pydantic-ai 1.0.12__tar.gz → 1.0.13__tar.gz

{pydantic_ai-1.0.12 → pydantic_ai-1.0.13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pydantic-ai
-Version: 1.0.12
+Version: 1.0.13
 Summary: Agent Framework / shim to use Pydantic with LLMs
 Project-URL: Homepage, https://ai.pydantic.dev
 Project-URL: Source, https://github.com/pydantic/pydantic-ai
@@ -26,13 +26,13 @@ Classifier: Topic :: Internet
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Requires-Python: >=3.10
-Requires-Dist: pydantic-ai-slim[ag-ui,anthropic,bedrock,cli,cohere,evals,google,groq,huggingface,logfire,mcp,mistral,openai,retries,temporal,vertexai]==1.0.12
+Requires-Dist: pydantic-ai-slim[ag-ui,anthropic,bedrock,cli,cohere,evals,google,groq,huggingface,logfire,mcp,mistral,openai,retries,temporal,vertexai]==1.0.13
 Provides-Extra: a2a
 Requires-Dist: fasta2a>=0.4.1; extra == 'a2a'
 Provides-Extra: dbos
-Requires-Dist: pydantic-ai-slim[dbos]==1.0.12; extra == 'dbos'
+Requires-Dist: pydantic-ai-slim[dbos]==1.0.13; extra == 'dbos'
 Provides-Extra: examples
-Requires-Dist: pydantic-ai-examples==1.0.12; extra == 'examples'
+Requires-Dist: pydantic-ai-examples==1.0.13; extra == 'examples'
 Description-Content-Type: text/markdown
 
 <div align="center">

{pydantic_ai-1.0.12 → pydantic_ai-1.0.13}/tests/evals/test_dataset.py

@@ -7,6 +7,7 @@ from pathlib import Path
 from typing import Any
 
 import pytest
+import yaml
 from dirty_equals import HasRepr, IsNumber
 from inline_snapshot import snapshot
 from pydantic import BaseModel, TypeAdapter
@@ -106,7 +107,7 @@ def example_cases() -> list[Case[TaskInput, TaskOutput, TaskMetadata]]:
 def example_dataset(
     example_cases: list[Case[TaskInput, TaskOutput, TaskMetadata]],
 ) -> Dataset[TaskInput, TaskOutput, TaskMetadata]:
-    return Dataset[TaskInput, TaskOutput, TaskMetadata](cases=example_cases)
+    return Dataset[TaskInput, TaskOutput, TaskMetadata](name='example', cases=example_cases)
 
 
 @pytest.fixture
@@ -820,10 +821,29 @@ async def test_serialization_to_yaml(example_dataset: Dataset[TaskInput, TaskOut
     # Test loading back
     loaded_dataset = Dataset[TaskInput, TaskOutput, TaskMetadata].from_file(yaml_path)
     assert len(loaded_dataset.cases) == 2
+    assert loaded_dataset.name == 'example'
     assert loaded_dataset.cases[0].name == 'case1'
     assert loaded_dataset.cases[0].inputs.query == 'What is 2+2?'
 
 
+async def test_deserializing_without_name(
+    example_dataset: Dataset[TaskInput, TaskOutput, TaskMetadata], tmp_path: Path
+):
+    """Test serializing a dataset to YAML."""
+    # Save the dataset
+    yaml_path = tmp_path / 'test_cases.yaml'
+    example_dataset.to_file(yaml_path)
+
+    # Rewrite the file _without_ a name to test deserializing a name-less file
+    obj = yaml.safe_load(yaml_path.read_text())
+    obj.pop('name', None)
+    yaml_path.write_text(yaml.dump(obj))
+
+    # Test loading results in the name coming from the filename stem
+    loaded_dataset = Dataset[TaskInput, TaskOutput, TaskMetadata].from_file(yaml_path)
+    assert loaded_dataset.name == 'test_cases'
+
+
 async def test_serialization_to_json(example_dataset: Dataset[TaskInput, TaskOutput, TaskMetadata], tmp_path: Path):
     """Test serializing a dataset to JSON."""
     json_path = tmp_path / 'test_cases.json'
@@ -855,6 +875,7 @@ def test_serialization_errors(tmp_path: Path):
 async def test_from_text():
     """Test creating a dataset from text."""
     dataset_dict = {
+        'name': 'my dataset',
         'cases': [
             {
                 'name': '1',
@@ -874,6 +895,7 @@ async def test_from_text():
     }
 
     loaded_dataset = Dataset[TaskInput, TaskOutput, TaskMetadata].from_text(json.dumps(dataset_dict))
+    assert loaded_dataset.name == 'my dataset'
     assert loaded_dataset.cases == snapshot(
         [
             Case(
@@ -1241,7 +1263,7 @@ async def test_dataset_evaluate_with_custom_name(example_dataset: Dataset[TaskIn
     async def task(inputs: TaskInput) -> TaskOutput:
         return TaskOutput(answer=inputs.query.upper())
 
