Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

fix(llm): handle Ollama chat params and JSON fences

MagMuellered5f46795789
diff --git a/browser_use/llm/ollama/chat.py b/browser_use/llm/ollama/chat.py
index bbec7030a..c10d25618 100644
--- a/browser_use/llm/ollama/chat.py
+++ b/browser_use/llm/ollama/chat.py
@@ -1,4 +1,5 @@
import logging
+import re
from collections.abc import Mapping
from dataclasses import dataclass
from typing import Any, TypeVar, overload
@@ -18,16 +19,17 @@ T = TypeVar('T', bound=BaseModel)
logger = logging.getLogger(__name__)
# These belong on AsyncClient.chat(), not in the model `options` dict.
-_TOP_LEVEL_CHAT_KEYS = frozenset({'format', 'stream'})
+_PASSTHROUGH_CHAT_KEYS = frozenset({'think', 'logprobs', 'top_logprobs', 'keep_alive'})
+_IGNORED_CHAT_KEYS = frozenset({'format', 'stream'})
+_JSON_FENCE_RE = re.compile(r'\A```[ \t]*(?:json)?[ \t]*\r?\n(?P<body>.*?)\r?\n?```[ \t]*\Z', re.IGNORECASE | re.DOTALL)
def _unwrap_json_content(content: str) -> str:
"""Strip markdown code fences that Ollama vision models often wrap around JSON."""
text = content.strip()
- if text.startswith('```json') and text.endswith('```'):
- return text[7:-3].strip()
- if text.startswith('```') and text.endswith('```'):
- return text[3:-3].strip()
+ match = _JSON_FENCE_RE.fullmatch(text)
+ if match:
+ return match.group('body').strip()
return text
@@ -72,26 +74,28 @@ class ChatOllama(BaseChatModel):
def name(self) -> str:
return self.model
- def _chat_options(self) -> Mapping[str, Any] | Options | None:
- """Return ollama_options with top-level chat keys removed.
+ def _split_chat_options(self) -> tuple[Mapping[str, Any] | Options | None, dict[str, Any]]:
+ """Split model options from supported top-level ``chat()`` parameters.
- `format` and `stream` are parameters of `AsyncClient.chat()`, not model
- runtime options. Putting them in `ollama_options` is ignored or fights
- the structured-output schema we already pass as `format=`.
+ ``format`` and ``stream`` cannot be honored here because this wrapper owns
+ the structured-output schema and requires a non-streaming response.
"""
options = self.ollama_options
if not options or not isinstance(options, Mapping):
- return options
+ return options, {}
- dropped = sorted(key for key in options if key in _TOP_LEVEL_CHAT_KEYS)
- if not dropped:
- return options
+ top_level = {key: options[key] for key in _PASSTHROUGH_CHAT_KEYS if key in options}
+ ignored = sorted(key for key in options if key in _IGNORED_CHAT_KEYS)
- logger.warning(
- 'Ignoring %s in ollama_options; those are top-level chat() parameters, not model options',
- ', '.join(dropped),
- )
- return {key: value for key, value in options.items() if key not in _TOP_LEVEL_CHAT_KEYS}
+ if ignored:
+ logger.warning(
+ 'Ignoring %s in ollama_options; ChatOllama controls structured output and streaming',
+ ', '.join(ignored),
+ )
+
+ extracted = _PASSTHROUGH_CHAT_KEYS | _IGNORED_CHAT_KEYS
+ model_options = {key: value for key, value in options.items() if key not in extracted}
+ return model_options, top_level
@overload
async def ainvoke(
@@ -107,12 +111,13 @@ class ChatOllama(BaseChatModel):
ollama_messages = OllamaMessageSerializer.serialize_messages(messages)
try:
- options = self._chat_options()
+ options, top_level = self._split_chat_options()
if output_format is None:
response = await self.get_client().chat(
model=self.model,
messages=ollama_messages,
options=options,
+ **top_level,
)
return ChatInvokeCompletion(completion=response.message.content or '', usage=None)
@@ -123,6 +128,7 @@ class ChatOllama(BaseChatModel):
messages=ollama_messages,
format=schema,
options=options,
+ **top_level,
)
completion = _unwrap_json_content(response.message.content or '')
diff --git a/tests/ci/models/test_llm_ollama.py b/tests/ci/models/test_llm_ollama.py
index d82c3e045..6c96b6f9e 100644
--- a/tests/ci/models/test_llm_ollama.py
+++ b/tests/ci/models/test_llm_ollama.py
@@ -20,12 +20,20 @@ def _client_returning(content: str) -> MagicMock:
return client
-async def test_drops_format_and_stream_from_ollama_options():
- """format/stream belong on chat(), not inside options (#5017)."""
+async def test_splits_top_level_chat_parameters_from_ollama_options():
+ """Top-level chat parameters must not be sent inside model options (#5017)."""
client = _client_returning('{"answer": "ok"}')
llm = ChatOllama(
model='test-model',
- ollama_options={'think': False, 'format': 'json', 'stream': False, 'num_ctx': 2048},
+ ollama_options={
+ 'think': False,
+ 'logprobs': True,
+ 'top_logprobs': 3,
+ 'keep_alive': '10m',
+ 'format': 'json',
+ 'stream': False,
+ 'num_ctx': 2048,
+ },
)
with patch.object(ChatOllama, 'get_client', return_value=client):
@@ -33,12 +41,18 @@ async def test_drops_format_and_stream_from_ollama_options():
assert result.completion.answer == 'ok'
kwargs = client.chat.await_args.kwargs
- assert kwargs['options'] == {'think': False, 'num_ctx': 2048}
+ assert kwargs['options'] == {'num_ctx': 2048}
+ assert kwargs['think'] is False
+ assert kwargs['logprobs'] is True
+ assert kwargs['top_logprobs'] == 3
+ assert kwargs['keep_alive'] == '10m'
assert kwargs['format'] == Answer.model_json_schema()
+ assert kwargs.get('stream') is None
-async def test_parses_json_wrapped_in_markdown_fences():
- client = _client_returning('```json\n{"answer": "ok"}\n```')
[email protected]('fence', ['```json', '```JSON', '``` json', '```'])
+async def test_parses_json_wrapped_in_markdown_fences(fence: str):
+ client = _client_returning(f'{fence}\n{{"answer": "ok"}}\n```')
llm = ChatOllama(model='test-model')
with patch.object(ChatOllama, 'get_client', return_value=client):