Compare commits

...

7 Commits

Author SHA1 Message Date
ParthSareen 4390741023 more wip 2025-09-17 11:01:58 -07:00
Mark Ward 9f41447f20 examples: make gpt-oss resilient for failed tool calls (#569)
test / test (push) Has been cancelled
test / lint (push) Has been cancelled
2025-09-02 13:58:36 -07:00
Parth Sareen da79e987f0 examples: fix gpt-oss-tools-stream for adding toolcalls (#568)
test / test (push) Has been cancelled
test / lint (push) Has been cancelled
2025-08-21 13:44:59 -07:00
Bryon Tjanaka c8392d6524 Fix link for thinking-levels.py (#567)
test / test (push) Has been cancelled
test / lint (push) Has been cancelled
Resolves #554
2025-08-20 00:19:07 -07:00
Parth Sareen 07ab287cdf examples/gpt-oss: fix examples (#566)
test / test (push) Waiting to run
test / lint (push) Waiting to run
2025-08-19 11:08:57 -07:00
dependabot[bot] b0f6b99ca6 build(deps): bump actions/checkout from 4 to 5 (#559)
test / test (push) Has been cancelled
test / lint (push) Has been cancelled
Bumps [actions/checkout](https://github.com/actions/checkout) from 4 to 5.
- [Release notes](https://github.com/actions/checkout/releases)
- [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md)
- [Commits](https://github.com/actions/checkout/compare/v4...v5)

---
updated-dependencies:
- dependency-name: actions/checkout
  dependency-version: '5'
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
2025-08-12 14:40:10 -07:00
Parth Sareen c87604c66f examples: add gpt-oss browser example (#558)
test / test (push) Has been cancelled
test / lint (push) Has been cancelled
2025-08-11 16:59:26 -07:00
10 changed files with 589 additions and 19 deletions
+1 -1
View File
@@ -13,7 +13,7 @@ jobs:
id-token: write
contents: write
steps:
- uses: actions/checkout@v4
- uses: actions/checkout@v5
- uses: actions/setup-python@v5
- uses: astral-sh/setup-uv@v5
with:
+2 -2
View File
@@ -10,7 +10,7 @@ jobs:
test:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/checkout@v5
- uses: astral-sh/setup-uv@v5
with:
enable-cache: true
@@ -19,7 +19,7 @@ jobs:
lint:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/checkout@v5
- uses: actions/setup-python@v5
- uses: astral-sh/setup-uv@v5
with:
+5 -3
View File
@@ -28,8 +28,10 @@ See [ollama/docs/api.md](https://github.com/ollama/ollama/blob/main/docs/api.md)
- [multi-tool.py](multi-tool.py) - Using multiple tools, with thinking enabled
#### gpt-oss
- [gpt-oss-tools.py](gpt-oss-tools.py) - Using tools with gpt-oss
- [gpt-oss-tools-stream.py](gpt-oss-tools-stream.py) - Using tools with gpt-oss, with streaming enabled
- [gpt-oss-tools.py](gpt-oss-tools.py)
- [gpt-oss-tools-stream.py](gpt-oss-tools-stream.py)
- [gpt-oss-tools-browser.py](gpt-oss-tools-browser.py) - Using browser research tools with gpt-oss
- [gpt-oss-tools-browser-stream.py](gpt-oss-tools-browser-stream.py) - Using browser research tools with gpt-oss, with streaming enabled
### Multimodal with Images - Chat with a multimodal (image chat) model
@@ -75,4 +77,4 @@ Requirement: `pip install tqdm`
- [thinking-generate.py](thinking-generate.py)
### Thinking (levels) - Choose the thinking level
- [thinking-levels.py](thinking-generate.py)
- [thinking-levels.py](thinking-levels.py)
+198
View File
@@ -0,0 +1,198 @@
# /// script
# requires-python = ">=3.11"
# dependencies = [
# "gpt-oss",
# "ollama",
# "rich",
# ]
# ///
import asyncio
