deepeval 3.6.7__py3-none-any.whl → 3.6.8__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- deepeval/_version.py +1 -1
- deepeval/errors.py +20 -2
- deepeval/evaluate/execute.py +725 -217
- deepeval/evaluate/types.py +1 -0
- deepeval/evaluate/utils.py +13 -3
- deepeval/integrations/crewai/__init__.py +2 -1
- deepeval/integrations/crewai/tool.py +71 -0
- deepeval/integrations/llama_index/__init__.py +0 -4
- deepeval/integrations/llama_index/handler.py +20 -21
- deepeval/integrations/pydantic_ai/instrumentator.py +125 -76
- deepeval/metrics/__init__.py +13 -0
- deepeval/metrics/base_metric.py +1 -0
- deepeval/metrics/contextual_precision/contextual_precision.py +27 -21
- deepeval/metrics/conversational_g_eval/__init__.py +3 -0
- deepeval/metrics/conversational_g_eval/conversational_g_eval.py +11 -7
- deepeval/metrics/dag/schema.py +1 -1
- deepeval/metrics/dag/templates.py +2 -2
- deepeval/metrics/goal_accuracy/__init__.py +1 -0
- deepeval/metrics/goal_accuracy/goal_accuracy.py +349 -0
- deepeval/metrics/goal_accuracy/schema.py +17 -0
- deepeval/metrics/goal_accuracy/template.py +235 -0
- deepeval/metrics/hallucination/hallucination.py +8 -8
- deepeval/metrics/mcp/mcp_task_completion.py +7 -2
- deepeval/metrics/mcp/multi_turn_mcp_use_metric.py +16 -6
- deepeval/metrics/mcp_use_metric/mcp_use_metric.py +2 -1
- deepeval/metrics/multimodal_metrics/multimodal_contextual_precision/multimodal_contextual_precision.py +32 -24
- deepeval/metrics/plan_adherence/__init__.py +1 -0
- deepeval/metrics/plan_adherence/plan_adherence.py +292 -0
- deepeval/metrics/plan_adherence/schema.py +11 -0
- deepeval/metrics/plan_adherence/template.py +170 -0
- deepeval/metrics/plan_quality/__init__.py +1 -0
- deepeval/metrics/plan_quality/plan_quality.py +292 -0
- deepeval/metrics/plan_quality/schema.py +11 -0
- deepeval/metrics/plan_quality/template.py +101 -0
- deepeval/metrics/step_efficiency/__init__.py +1 -0
- deepeval/metrics/step_efficiency/schema.py +11 -0
- deepeval/metrics/step_efficiency/step_efficiency.py +234 -0
- deepeval/metrics/step_efficiency/template.py +256 -0
- deepeval/metrics/task_completion/task_completion.py +1 -0
- deepeval/metrics/tool_correctness/schema.py +6 -0
- deepeval/metrics/tool_correctness/template.py +88 -0
- deepeval/metrics/tool_correctness/tool_correctness.py +226 -22
- deepeval/metrics/tool_use/__init__.py +1 -0
- deepeval/metrics/tool_use/schema.py +19 -0
- deepeval/metrics/tool_use/template.py +220 -0
- deepeval/metrics/tool_use/tool_use.py +458 -0
- deepeval/metrics/topic_adherence/__init__.py +1 -0
- deepeval/metrics/topic_adherence/schema.py +16 -0
- deepeval/metrics/topic_adherence/template.py +162 -0
- deepeval/metrics/topic_adherence/topic_adherence.py +355 -0
- deepeval/models/embedding_models/azure_embedding_model.py +37 -36
- deepeval/models/embedding_models/local_embedding_model.py +30 -32
- deepeval/models/embedding_models/ollama_embedding_model.py +18 -20
- deepeval/models/embedding_models/openai_embedding_model.py +22 -31
- deepeval/openai/extractors.py +61 -16
- deepeval/openai/patch.py +8 -12
- deepeval/openai/types.py +1 -1
- deepeval/openai/utils.py +108 -1
- deepeval/prompt/prompt.py +1 -0
- deepeval/prompt/utils.py +43 -14
- deepeval/synthesizer/synthesizer.py +11 -10
- deepeval/test_case/llm_test_case.py +6 -2
- deepeval/test_run/test_run.py +190 -207
- deepeval/tracing/__init__.py +2 -1
- deepeval/tracing/otel/exporter.py +3 -4
- deepeval/tracing/otel/utils.py +23 -4
- deepeval/tracing/trace_context.py +53 -38
- deepeval/tracing/tracing.py +23 -0
- deepeval/tracing/types.py +16 -14
- deepeval/utils.py +21 -0
- {deepeval-3.6.7.dist-info → deepeval-3.6.8.dist-info}/METADATA +1 -1
- {deepeval-3.6.7.dist-info → deepeval-3.6.8.dist-info}/RECORD +75 -53
- deepeval/integrations/llama_index/agent/patched.py +0 -68
- deepeval/tracing/message_types/__init__.py +0 -10
- deepeval/tracing/message_types/base.py +0 -6
- deepeval/tracing/message_types/messages.py +0 -14
- deepeval/tracing/message_types/tools.py +0 -18
- {deepeval-3.6.7.dist-info → deepeval-3.6.8.dist-info}/LICENSE.md +0 -0
- {deepeval-3.6.7.dist-info → deepeval-3.6.8.dist-info}/WHEEL +0 -0
- {deepeval-3.6.7.dist-info → deepeval-3.6.8.dist-info}/entry_points.txt +0 -0
|
@@ -1,14 +0,0 @@
|
|
|
1
|
-
from typing import Literal, Dict, Any
|
|
2
|
-
from .base import BaseMessage
|
|
3
|
-
|
|
4
|
-
|
|
5
|
-
class TextMessage(BaseMessage):
|
|
6
|
-
type: Literal["text", "thinking"]
|
|
7
|
-
content: str
|
|
8
|
-
|
|
9
|
-
|
|
10
|
-
class ToolCallMessage(BaseMessage):
|
|
11
|
-
"""This is a message for tool calls in response.choices[0].message.tool_calls"""
|
|
12
|
-
|
|
13
|
-
name: str
|
|
14
|
-
args: Dict[str, Any]
|
|
@@ -1,18 +0,0 @@
|
|
|
1
|
-
from typing import Any, Optional, Dict
|
|
2
|
-
from pydantic import BaseModel
|
|
3
|
-
|
|
4
|
-
|
|
5
|
-
class BaseTool(BaseModel):
|
|
6
|
-
name: str
|
|
7
|
-
description: Optional[str] = None
|
|
8
|
-
|
|
9
|
-
|
|
10
|
-
class ToolSchema(BaseTool):
|
|
11
|
-
parameters: Dict[str, Any]
|
|
12
|
-
is_called: Optional[bool] = False
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
class ToolOutput(BaseTool):
|
|
16
|
-
"""Output of the tool function"""
|
|
17
|
-
|
|
18
|
-
output: Any
|
|
File without changes
|
|
File without changes
|
|
File without changes
|