-    report = await example_dataset.evaluate(task, name='custom_task')
+    report = await example_dataset.evaluate(task, task_name='custom_task')
     assert report.name == 'custom_task'
 
 
@@ -1491,16 +1513,26 @@ async def test_evaluate_async_logfire(
             (
                 'evaluate {name}',
                 {
-                    'name': 'mock_async_task',
-                    'n_cases': 2,
                     'assertion_pass_rate': 1.0,
-                    'logfire.msg_template': 'evaluate {name}',
-                    'logfire.msg': 'evaluate mock_async_task',
-                    'logfire.span_type': 'span',
+                    'dataset_name': 'example',
+                    'gen_ai.operation.name': 'experiment',
                     'logfire.json_schema': {
+                        'properties': {
+                            'assertion_pass_rate': {},
+                            'dataset_name': {},
+                            'gen_ai.operation.name': {},
+                            'n_cases': {},
+                            'name': {},
+                            'task_name': {},
+                        },
                         'type': 'object',
-                        'properties': {'name': {}, 'n_cases': {}, 'assertion_pass_rate': {}},
                     },
+                    'logfire.msg': 'evaluate mock_async_task',
+                    'logfire.msg_template': 'evaluate {name}',
+                    'logfire.span_type': 'span',
+                    'n_cases': 2,
+                    'name': 'mock_async_task',
+                    'task_name': 'mock_async_task',
                 },
             ),
             (

{pydantic_ai-1.0.12 → pydantic_ai-1.0.13}/tests/models/test_fallback.py

@@ -169,6 +169,7 @@ def test_first_failed_instrumented(capfire: CaptureLogfire) -> None:
                 'attributes': {
                     'model_name': 'fallback:function:failure_response:,function:success_response:',
                     'agent_name': 'agent',
+                    'gen_ai.agent.name': 'agent',
                     'logfire.msg': 'agent run',
                     'logfire.span_type': 'span',
                     'gen_ai.usage.input_tokens': 51,
@@ -268,6 +269,7 @@ async def test_first_failed_instrumented_stream(capfire: CaptureLogfire) -> None
                 'attributes': {
                     'model_name': 'fallback:function::failure_response_stream,function::success_response_stream',
                     'agent_name': 'agent',
+                    'gen_ai.agent.name': 'agent',
                     'logfire.msg': 'agent run',
                     'logfire.span_type': 'span',
                     'gen_ai.usage.input_tokens': 50,
@@ -375,6 +377,7 @@ def test_all_failed_instrumented(capfire: CaptureLogfire) -> None:
                 'attributes': {
                     'model_name': 'fallback:function:failure_response:,function:failure_response:',
                     'agent_name': 'agent',
+                    'gen_ai.agent.name': 'agent',
                     'logfire.msg': 'agent run',
                     'logfire.span_type': 'span',
                     'pydantic_ai.all_messages': [{'role': 'user', 'parts': [{'type': 'text', 'content': 'hello'}]}],

{pydantic_ai-1.0.12 → pydantic_ai-1.0.13}/tests/models/test_huggingface.py

@@ -893,51 +893,6 @@ text 2\
     )
 
