azure-ai-evaluation 1.2.0__py3-none-any.whl → 1.4.0__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.
Potentially problematic release.
This version of azure-ai-evaluation might be problematic. Click here for more details.
- azure/ai/evaluation/__init__.py +42 -14
- azure/ai/evaluation/_azure/_models.py +6 -6
- azure/ai/evaluation/_common/constants.py +6 -2
- azure/ai/evaluation/_common/rai_service.py +38 -4
- azure/ai/evaluation/_common/raiclient/__init__.py +34 -0
- azure/ai/evaluation/_common/raiclient/_client.py +128 -0
- azure/ai/evaluation/_common/raiclient/_configuration.py +87 -0
- azure/ai/evaluation/_common/raiclient/_model_base.py +1235 -0
- azure/ai/evaluation/_common/raiclient/_patch.py +20 -0
- azure/ai/evaluation/_common/raiclient/_serialization.py +2050 -0
- azure/ai/evaluation/_common/raiclient/_version.py +9 -0
- azure/ai/evaluation/_common/raiclient/aio/__init__.py +29 -0
- azure/ai/evaluation/_common/raiclient/aio/_client.py +130 -0
- azure/ai/evaluation/_common/raiclient/aio/_configuration.py +87 -0
- azure/ai/evaluation/_common/raiclient/aio/_patch.py +20 -0
- azure/ai/evaluation/_common/raiclient/aio/operations/__init__.py +25 -0
- azure/ai/evaluation/_common/raiclient/aio/operations/_operations.py +981 -0
- azure/ai/evaluation/_common/raiclient/aio/operations/_patch.py +20 -0
- azure/ai/evaluation/_common/raiclient/models/__init__.py +60 -0
- azure/ai/evaluation/_common/raiclient/models/_enums.py +18 -0
- azure/ai/evaluation/_common/raiclient/models/_models.py +651 -0
- azure/ai/evaluation/_common/raiclient/models/_patch.py +20 -0
- azure/ai/evaluation/_common/raiclient/operations/__init__.py +25 -0
- azure/ai/evaluation/_common/raiclient/operations/_operations.py +1225 -0
- azure/ai/evaluation/_common/raiclient/operations/_patch.py +20 -0
- azure/ai/evaluation/_common/raiclient/py.typed +1 -0
- azure/ai/evaluation/_common/utils.py +30 -10
- azure/ai/evaluation/_constants.py +10 -0
- azure/ai/evaluation/_converters/__init__.py +3 -0
- azure/ai/evaluation/_converters/_ai_services.py +804 -0
- azure/ai/evaluation/_converters/_models.py +302 -0
- azure/ai/evaluation/_evaluate/_batch_run/__init__.py +10 -3
- azure/ai/evaluation/_evaluate/_batch_run/_run_submitter_client.py +104 -0
- azure/ai/evaluation/_evaluate/_batch_run/batch_clients.py +82 -0
- azure/ai/evaluation/_evaluate/_eval_run.py +1 -1
- azure/ai/evaluation/_evaluate/_evaluate.py +36 -4
- azure/ai/evaluation/_evaluators/_bleu/_bleu.py +23 -3
- azure/ai/evaluation/_evaluators/_code_vulnerability/__init__.py +5 -0
- azure/ai/evaluation/_evaluators/_code_vulnerability/_code_vulnerability.py +120 -0
- azure/ai/evaluation/_evaluators/_coherence/_coherence.py +21 -2
- azure/ai/evaluation/_evaluators/_common/_base_eval.py +43 -3
- azure/ai/evaluation/_evaluators/_common/_base_multi_eval.py +3 -1
- azure/ai/evaluation/_evaluators/_common/_base_prompty_eval.py +43 -4
- azure/ai/evaluation/_evaluators/_common/_base_rai_svc_eval.py +16 -4
- azure/ai/evaluation/_evaluators/_content_safety/_content_safety.py +42 -5