import json
from typing import Iterator, Optional
from gpt_oss.tools.simple_browser import ExaBackend, SimpleBrowserTool
from openai_harmony import Author, Role, TextContent
from openai_harmony import Message as HarmonyMessage
from rich import print
from ollama import Client
from ollama._types import ChatResponse
_backend = ExaBackend(source='web')
_browser_tool = SimpleBrowserTool(backend=_backend)
def heading(text):
print(text)
print('=' * (len(text) + 3))
async def _browser_search_async(query: str, topn: int = 10, source: str | None = None) -> str:
# map Ollama message to Harmony format
harmony_message = HarmonyMessage(
author=Author(role=Role.USER),
content=[TextContent(text=json.dumps({'query': query, 'topn': topn}))],
recipient='browser.search',
)
result_text: str = ''
async for response in _browser_tool._process(harmony_message):
if response.content:
for content in response.content:
if isinstance(content, TextContent):
result_text += content.text
return result_text or f'No results for query: {query}'
async def _browser_open_async(id: int | str = -1, cursor: int = -1, loc: int = -1, num_lines: int = -1, *, view_source: bool = False, source: str | None = None) -> str:
payload = {'id': id, 'cursor': cursor, 'loc': loc, 'num_lines': num_lines, 'view_source': view_source, 'source': source}
harmony_message = HarmonyMessage(
author=Author(role=Role.USER),
content=[TextContent(text=json.dumps(payload))],
recipient='browser.open',
)
result_text: str = ''
async for response in _browser_tool._process(harmony_message):
if response.content:
for content in response.content:
if isinstance(content, TextContent):
result_text += content.text
return result_text or f'Could not open: {id}'
async def _browser_find_async(pattern: str, cursor: int = -1) -> str:
payload = {'pattern': pattern, 'cursor': cursor}
harmony_message = HarmonyMessage(
author=Author(role=Role.USER),
content=[TextContent(text=json.dumps(payload))],
recipient='browser.find',
)
result_text: str = ''
async for response in _browser_tool._process(harmony_message):
if response.content:
for content in response.content:
if isinstance(content, TextContent):
result_text += content.text
return result_text or f'Pattern not found: {pattern}'
def browser_search(query: str, topn: int = 10, source: Optional[str] = None) -> str:
return asyncio.run(_browser_search_async(query=query, topn=topn, source=source))
def browser_open(id: int | str = -1, cursor: int = -1, loc: int = -1, num_lines: int = -1, *, view_source: bool = False, source: Optional[str] = None) -> str:
return asyncio.run(_browser_open_async(id=id, cursor=cursor, loc=loc, num_lines=num_lines, view_source=view_source, source=source))
def browser_find(pattern: str, cursor: int = -1) -> str:
return asyncio.run(_browser_find_async(pattern=pattern, cursor=cursor))
# Schema definitions for each browser tool
browser_search_schema = {
'type': 'function',
'function': {
'name': 'browser.search',
},
}
browser_open_schema = {
'type': 'function',
'function': {
'name': 'browser.open',
},
}
browser_find_schema = {
'type': 'function',
'function': {
'name': 'browser.find',
},
}
available_tools = {
'browser.search': browser_search,
'browser.open': browser_open,
'browser.find': browser_find,
}
model = 'gpt-oss:20b'
print('Model: ', model, '\n')
prompt = 'What is Ollama?'