 
-@pytest.mark.parametrize('strict', [True, False, None])
-async def test_tool_strict_mode(allow_model_requests: None, strict: bool | None):
-    tool_call = ChatCompletionOutputToolCall.parse_obj_as_instance(  # type:ignore
-        {
-            'function': ChatCompletionOutputFunctionDefinition.parse_obj_as_instance(  # type:ignore
-                {
-                    'name': 'my_tool',
-                    'arguments': '{"x": 42}',
-                }
-            ),
-            'id': '1',
-            'type': 'function',
-        }
-    )
-    responses = [
-        completion_message(
-            ChatCompletionOutputMessage.parse_obj_as_instance(  # type:ignore
-                {
-                    'content': None,
-                    'role': 'assistant',
-                    'tool_calls': [tool_call],
-                }
-            )
-        ),
-        completion_message(ChatCompletionOutputMessage(content='final response', role='assistant')),  # type: ignore
-    ]
-    mock_client = MockHuggingFace.create_mock(responses)
-    model = HuggingFaceModel('hf-model', provider=HuggingFaceProvider(hf_client=mock_client, api_key='x'))
-    agent = Agent(model)
-
-    @agent.tool_plain(strict=strict)
-    def my_tool(x: int) -> int:
-        return x
-
-    result = await agent.run('hello')
-    assert result.output == 'final response'
-
-    kwargs = get_mock_chat_completion_kwargs(mock_client)[0]
-    tools = kwargs['tools']
-    if strict is not None:
-        assert tools[0]['function']['strict'] is strict
-    else:
-        assert 'strict' not in tools[0]['function']
-
-
 @pytest.mark.parametrize(
     'content_item, error_message',
     [

{pydantic_ai-1.0.12 → pydantic_ai-1.0.13}/tests/models/test_model_settings.py

@@ -2,7 +2,12 @@
 
 from __future__ import annotations
 
-from pydantic_ai import Agent, ModelMessage, ModelResponse, TextPart
+import asyncio
+
+from pydantic_ai import Agent
+from pydantic_ai.direct import model_request as direct_model_request
+from pydantic_ai.messages import ModelMessage, ModelRequest, ModelResponse, TextPart
+from pydantic_ai.models import ModelRequestParameters
 from pydantic_ai.models.function import AgentInfo, FunctionModel
 from pydantic_ai.models.instrumented import InstrumentedModel
 from pydantic_ai.models.test import TestModel
@@ -161,3 +166,39 @@ def test_empty_settings_objects():
     assert captured_settings is not None
     assert captured_settings.get('temperature') == 0.75
     assert len(captured_settings) == 1  # Only one setting should be present
+
+
+def test_direct_model_request_merges_model_settings():
+    """Ensure direct requests merge model defaults with provided run settings."""
+
+    captured_settings = None
+
+    async def capture(messages: list[ModelMessage], agent_info: AgentInfo) -> ModelResponse:
+        nonlocal captured_settings
+        captured_settings = agent_info.model_settings
+        return ModelResponse(parts=[TextPart('ok')])
+
+    model = FunctionModel(
+        capture,
+        settings=ModelSettings(max_tokens=50, temperature=0.3),
+    )
+
+    messages: list[ModelMessage] = [ModelRequest.user_text_prompt('hi')]
+    run_settings = ModelSettings(temperature=0.9, top_p=0.2)
+
+    async def _run() -> ModelResponse:
+        return await direct_model_request(
+            model,
+            messages,
+            model_settings=run_settings,
+            model_request_parameters=ModelRequestParameters(),
+        )
+
+    response = asyncio.run(_run())
+
+    assert response.parts == [TextPart('ok')]
+    assert captured_settings == {
+        'max_tokens': 50,
+        'temperature': 0.9,
+        'top_p': 0.2,
+    }