- azure/ai/evaluation/_evaluators/_content_safety/_hate_unfairness.py +15 -0
- azure/ai/evaluation/_evaluators/_content_safety/_self_harm.py +15 -0
- azure/ai/evaluation/_evaluators/_content_safety/_sexual.py +15 -0
- azure/ai/evaluation/_evaluators/_content_safety/_violence.py +15 -0
- azure/ai/evaluation/_evaluators/_f1_score/_f1_score.py +28 -4
- azure/ai/evaluation/_evaluators/_fluency/_fluency.py +21 -2
- azure/ai/evaluation/_evaluators/_gleu/_gleu.py +26 -3
- azure/ai/evaluation/_evaluators/_groundedness/_groundedness.py +21 -3
- azure/ai/evaluation/_evaluators/_intent_resolution/__init__.py +7 -0
- azure/ai/evaluation/_evaluators/_intent_resolution/_intent_resolution.py +152 -0
- azure/ai/evaluation/_evaluators/_intent_resolution/intent_resolution.prompty +161 -0
- azure/ai/evaluation/_evaluators/_meteor/_meteor.py +26 -3
- azure/ai/evaluation/_evaluators/_qa/_qa.py +51 -7
- azure/ai/evaluation/_evaluators/_relevance/_relevance.py +26 -2
- azure/ai/evaluation/_evaluators/_response_completeness/__init__.py +7 -0
- azure/ai/evaluation/_evaluators/_response_completeness/_response_completeness.py +157 -0
- azure/ai/evaluation/_evaluators/_response_completeness/response_completeness.prompty +99 -0
- azure/ai/evaluation/_evaluators/_retrieval/_retrieval.py +21 -2
- azure/ai/evaluation/_evaluators/_rouge/_rouge.py +113 -4
- azure/ai/evaluation/_evaluators/_service_groundedness/_service_groundedness.py +23 -3
- azure/ai/evaluation/_evaluators/_similarity/_similarity.py +24 -5
- azure/ai/evaluation/_evaluators/_task_adherence/__init__.py +7 -0
- azure/ai/evaluation/_evaluators/_task_adherence/_task_adherence.py +148 -0
- azure/ai/evaluation/_evaluators/_task_adherence/task_adherence.prompty +117 -0
- azure/ai/evaluation/_evaluators/_tool_call_accuracy/__init__.py +9 -0
- azure/ai/evaluation/_evaluators/_tool_call_accuracy/_tool_call_accuracy.py +292 -0
- azure/ai/evaluation/_evaluators/_tool_call_accuracy/tool_call_accuracy.prompty +71 -0
- azure/ai/evaluation/_evaluators/_ungrounded_attributes/__init__.py +5 -0
- azure/ai/evaluation/_evaluators/_ungrounded_attributes/_ungrounded_attributes.py +103 -0
- azure/ai/evaluation/_evaluators/_xpia/xpia.py +2 -0
- azure/ai/evaluation/_exceptions.py +5 -1
- azure/ai/evaluation/_legacy/__init__.py +3 -0
- azure/ai/evaluation/_legacy/_batch_engine/__init__.py +9 -0
- azure/ai/evaluation/_legacy/_batch_engine/_config.py +45 -0
- azure/ai/evaluation/_legacy/_batch_engine/_engine.py +368 -0
- azure/ai/evaluation/_legacy/_batch_engine/_exceptions.py +88 -0
- azure/ai/evaluation/_legacy/_batch_engine/_logging.py +292 -0
- azure/ai/evaluation/_legacy/_batch_engine/_openai_injector.py +23 -0
- azure/ai/evaluation/_legacy/_batch_engine/_result.py +99 -0
- azure/ai/evaluation/_legacy/_batch_engine/_run.py +121 -0
- azure/ai/evaluation/_legacy/_batch_engine/_run_storage.py +128 -0
- azure/ai/evaluation/_legacy/_batch_engine/_run_submitter.py +217 -0
- azure/ai/evaluation/_legacy/_batch_engine/_status.py +25 -0
- azure/ai/evaluation/_legacy/_batch_engine/_trace.py +105 -0