print('You: ', prompt, '\n')
messages = [{'role': 'user', 'content': prompt}]
client = Client()
# gpt-oss can call tools while "thinking"
# a loop is needed to call the tools and get the results
final = True
while True:
response_stream: Iterator[ChatResponse] = client.chat(
model=model,
messages=messages,
tools=[browser_search_schema, browser_open_schema, browser_find_schema],
options={'num_ctx': 8192}, # 8192 is the recommended lower limit for the context window
stream=True,
)
tool_calls = []
thinking = ''
content = ''
for chunk in response_stream:
if chunk.message.tool_calls:
tool_calls.extend(chunk.message.tool_calls)
if chunk.message.content:
if not (chunk.message.thinking or chunk.message.thinking == '') and final:
heading('\n\nFinal result: ')
final = False
print(chunk.message.content, end='', flush=True)
if chunk.message.thinking:
thinking += chunk.message.thinking
print(chunk.message.thinking, end='', flush=True)
if thinking != '':
messages.append({'role': 'assistant', 'content': thinking, 'tool_calls': tool_calls})
print()
if tool_calls:
for tool_call in tool_calls:
tool_name = tool_call.function.name
args = tool_call.function.arguments or {}
function_to_call = available_tools.get(tool_name)
if function_to_call:
heading(f'\nCalling tool: {tool_name}')
if args:
print(f'Arguments: {args}')
try:
result = function_to_call(**args)
print(f'Tool result: {result[:200]}')
if len(result) > 200:
heading('... [truncated]')
print()
result_message = {'role': 'tool', 'content': result, 'tool_name': tool_name}
messages.append(result_message)
except Exception as e:
err = f'Error from {tool_name}: {e}'
print(err)
messages.append({'role': 'tool', 'content': err, 'tool_name': tool_name})
else:
print(f'Tool {tool_name} not found')
else:
# no more tool calls, we can stop the loop
break
+175
View File
@@ -0,0 +1,175 @@
# /// script
# requires-python = ">=3.11"
# dependencies = [
# "gpt-oss",
# "ollama",
# "rich",
# ]
# ///
import asyncio
import json
from typing import Optional
from gpt_oss.tools.simple_browser import ExaBackend, SimpleBrowserTool
from openai_harmony import Author, Role, TextContent
from openai_harmony import Message as HarmonyMessage
from ollama import Client
_backend = ExaBackend(source='web')
_browser_tool = SimpleBrowserTool(backend=_backend)
def heading(text):
print(text)
print('=' * (len(text) + 3))
async def _browser_search_async(query: str, topn: int = 10, source: str | None = None) -> str:
# map Ollama message to Harmony format
harmony_message = HarmonyMessage(
author=Author(role=Role.USER),
content=[TextContent(text=json.dumps({'query': query, 'topn': topn}))],
recipient='browser.search',
)
result_text: str = ''
async for response in _browser_tool._process(harmony_message):
if response.content:
for content in response.content:
if isinstance(content, TextContent):
result_text += content.text
return result_text or f'No results for query: {query}'
async def _browser_open_async(id: int | str = -1, cursor: int = -1, loc: int = -1, num_lines: int = -1, *, view_source: bool = False, source: str | None = None) -> str:
payload = {'id': id, 'cursor': cursor, 'loc': loc, 'num_lines': num_lines, 'view_source': view_source, 'source': source}
harmony_message = HarmonyMessage(
author=Author(role=Role.USER),
content=[TextContent(text=json.dumps(payload))],
recipient='browser.open',
)
result_text: str = ''
async for response in _browser_tool._process(harmony_message):
if response.content:
for content in response.content:
if isinstance(content, TextContent):
result_text += content.text
return result_text or f'Could not open: {id}'
async def _browser_find_async(pattern: str, cursor: int = -1) -> str:
payload = {'pattern': pattern, 'cursor': cursor}
harmony_message = HarmonyMessage(
author=Author(role=Role.USER),
content=[TextContent(text=json.dumps(payload))],
recipient='browser.find',
)
result_text: str = ''
async for response in _browser_tool._process(harmony_message):
if response.content:
for content in response.content:
if isinstance(content, TextContent):
result_text += content.text
return result_text or f'Pattern not found: {pattern}'
def browser_search(query: str, topn: int = 10, source: Optional[str] = None) -> str:
return asyncio.run(_browser_search_async(query=query, topn=topn, source=source))
def browser_open(id: int | str = -1, cursor: int = -1, loc: int = -1, num_lines: int = -1, *, view_source: bool = False, source: Optional[str] = None) -> str:
return asyncio.run(_browser_open_async(id=id, cursor=cursor, loc=loc, num_lines=num_lines, view_source=view_source, source=source))
def browser_find(pattern: str, cursor: int = -1) -> str:
return asyncio.run(_browser_find_async(pattern=pattern, cursor=cursor))
# Schema definitions for each browser tool
browser_search_schema = {
'type': 'function',
'function': {
'name': 'browser.search',
},
}
browser_open_schema = {
'type': 'function',
'function': {
'name': 'browser.open',
},
}
browser_find_schema = {
'type': 'function',
'function': {
'name': 'browser.find',
},
}
available_tools = {
'browser.search': browser_search,
'browser.open': browser_open,
'browser.find': browser_find,
}
model = 'gpt-oss:20b'
print('Model: ', model, '\n')
prompt = 'What is Ollama?'