{pydantic_ai-1.0.12 → pydantic_ai-1.0.13}/tests/test_agent.py

@@ -1,3 +1,4 @@
+import asyncio
 import json
 import re
 import sys
@@ -5148,3 +5149,231 @@ async def test_consecutive_model_responses_in_history():
             ),
         ]
     )
+
+
+def test_override_instructions_basic():
+    """Test that override can override instructions."""
+    agent = Agent('test')
+
+    @agent.instructions
+    def instr_fn() -> str:
+        return 'SHOULD_BE_IGNORED'
+
+    with capture_run_messages() as base_messages:
+        agent.run_sync('Hello', model=TestModel(custom_output_text='baseline'))
+
+    base_req = base_messages[0]
+    assert isinstance(base_req, ModelRequest)
+    assert base_req.instructions == 'SHOULD_BE_IGNORED'
+
+    with agent.override(instructions='OVERRIDE'):
+        with capture_run_messages() as messages:
+            agent.run_sync('Hello', model=TestModel(custom_output_text='ok'))
+
+    req = messages[0]
+    assert isinstance(req, ModelRequest)
+    assert req.instructions == 'OVERRIDE'
+
+
+def test_override_reset_after_context():
+    """Test that instructions are reset after exiting the override context."""
+    agent = Agent('test', instructions='ORIG')
+
+    with agent.override(instructions='NEW'):
+        with capture_run_messages() as messages_new:
+            agent.run_sync('Hi', model=TestModel(custom_output_text='ok'))
+
+    with capture_run_messages() as messages_orig:
+        agent.run_sync('Hi', model=TestModel(custom_output_text='ok'))
+
+    req_new = messages_new[0]
+    assert isinstance(req_new, ModelRequest)
+    req_orig = messages_orig[0]
+    assert isinstance(req_orig, ModelRequest)
+    assert req_new.instructions == 'NEW'
+    assert req_orig.instructions == 'ORIG'
+
+
+def test_override_none_clears_instructions():
+    """Test that passing None for instructions clears all instructions."""
+    agent = Agent('test', instructions='BASE')
+
+    @agent.instructions
+    def instr_fn() -> str:  # pragma: no cover - ignored under override
+        return 'ALSO_BASE'
+
+    with agent.override(instructions=None):
+        with capture_run_messages() as messages:
+            agent.run_sync('Hello', model=TestModel(custom_output_text='ok'))
+
+    req = messages[0]
+    assert isinstance(req, ModelRequest)
+    assert req.instructions is None
+
+
+def test_override_instructions_callable_replaces_functions():
+    """Override with a callable should replace existing instruction functions."""
+    agent = Agent('test')
+
+    @agent.instructions
+    def base_fn() -> str:
+        return 'BASE_FN'
+
+    def override_fn() -> str:
+        return 'OVERRIDE_FN'
+
+    with capture_run_messages() as base_messages:
+        agent.run_sync('Hello', model=TestModel(custom_output_text='baseline'))
+
+    base_req = base_messages[0]
+    assert isinstance(base_req, ModelRequest)
+    assert base_req.instructions is not None
+    assert 'BASE_FN' in base_req.instructions
+
+    with agent.override(instructions=override_fn):
+        with capture_run_messages() as messages:
+            agent.run_sync('Hello', model=TestModel(custom_output_text='ok'))
+
+    req = messages[0]
+    assert isinstance(req, ModelRequest)
+    assert req.instructions == 'OVERRIDE_FN'
+    assert 'BASE_FN' not in req.instructions
+
+
+async def test_override_instructions_async_callable():
+    """Override with an async callable should be awaited."""
+    agent = Agent('test')
+
+    async def override_fn() -> str:
+        await asyncio.sleep(0)
+        return 'ASYNC_FN'
+
+    with agent.override(instructions=override_fn):
+        with capture_run_messages() as messages:
+            await agent.run('Hi', model=TestModel(custom_output_text='ok'))
+
+    req = messages[0]
+    assert isinstance(req, ModelRequest)
+    assert req.instructions == 'ASYNC_FN'
+
+
+def test_override_instructions_sequence_mixed_types():
+    """Override can mix literal strings and functions."""
+    agent = Agent('test', instructions='BASE')
+
+    def override_fn() -> str:
+        return 'FUNC_PART'
+
+    def override_fn_2() -> str:
+        return 'FUNC_PART_2'
+
+    with agent.override(instructions=['OVERRIDE1', override_fn, 'OVERRIDE2', override_fn_2]):
+        with capture_run_messages() as messages:
+            agent.run_sync('Hello', model=TestModel(custom_output_text='ok'))
+
+    req = messages[0]
+    assert isinstance(req, ModelRequest)
+    assert req.instructions == 'OVERRIDE1\nOVERRIDE2\n\nFUNC_PART\n\nFUNC_PART_2'
+    assert 'BASE' not in req.instructions
+
+
+async def test_override_concurrent_isolation():