- azure/ai/evaluation/_legacy/_batch_engine/_utils.py +82 -0
- azure/ai/evaluation/_legacy/_batch_engine/_utils_deprecated.py +131 -0
- azure/ai/evaluation/_legacy/prompty/__init__.py +36 -0
- azure/ai/evaluation/_legacy/prompty/_connection.py +182 -0
- azure/ai/evaluation/_legacy/prompty/_exceptions.py +59 -0
- azure/ai/evaluation/_legacy/prompty/_prompty.py +313 -0
- azure/ai/evaluation/_legacy/prompty/_utils.py +545 -0
- azure/ai/evaluation/_legacy/prompty/_yaml_utils.py +99 -0
- azure/ai/evaluation/_red_team/__init__.py +3 -0
- azure/ai/evaluation/_red_team/_attack_objective_generator.py +192 -0
- azure/ai/evaluation/_red_team/_attack_strategy.py +42 -0
- azure/ai/evaluation/_red_team/_callback_chat_target.py +74 -0
- azure/ai/evaluation/_red_team/_default_converter.py +21 -0
- azure/ai/evaluation/_red_team/_red_team.py +1858 -0
- azure/ai/evaluation/_red_team/_red_team_result.py +246 -0
- azure/ai/evaluation/_red_team/_utils/__init__.py +3 -0
- azure/ai/evaluation/_red_team/_utils/constants.py +64 -0
- azure/ai/evaluation/_red_team/_utils/formatting_utils.py +164 -0
- azure/ai/evaluation/_red_team/_utils/logging_utils.py +139 -0
- azure/ai/evaluation/_red_team/_utils/strategy_utils.py +188 -0
- azure/ai/evaluation/_safety_evaluation/__init__.py +3 -0
- azure/ai/evaluation/_safety_evaluation/_generated_rai_client.py +0 -0
- azure/ai/evaluation/_safety_evaluation/_safety_evaluation.py +741 -0
- azure/ai/evaluation/_version.py +2 -1
- azure/ai/evaluation/simulator/_adversarial_scenario.py +3 -1
- azure/ai/evaluation/simulator/_adversarial_simulator.py +61 -27
- azure/ai/evaluation/simulator/_conversation/__init__.py +4 -5
- azure/ai/evaluation/simulator/_conversation/_conversation.py +4 -0
- azure/ai/evaluation/simulator/_model_tools/_generated_rai_client.py +145 -0
- azure/ai/evaluation/simulator/_model_tools/_proxy_completion_model.py +2 -0
- azure/ai/evaluation/simulator/_model_tools/_rai_client.py +71 -1
- {azure_ai_evaluation-1.2.0.dist-info → azure_ai_evaluation-1.4.0.dist-info}/METADATA +75 -15
- azure_ai_evaluation-1.4.0.dist-info/RECORD +197 -0
- {azure_ai_evaluation-1.2.0.dist-info → azure_ai_evaluation-1.4.0.dist-info}/WHEEL +1 -1
- azure/ai/evaluation/_evaluators/_multimodal/__init__.py +0 -20
- azure/ai/evaluation/_evaluators/_multimodal/_content_safety_multimodal.py +0 -132
- azure/ai/evaluation/_evaluators/_multimodal/_content_safety_multimodal_base.py +0 -55
- azure/ai/evaluation/_evaluators/_multimodal/_hate_unfairness.py +0 -100
- azure/ai/evaluation/_evaluators/_multimodal/_protected_material.py +0 -124
- azure/ai/evaluation/_evaluators/_multimodal/_self_harm.py +0 -100
- azure/ai/evaluation/_evaluators/_multimodal/_sexual.py +0 -100
- azure/ai/evaluation/_evaluators/_multimodal/_violence.py +0 -100
- azure_ai_evaluation-1.2.0.dist-info/RECORD +0 -125
- {azure_ai_evaluation-1.2.0.dist-info → azure_ai_evaluation-1.4.0.dist-info}/NOTICE.txt +0 -0
- {azure_ai_evaluation-1.2.0.dist-info → azure_ai_evaluation-1.4.0.dist-info}/top_level.txt +0 -0