print('You: ', prompt, '\n')
messages = [{'role': 'user', 'content': prompt}]
client = Client()
while True:
response = client.chat(
model=model,
messages=messages,
tools=[browser_search_schema, browser_open_schema, browser_find_schema],
options={'num_ctx': 8192}, # 8192 is the recommended lower limit for the context window
)
if hasattr(response.message, 'thinking') and response.message.thinking:
heading('Thinking')
print(response.message.thinking.strip() + '\n')
if hasattr(response.message, 'content') and response.message.content:
heading('Assistant')
print(response.message.content.strip() + '\n')
# add message to chat history
messages.append(response.message)
if response.message.tool_calls:
for tool_call in response.message.tool_calls:
tool_name = tool_call.function.name
args = tool_call.function.arguments or {}
function_to_call = available_tools.get(tool_name)
if not function_to_call:
print(f'Unknown tool: {tool_name}')
continue
try:
result = function_to_call(**args)
heading(f'Tool: {tool_name}')
if args:
print(f'Arguments: {args}')
print(result[:200])
if len(result) > 200:
print('... [truncated]')
print()
messages.append({'role': 'tool', 'content': result, 'tool_name': tool_name})
except Exception as e:
err = f'Error from {tool_name}: {e}'
print(err)
messages.append({'role': 'tool', 'content': err, 'tool_name': tool_name})
else:
# break on no more tool calls
break
+36 -8
View File
@@ -1,7 +1,17 @@
# /// script
# requires-python = ">=3.11"
# dependencies = [
# "gpt-oss",
# "ollama",
# "rich",
# ]
# ///
import random
from typing import Iterator
from ollama import chat
from rich import print
from ollama import Client
from ollama._types import ChatResponse
@@ -40,37 +50,55 @@ available_tools = {'get_weather': get_weather, 'get_weather_conditions': get_wea
messages = [{'role': 'user', 'content': 'What is the weather like in London? What are the conditions in Toronto?'}]
client = Client(
# Ollama Turbo
# host="https://ollama.com", headers={'Authorization': (os.getenv('OLLAMA_API_KEY'))}
)
model = 'gpt-oss:20b'
# gpt-oss can call tools while "thinking"
# a loop is needed to call the tools and get the results
final = True
while True:
response_stream: Iterator[ChatResponse] = chat(model=model, messages=messages, tools=[get_weather, get_weather_conditions], stream=True)
response_stream: Iterator[ChatResponse] = client.chat(model=model, messages=messages, tools=[get_weather, get_weather_conditions], stream=True)
tool_calls = []
thinking = ''
content = ''
for chunk in response_stream:
if chunk.message.tool_calls:
tool_calls.extend(chunk.message.tool_calls)
if chunk.message.content:
if not (chunk.message.thinking or chunk.message.thinking == '') and final:
print('\nFinal result: ')
print('\n\n' + '=' * 10)
print('Final result: ')
final = False
print(chunk.message.content, end='', flush=True)
if chunk.message.thinking:
# accumulate thinking
thinking += chunk.message.thinking
print(chunk.message.thinking, end='', flush=True)
if thinking != '' or content != '' or len(tool_calls) > 0:
messages.append({'role': 'assistant', 'thinking': thinking, 'content': content, 'tool_calls': tool_calls})
print()
if chunk.message.tool_calls:
for tool_call in chunk.message.tool_calls:
if tool_calls:
for tool_call in tool_calls:
function_to_call = available_tools.get(tool_call.function.name)
if function_to_call:
print('\nCalling tool: ', tool_call.function.name, 'with arguments: ', tool_call.function.arguments)
print('\nCalling tool:', tool_call.function.name, 'with arguments: ', tool_call.function.arguments)
result = function_to_call(**tool_call.function.arguments)
print('Tool result: ', result + '\n')
messages.append(chunk.message)
messages.append({'role': 'tool', 'content': result, 'tool_name': tool_call.function.name})