+    """Test that concurrent overrides are isolated from each other."""
+    agent = Agent('test', instructions='ORIG')
+
+    async def run_with(instr: str) -> str | None:
+        with agent.override(instructions=instr):
+            with capture_run_messages() as messages:
+                await agent.run('Hi', model=TestModel(custom_output_text='ok'))
+            req = messages[0]
+            assert isinstance(req, ModelRequest)
+            return req.instructions
+
+    a, b = await asyncio.gather(
+        run_with('A'),
+        run_with('B'),
+    )
+
+    assert a == 'A'
+    assert b == 'B'
+
+
+def test_override_replaces_instructions():
+    """Test overriding instructions replaces the base instructions."""
+    agent = Agent('test', instructions='ORIG_INSTR')
+
+    with agent.override(instructions='NEW_INSTR'):
+        with capture_run_messages() as messages:
+            agent.run_sync('Hi', model=TestModel(custom_output_text='ok'))
+
+    req = messages[0]
+    assert isinstance(req, ModelRequest)
+    assert req.instructions == 'NEW_INSTR'
+
+
+def test_override_nested_contexts():
+    """Test nested override contexts."""
+    agent = Agent('test', instructions='ORIG')
+
+    with agent.override(instructions='OUTER'):
+        with capture_run_messages() as outer_messages:
+            agent.run_sync('Hi', model=TestModel(custom_output_text='ok'))
+
+        with agent.override(instructions='INNER'):
+            with capture_run_messages() as inner_messages:
+                agent.run_sync('Hi', model=TestModel(custom_output_text='ok'))
+
+    outer_req = outer_messages[0]
+    assert isinstance(outer_req, ModelRequest)
+    inner_req = inner_messages[0]
+    assert isinstance(inner_req, ModelRequest)
+
+    assert outer_req.instructions == 'OUTER'
+    assert inner_req.instructions == 'INNER'
+
+
+async def test_override_async_run():
+    """Test override with async run method."""
+    agent = Agent('test', instructions='ORIG')
+
+    with agent.override(instructions='ASYNC_OVERRIDE'):
+        with capture_run_messages() as messages:
+            await agent.run('Hi', model=TestModel(custom_output_text='ok'))
+
+    req = messages[0]
+    assert isinstance(req, ModelRequest)
+    assert req.instructions == 'ASYNC_OVERRIDE'
+
+
+def test_override_with_dynamic_prompts():
+    """Test override interacting with dynamic prompts."""
+    agent = Agent('test')
+
+    dynamic_value = 'DYNAMIC'
+
+    @agent.system_prompt
+    def dynamic_sys() -> str:
+        return dynamic_value
+
+    @agent.instructions
+    def dynamic_instr() -> str:
+        return 'DYNAMIC_INSTR'
+
+    with capture_run_messages() as base_messages:
+        agent.run_sync('Hi', model=TestModel(custom_output_text='baseline'))
+
+    base_req = base_messages[0]
+    assert isinstance(base_req, ModelRequest)
+    assert base_req.instructions == 'DYNAMIC_INSTR'
+
+    # Override should take precedence over dynamic instructions but leave system prompts intact
+    with agent.override(instructions='OVERRIDE_INSTR'):
+        with capture_run_messages() as messages:
+            agent.run_sync('Hi', model=TestModel(custom_output_text='ok'))
+
+    req = messages[0]
+    assert isinstance(req, ModelRequest)
+    assert req.instructions == 'OVERRIDE_INSTR'
+    sys_texts = [p.content for p in req.parts if isinstance(p, SystemPromptPart)]
+    # The dynamic system prompt should still be present since overrides target instructions only
+    assert dynamic_value in sys_texts

{pydantic_ai-1.0.12 → pydantic_ai-1.0.13}/tests/test_direct.py

@@ -7,8 +7,16 @@ from unittest.mock import AsyncMock, patch
 import pytest
 from inline_snapshot import snapshot
 
-from pydantic_ai import (
-    Agent,
+from pydantic_ai import Agent
+from pydantic_ai.direct import (
+    StreamedResponseSync,
+    _prepare_model,  # pyright: ignore[reportPrivateUsage]
+    model_request,
+    model_request_stream,
+    model_request_stream_sync,
+    model_request_sync,
+)
+from pydantic_ai.messages import (
     FinalResultEvent,
     ModelMessage,
     ModelRequest,
@@ -19,14 +27,6 @@ from pydantic_ai import (
     TextPartDelta,
     ToolCallPart,
 )
-from pydantic_ai.direct import (
-    StreamedResponseSync,
-    _prepare_model,  # pyright: ignore[reportPrivateUsage]
-    model_request,
-    model_request_stream,
-    model_request_stream_sync,
-    model_request_sync,
-)
 from pydantic_ai.models import ModelRequestParameters
 from pydantic_ai.models.instrumented import InstrumentedModel
 from pydantic_ai.models.test import TestModel