|
@@ -0,0 +1,36 @@
|
|
|
1
|
+
# ---------------------------------------------------------
|
|
2
|
+
# Copyright (c) Microsoft Corporation. All rights reserved.
|
|
3
|
+
# ---------------------------------------------------------
|
|
4
|
+
|
|
5
|
+
from azure.ai.evaluation._legacy.prompty._prompty import AsyncPrompty
|
|
6
|
+
from azure.ai.evaluation._legacy.prompty._connection import Connection, OpenAIConnection, AzureOpenAIConnection
|
|
7
|
+
from azure.ai.evaluation._legacy.prompty._exceptions import (
|
|
8
|
+
PromptyException,
|
|
9
|
+
MissingRequiredInputError,
|
|
10
|
+
InvalidInputError,
|
|
11
|
+
JinjaTemplateError,
|
|
12
|
+
NotSupportedError,
|
|
13
|
+
)
|
|
14
|
+
|
|
15
|
+
# =========================================================================================================
|
|
16
|
+
# NOTE: All of the code here is largely copy of code from Promptflow. Generally speaking, the following
|
|
17
|
+
# changes were made:
|
|
18
|
+
# - Added type annotations
|
|
19
|
+
# - Legacy or deprecated functionality has been removed (e.g. no more support for completions API)
|
|
20
|
+
# - Reworked the way images are handled to 1) Reduce the amount of code brought over, 2) remove
|
|
21
|
+
# the need to do two passes over the template to insert images, 3) remove the completely unnecessary
|
|
22
|
+
# loading of image data from the internet when it is not actually needed
|
|
23
|
+
# - Minor obvious tweaks to improve code readability, and removal of unused code paths
|
|
24
|
+
# =========================================================================================================
|
|
25
|
+
|
|
26
|
+
__all__ = [
|
|
27
|
+
"AsyncPrompty",
|
|
28
|
+
"Connection",
|
|
29
|
+
"AzureOpenAIConnection",
|
|
30
|
+
"OpenAIConnection",
|
|
31
|
+
"PromptyException",
|
|
32
|
+
"MissingRequiredInputError",
|
|
33
|
+
"InvalidInputError",
|
|
34
|
+
"JinjaTemplateError",
|
|
35
|
+
"NotSupportedError",
|
|
36
|
+
]
|
|
@@ -0,0 +1,182 @@
|
|
|
1
|
+
# ---------------------------------------------------------
|
|
2
|
+
# Copyright (c) Microsoft Corporation. All rights reserved.
|
|
3
|
+
# ---------------------------------------------------------
|
|
4
|
+
|
|
5
|
+
import os
|
|
6
|
+
import re
|
|
7
|
+
from abc import ABC, abstractmethod
|
|
8
|
+
from dataclasses import dataclass
|
|
9
|
+
from typing import Any, ClassVar, Mapping, Optional, Set, Union
|
|
10
|
+
|
|
11
|
+
from azure.ai.evaluation._legacy.prompty._exceptions import MissingRequiredInputError
|
|
12
|
+
from azure.ai.evaluation._legacy.prompty._utils import dataclass_from_dict
|
|
13
|
+
|
|
14
|
+
|
|
15
|
+
ENV_VAR_PATTERN = re.compile(r"^\$\{env:(.*)\}$")
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
def _parse_environment_variable(value: Union[str, Any]) -> Union[str, Any]:
|
|
19
|
+
"""Get environment variable from ${env:ENV_NAME}. If not found, return original value.
|
|
20
|
+
|
|
21
|
+
:param value: The value to parse.
|
|
22
|
+
:type value: str | Any
|
|
23
|
+
:return: The parsed value
|
|
24
|
+
:rtype: str | Any"""
|
|
25
|
+
if not isinstance(value, str):
|
|
26
|
+
return value
|
|
27
|
+
|
|
28
|
+
result = re.match(ENV_VAR_PATTERN, value)
|
|
29
|
+
if result:
|
|
30
|
+
env_name = result.groups()[0]
|
|
31
|
+
return os.environ.get(env_name, value)
|
|
32
|
+
|
|
33
|
+
return value
|
|
34
|
+
|
|
35
|
+
|
|
36
|
+
def _is_empty_connection_config(connection_dict: Mapping[str, Any]) -> bool:
|
|
37
|
+
ignored_fields = set(["azure_deployment", "model"])
|
|
38
|
+
keys = {k for k, v in connection_dict.items() if v}
|
|
39
|
+
return len(keys - ignored_fields) == 0
|
|
40
|
+
|
|
41
|
+
|
|
42
|
+
@dataclass
|
|
43
|
+
class Connection(ABC):
|
|
44
|
+
"""Base class for all connection classes."""
|
|
45
|
+
|
|
46
|
+
@property
|
|
47
|
+
@abstractmethod
|
|
48
|
+
def type(self) -> str:
|
|
49
|
+
"""Gets the type of the connection.
|
|
50
|
+
|
|
51
|
+
:return: The type of the connection.
|
|
52
|
+
:rtype: str"""
|
|
53
|
+
...
|
|
54
|
+
|
|
55
|
+
@abstractmethod
|
|
56
|
+
def is_valid(self, missing_fields: Optional[Set[str]] = None) -> bool:
|
|
57
|
+
"""Check if the connection is valid.
|
|
58
|
+
|
|
59
|
+
:param missing_fields: If set, this will be populated with the missing required fields.
|
|
60
|
+
:type missing_fields: Set[str] | None
|
|
61
|
+
:return: True if the connection is valid, False otherwise.