result_message = {'role': 'tool', 'content': result, 'tool_name': tool_call.function.name}
messages.append(result_message)
else:
print(f'Tool {tool_call.function.name} not found')
messages.append({'role': 'tool', 'content': f'Tool {tool_call.function.name} not found', 'tool_name': tool_call.function.name})
else:
# no more tool calls, we can stop the loop
+19 -5
View File
@@ -1,6 +1,16 @@
# /// script
# requires-python = ">=3.11"
# dependencies = [
# "gpt-oss",
# "ollama",
# "rich",
# ]
# ///
import random
from ollama import chat
from rich import print
from ollama import Client
from ollama._types import ChatResponse
@@ -40,11 +50,15 @@ available_tools = {'get_weather': get_weather, 'get_weather_conditions': get_wea
messages = [{'role': 'user', 'content': 'What is the weather like in London? What are the conditions in Toronto?'}]
client = Client(
# Ollama Turbo
# host="https://ollama.com", headers={'Authorization': (os.getenv('OLLAMA_API_KEY'))}
)
model = 'gpt-oss:20b'
# gpt-oss can call tools while "thinking"
# a loop is needed to call the tools and get the results
while True:
response: ChatResponse = chat(model=model, messages=messages, tools=[get_weather, get_weather_conditions])
response: ChatResponse = client.chat(model=model, messages=messages, tools=[get_weather, get_weather_conditions])
if response.message.content:
print('Content: ')
@@ -53,18 +67,18 @@ while True:
print('Thinking: ')
print(response.message.thinking + '\n')
messages.append(response.message)
if response.message.tool_calls:
for tool_call in response.message.tool_calls:
function_to_call = available_tools.get(tool_call.function.name)
if function_to_call:
result = function_to_call(**tool_call.function.arguments)
print('Result from tool call name: ', tool_call.function.name, 'with arguments: ', tool_call.function.arguments, 'result: ', result + '\n')
messages.append(response.message)
messages.append({'role': 'tool', 'content': result, 'tool_name': tool_call.function.name})
else:
print(f'Tool {tool_call.function.name} not found')
messages.append({'role': 'tool', 'content': f'Tool {tool_call.function.name} not found', 'tool_name': tool_call.function.name})
else:
# no more tool calls, we can stop the loop
break
+11
View File
@@ -1,4 +1,8 @@
from ollama._client import AsyncClient, Client
from ollama._browser import (
Browser
)
from ollama._types import (
ChatResponse,
EmbeddingsResponse,
@@ -15,6 +19,8 @@ from ollama._types import (
ShowResponse,
StatusResponse,
Tool,
WebSearchResponse,
WebCrawlResponse,
)
__all__ = [
@@ -35,6 +41,9 @@ __all__ = [
'ShowResponse',
'StatusResponse',
'Tool',
'WebSearchResponse',
'WebCrawlResponse',
'Browser',
]
_client = Client()
@@ -51,3 +60,5 @@ list = _client.list
copy = _client.copy
show = _client.show
ps = _client.ps
websearch = _client.websearch
webcrawl = _client.webcrawl
+45
View File
@@ -66,6 +66,10 @@ from ollama._types import (
ShowResponse,
StatusResponse,
Tool,
WebCrawlRequest,
WebCrawlResponse,
WebSearchRequest,
WebSearchResponse,
)
T = TypeVar('T')
@@ -102,6 +106,8 @@ class BaseClient:
'Content-Type': 'application/json',
'Accept': 'application/json',
'User-Agent': f'ollama-python/{__version__} ({platform.machine()} {platform.system().lower()}) Python/{platform.python_version()}',
# TODO: this is to make the client feel good
'Authorization': f'Bearer {(headers or {}).get("Authorization") or os.getenv("OLLAMA_API_KEY")}' if (headers or {}).get("Authorization") or os.getenv("OLLAMA_API_KEY") else None,
}.items()
},
**kwargs,
@@ -622,6 +628,45 @@ class Client(BaseClient):
'/api/ps',
)
def websearch(self, query: str, max_results: int = 3) -> WebSearchResponse:
"""
Perform a web search using ollama.com.
