': [{'role': 'user', 'content': 'Why is the sky blue?'}],
+      'tools': [],
       'stream': False,
       'format': '',
       'options': {},
@@ -560,6 +564,7 @@ def generate():
     json={
       'model': 'dummy',
       'messages': [{'role': 'user', 'content': 'Why is the sky blue?'}],
+      'tools': [],
       'stream': True,
       'format': '',
       'options': {},
@@ -590,6 +595,7 @@ async def test_async_client_chat_images(httpserver: HTTPServer):
           'images': ['iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAIAAACQd1PeAAAADElEQVR4nGNgYGAAAAAEAAH2FzhVAAAAAElFTkSuQmCC'],
         },
       ],
+      'tools': [],
       'stream': False,
       'format': '',
       'options': {},