|
|
62
|
+
:rtype: bool"""
|
|
63
|
+
...
|
|
64
|
+
|
|
65
|
+
@staticmethod
|
|
66
|
+
def parse_from_config(model_configuration: Mapping[str, Any]) -> "Connection":
|
|
67
|
+
"""Parse a connection from a model configuration.
|
|
68
|
+
|
|
69
|
+
:param model_configuration: The model configuration.
|
|
70
|
+
:type model_configuration: Mapping[str, Any]
|
|
71
|
+
:return: The connection.
|
|
72
|
+
:rtype: Connection
|
|
73
|
+
"""
|
|
74
|
+
connection: Connection
|
|
75
|
+
connection_dict = {k: _parse_environment_variable(v) for k, v in model_configuration.items()}
|
|
76
|
+
connection_type = connection_dict.pop("type", "")
|
|
77
|
+
|
|
78
|
+
if connection_type in [AzureOpenAIConnection.TYPE, "azure_openai"]:
|
|
79
|
+
if _is_empty_connection_config(connection_dict):
|
|
80
|
+
connection = AzureOpenAIConnection.from_env()
|
|
81
|
+
else:
|
|
82
|
+
connection = dataclass_from_dict(AzureOpenAIConnection, connection_dict)
|
|
83
|
+
|
|
84
|
+
elif connection_type in [OpenAIConnection.TYPE, "openai"]:
|
|
85
|
+
if _is_empty_connection_config(connection_dict):
|
|
86
|
+
connection = OpenAIConnection.from_env()
|
|
87
|
+
else:
|
|
88
|
+
connection = dataclass_from_dict(OpenAIConnection, connection_dict)
|
|
89
|
+
|
|
90
|
+
else:
|
|
91
|
+
error_message = (
|
|
92
|
+
f"'{connection_type}' is not a supported connection type. Valid values are "
|
|
93
|
+
f"[{AzureOpenAIConnection.TYPE}, {OpenAIConnection.TYPE}]"
|
|
94
|
+
)
|
|
95
|
+
raise MissingRequiredInputError(error_message)
|
|
96
|
+
|
|
97
|
+
missing_fields: Set[str] = set()
|
|
98
|
+
if not connection.is_valid(missing_fields):
|
|
99
|
+
raise MissingRequiredInputError(
|
|
100
|
+
f"The following required fields are missing for connection {connection.type}: "
|
|
101
|
+
f"{', '.join(missing_fields)}"
|
|
102
|
+
)
|
|
103
|
+
|
|
104
|
+
return connection
|
|
105
|
+
|
|
106
|
+
|
|
107
|
+
@dataclass
|
|
108
|
+
class OpenAIConnection(Connection):
|
|
109
|
+
"""Connection class for OpenAI endpoints."""
|
|
110
|
+
|
|
111
|
+
base_url: str
|
|
112
|
+
api_key: Optional[str] = None
|
|
113
|
+
organization: Optional[str] = None
|
|
114
|
+
|
|
115
|
+
TYPE: ClassVar[str] = "openai"
|
|
116
|
+
|
|
117
|
+
@property
|
|
118
|
+
def type(self) -> str:
|
|
119
|
+
return OpenAIConnection.TYPE
|
|
120
|
+
|
|
121
|
+
@classmethod
|
|
122
|
+
def from_env(cls) -> "OpenAIConnection":
|
|
123
|
+
return cls(
|
|
124
|
+
base_url=os.environ.get("OPENAI_BASE_URL", ""),
|
|
125
|
+
api_key=os.environ.get("OPENAI_API_KEY"),
|
|
126
|
+
organization=os.environ.get("OPENAI_ORG_ID"),
|
|
127
|
+
)
|
|
128
|
+
|
|
129
|
+
def is_valid(self, missing_fields: Optional[Set[str]] = None) -> bool:
|
|
130
|
+
if missing_fields is None:
|
|
131
|
+
missing_fields = set()
|
|
132
|
+
if not self.base_url:
|
|
133
|
+
missing_fields.add("base_url")
|
|
134
|
+
if not self.api_key:
|
|
135
|
+
missing_fields.add("api_key")
|
|
136
|
+
if not self.organization:
|
|
137
|
+
missing_fields.add("organization")
|
|
138
|
+
return not bool(missing_fields)
|
|
139
|
+
|
|
140
|
+
|
|
141
|
+
@dataclass
|
|
142
|
+
class AzureOpenAIConnection(Connection):
|
|
143
|
+
"""Connection class for Azure OpenAI endpoints."""