Args:
query: The query to search for max_results: The maximum number of results to return.
Returns:
WebSearchResponse with the search results
"""
return self._request(
WebSearchResponse,
'POST',
'https://ollama.com/api/web_search',
json=WebSearchRequest(
queries=[query],
max_results=max_results,
).model_dump(exclude_none=True),
)
def webcrawl(self, urls: Sequence[str]) -> WebCrawlResponse:
"""
Gets the content of web pages for the provided URLs.
Args:
urls: The URLs to crawl
Returns:
WebCrawlResponse with the crawl results
"""
return self._request(
WebCrawlResponse,
'POST',
'https://ollama.com/api/web_crawl',
json=WebCrawlRequest(
urls=urls,
).model_dump(exclude_none=True),
)
class AsyncClient(BaseClient):
def __init__(self, host: Optional[str] = None, **kwargs) -> None:
+97
View File
@@ -538,6 +538,103 @@ class ProcessResponse(SubscriptableBaseModel):
models: Sequence[Model]
class WebSearchRequest(SubscriptableBaseModel):
queries: Sequence[str]
max_results: Optional[int] = None
class SearchResult(SubscriptableBaseModel):
title: str
url: str
content: str
metadata: Optional['SearchResultMetadata'] = None
class CrawlResult(SubscriptableBaseModel):
title: str
url: str
content: str
links: Optional[Sequence[str]] = None
metadata: Optional['CrawlResultMetadata'] = None
class SearchResultContent(SubscriptableBaseModel):
snippet: str
full_text: str
class SearchResultMetadata(SubscriptableBaseModel):
published_date: Optional[str] = None
author: Optional[str] = None
class WebSearchResponse(SubscriptableBaseModel):
results: Mapping[str, Sequence[SearchResult]]
success: bool
errors: Optional[Sequence[str]] = None
def __str__(self) -> str:
if not self.success:
error_msg = ', '.join(self.errors) if self.errors else 'Unknown error'
return f'Web search failed: {error_msg}'
output = []
for query, search_results in self.results.items():
output.append(f'Search results for "{query}":')
for i, result in enumerate(search_results, 1):
output.append(f'{i}. {result.title}')
output.append(f' URL: {result.url}')
output.append(f' Content: {result.content}')
if result.metadata and result.metadata.published_date:
output.append(f' Published: {result.metadata.published_date}')
if result.metadata and result.metadata.author:
output.append(f' Author: {result.metadata.author}')
output.append('')
return '\n'.join(output).rstrip()
class WebCrawlRequest(SubscriptableBaseModel):
urls: Sequence[str]
class CrawlResultContent(SubscriptableBaseModel):
# provides the first 200 characters of the full text
snippet: str
full_text: str
class CrawlResultMetadata(SubscriptableBaseModel):
published_date: Optional[str] = None
author: Optional[str] = None
class WebCrawlResponse(SubscriptableBaseModel):
results: Mapping[str, Sequence[CrawlResult]]
success: bool
errors: Optional[Sequence[str]] = None
def __str__(self) -> str:
if not self.success:
error_msg = ', '.join(self.errors) if self.errors else 'Unknown error'
return f'Web crawl failed: {error_msg}'
output = []
for url, crawl_results in self.results.items():
output.append(f'Crawl results for "{url}":')
for i, result in enumerate(crawl_results, 1):
output.append(f'{i}. {result.title}')
output.append(f' URL: {result.url}')
output.append(f' Content: {result.content}')
if result.links:
output.append(f' Links: {", ".join(result.links)}')
if result.metadata and result.metadata.published_date:
output.append(f' Published: {result.metadata.published_date}')
if result.metadata and result.metadata.author:
output.append(f' Author: {result.metadata.author}')
output.append('')
return '\n'.join(output).rstrip()
class RequestError(Exception):
"""
Common class for request errors.