|
|
144
|
+
|
|
145
|
+
azure_endpoint: str
|
|
146
|
+
api_key: Optional[str] = None # TODO ralphe: Replace this TokenCredential to allow for more flexible authentication
|
|
147
|
+
azure_deployment: Optional[str] = None
|
|
148
|
+
api_version: Optional[str] = None
|
|
149
|
+
resource_id: Optional[str] = None
|
|
150
|
+
|
|
151
|
+
TYPE: ClassVar[str] = "azure_openai"
|
|
152
|
+
|
|
153
|
+
@property
|
|
154
|
+
def type(self) -> str:
|
|
155
|
+
return AzureOpenAIConnection.TYPE
|
|
156
|
+
|
|
157
|
+
@classmethod
|
|
158
|
+
def from_env(cls) -> "AzureOpenAIConnection":
|
|
159
|
+
return cls(
|
|
160
|
+
azure_endpoint=os.environ.get("AZURE_OPENAI_ENDPOINT", ""),
|
|
161
|
+
api_key=os.environ.get("AZURE_OPENAI_API_KEY"),
|
|
162
|
+
azure_deployment=os.environ.get("AZURE_OPENAI_DEPLOYMENT"),
|
|
163
|
+
api_version=os.environ.get("AZURE_OPENAI_API_VERSION", "2024-02-01"),
|
|
164
|
+
)
|
|
165
|
+
|
|
166
|
+
def __post_init__(self):
|
|
167
|
+
# set default API version
|
|
168
|
+
if not self.api_version:
|
|
169
|
+
self.api_version = "2024-02-01"
|
|
170
|
+
|
|
171
|
+
def is_valid(self, missing_fields: Optional[Set[str]] = None) -> bool:
|
|
172
|
+
if missing_fields is None:
|
|
173
|
+
missing_fields = set()
|
|
174
|
+
if not self.azure_endpoint:
|
|
175
|
+
missing_fields.add("azure_endpoint")
|
|
176
|
+
if not self.api_key:
|
|
177
|
+
missing_fields.add("api_key")
|
|
178
|
+
if not self.azure_deployment:
|
|
179
|
+
missing_fields.add("azure_deployment")
|
|
180
|
+
if not self.api_version:
|
|
181
|
+
missing_fields.add("api_version")
|
|
182
|
+
return not bool(missing_fields)
|
|
@@ -0,0 +1,59 @@
|
|
|
1
|
+
# ---------------------------------------------------------
|
|
2
|
+
# Copyright (c) Microsoft Corporation. All rights reserved.
|
|
3
|
+
# ---------------------------------------------------------
|
|
4
|
+
|
|
5
|
+
from azure.ai.evaluation._exceptions import ErrorCategory, ErrorBlame, ErrorTarget, EvaluationException
|
|
6
|
+
|
|
7
|
+
|
|
8
|
+
class PromptyException(EvaluationException):
|
|
9
|
+
"""Exception class for Prompty related errors.
|
|
10
|
+
|
|
11
|
+
This exception is used to indicate that the error was caused by Prompty execution.
|
|
12
|
+
|
|
13
|
+
:param message: The error message.
|
|
14
|
+
:type message: str
|
|
15
|
+
"""
|
|
16
|
+
|
|
17
|
+
def __init__(self, message: str, **kwargs):
|
|
18
|
+
kwargs.setdefault("category", ErrorCategory.INVALID_VALUE)
|
|
19
|
+
kwargs.setdefault("target", ErrorTarget.UNKNOWN)
|
|
20
|
+
kwargs.setdefault("blame", ErrorBlame.USER_ERROR)
|
|
21
|
+
|
|
22
|
+
super().__init__(message, **kwargs)
|
|
23
|
+
|
|
24
|
+
|
|
25
|
+
class MissingRequiredInputError(PromptyException):
|
|
26
|
+
"""Exception raised when missing required input"""
|
|
27
|
+
|
|
28
|
+
def __init__(self, message: str, **kwargs):
|
|
29
|
+
kwargs.setdefault("category", ErrorCategory.MISSING_FIELD)
|
|
30
|
+
kwargs.setdefault("target", ErrorTarget.EVAL_RUN)
|
|
31
|
+
super().__init__(message, **kwargs)
|
|
32
|
+
|
|
33
|
+
|
|
34
|
+
class InvalidInputError(PromptyException):
|
|
35
|
+
"""Exception raised when an input is invalid, could not be loaded, or is not the expected format."""
|
|
36
|
+
|
|
37
|
+
def __init__(self, message: str, **kwargs):
|
|
38
|
+
kwargs.setdefault("category", ErrorCategory.INVALID_VALUE)
|
|
39
|
+
kwargs.setdefault("target", ErrorTarget.EVAL_RUN)
|
|
40
|
+
super().__init__(message, **kwargs)
|
|
41
|
+
|
|
42
|
+
|
|
43
|
+
class JinjaTemplateError(PromptyException):
|
|
44
|
+
"""Exception raised when the Jinja template is invalid or could not be rendered."""
|
|
45
|
+
|
|
46
|
+
def __init__(self, message: str, **kwargs):
|
|
47
|
+
kwargs.setdefault("category", ErrorCategory.INVALID_VALUE)
|
|
48
|
+
kwargs.setdefault("target", ErrorTarget.EVAL_RUN)
|
|
49
|
+
super().__init__(message, **kwargs)
|
|
50
|
+
|
|
51
|
+
|
|
52
|
+
class NotSupportedError(PromptyException):
|
|
53
|
+
"""Exception raised when the operation is not supported."""
|
|
54
|
+
|
|
55
|
+
def __init__(self, message: str, **kwargs):
|
|
56
|
+
kwargs.setdefault("category", ErrorCategory.INVALID_VALUE)
|
|
57
|
+
kwargs.setdefault("target", ErrorTarget.UNKNOWN)
|
|
58
|
+
kwargs.setdefault("blame", ErrorBlame.SYSTEM_ERROR)
|
|
59
|
+
super().__init__(message, **kwargs)
|
|
@@ -0,0 +1,313 @@
|
|
|
1
|
+
# ---------------------------------------------------------
|
|
2
|
+
# Copyright (c) Microsoft Corporation. All rights reserved.
|
|
3
|
+
# ---------------------------------------------------------
|
|
4
|
+
|
|
5
|
+
import re
|
|
6
|
+
from os import PathLike
|
|
7
|
+
from pathlib import Path
|
|
8
|
+
from typing import Any, AsyncGenerator, Dict, Final, List, Mapping, Optional, Sequence, Tuple, Union, cast
|
|
9
|
+
|
|
10
|
+
from openai import AsyncAzureOpenAI, AsyncOpenAI, NotGiven
|
|
11
|
+
|
|
12
|
+
from azure.ai.evaluation._exceptions import ErrorTarget
|
|
13
|
+
from azure.ai.evaluation._constants import DefaultOpenEncoding
|
|
14
|
+
from azure.ai.evaluation._legacy.prompty._exceptions import (
|
|
15
|
+
InvalidInputError,
|
|
16
|
+
PromptyException,
|
|
17
|
+
MissingRequiredInputError,
|
|
18
|
+
NotSupportedError,
|
|
19
|
+
)
|
|
20
|
+
from azure.ai.evaluation._legacy.prompty._connection import AzureOpenAIConnection, Connection, OpenAIConnection
|
|
21
|
+
from azure.ai.evaluation._legacy.prompty._yaml_utils import load_yaml_string
|
|
22
|
+
from azure.ai.evaluation._legacy.prompty._utils import (
|
|
23
|
+
dataclass_from_dict,
|
|
24
|
+
PromptyModelConfiguration,
|
|
25
|
+
OpenAIChatResponseType,
|
|
26
|
+
build_messages,
|
|
27
|
+
format_llm_response,
|
|
28
|
+
prepare_open_ai_request_params,
|
|
29
|
+
resolve_references,
|
|
30
|
+
update_dict_recursively,
|
|
31
|
+
)
|
|
32
|
+
|
|
33
|
+
|
|
34
|
+
PROMPTY_EXTENSION: Final[str] = ".prompty"
|
|
35
|
+
|
|
36
|
+
|
|
37
|
+
class AsyncPrompty:
|
|
38
|
+
"""A prompty is a prompt with predefined metadata like inputs, and can be executed directly like a flow.
|
|
39
|
+
A prompty is represented as a templated markdown file with a modified front matter.
|
|
40
|
+
The front matter is a yaml file that contains meta fields like model configuration, inputs, etc..
|
|
41
|
+
|
|
42
|
+
Prompty example:
|
|
43
|
+
.. code-block::
|
|
44
|
+
|
|
45
|
+
---
|
|
46
|
+
name: Hello Prompty
|
|
47
|
+
description: A basic prompt
|
|
48
|
+
model:
|
|
49
|
+
api: chat
|
|
50
|
+
configuration:
|
|
51
|
+
type: azure_openai
|
|
52
|
+
azure_deployment: gpt-35-turbo
|
|
53
|
+
api_key="${env:AZURE_OPENAI_API_KEY}",
|
|
54
|
+
api_version=${env:AZURE_OPENAI_API_VERSION}",
|
|
55
|
+
azure_endpoint="${env:AZURE_OPENAI_ENDPOINT}",
|
|
56
|
+
parameters:
|
|
57
|
+
max_tokens: 128
|
|
58
|
+
temperature: 0.2
|
|
59
|
+
inputs:
|
|
60
|
+
text:
|
|
61
|
+
type: string
|
|
62
|
+
---
|
|
63
|
+
system:
|
|
64
|
+
Write a simple {{text}} program that displays the greeting message.
|
|
65
|
+
|
|
66
|
+
Prompty as function example:
|
|
67
|
+
|
|
68
|
+
.. code-block:: python
|
|
69
|
+
|
|
70
|
+
from azure.ai.evaluation._legacy.prompty import AsyncPrompty
|
|
71
|
+
prompty = Prompty(path="path/to/prompty.prompty")
|
|
72
|
+
result = prompty(input_a=1, input_b=2)
|
|
73
|
+
|
|
74
|
+
# Override model config with dict
|
|
75
|
+
model_config = {
|
|
76
|
+
"api": "chat",
|
|
77
|
+
"configuration": {
|
|
78
|
+
"type": "azure_openai",
|
|
79
|
+
"azure_deployment": "gpt-35-turbo",
|
|
80
|
+
"api_key": "${env:AZURE_OPENAI_API_KEY}",
|
|
81
|
+
"api_version": "${env:AZURE_OPENAI_API_VERSION}",
|
|
82
|
+
"azure_endpoint": "${env:AZURE_OPENAI_ENDPOINT}",
|
|
83
|
+
},
|
|
84
|
+
"parameters": {
|
|
85
|
+
"max_token": 512
|
|
86
|
+
}
|
|
87
|
+
}
|
|
88
|
+
prompty = Prompty.load(source="path/to/prompty.prompty", model=model_config)
|
|
89
|
+
result = prompty(input_a=1, input_b=2)
|
|
90
|
+
|
|
91
|
+
# Override model config with configuration
|
|
92
|
+
from azure.ai.evaluation._legacy.prompty._connection import AzureOpenAIConnection
|
|
93
|
+
model_config = {
|
|
94
|
+
"api": "chat",
|
|
95
|
+
"configuration": AzureOpenAIModelConfiguration(
|
|
96
|
+
azure_deployment="gpt-35-turbo",
|
|
97
|
+
api_key="${env:AZURE_OPENAI_API_KEY}",
|
|
98
|
+
api_version="${env:AZURE_OPENAI_API_VERSION}",
|
|
99
|
+
azure_endpoint="${env:AZURE_OPENAI_ENDPOINT}",
|
|
100
|
+
),
|
|
101
|
+
"parameters": {
|
|
102
|
+
"max_token": 512
|
|
103
|
+
}
|
|
104
|
+
}
|
|
105
|
+
prompty = Prompty(path="path/to/prompty.prompty", model=model_config)
|
|
106
|
+
result = prompty(input_a=1, input_b=2)
|
|
107
|
+
|
|
108
|
+
# Override model config with created connection
|
|
109
|
+
from azure.ai.evaluation._legacy.prompty._connection import AzureOpenAIConnection
|
|
110
|
+
model_config = {
|
|
111
|
+
"api": "chat",
|
|
112
|
+
"configuration": AzureOpenAIModelConfiguration(
|
|
113
|
+
connection="azure_open_ai_connection",
|
|
114
|
+
azure_deployment="gpt-35-turbo",
|
|
115
|
+
),
|
|
116
|
+
"parameters": {
|
|
117
|
+
"max_token": 512
|
|
118
|
+
}
|
|
119
|
+
}
|
|
120
|
+
prompty = Prompty(path="path/to/prompty.prompty", model=model_config)
|
|
121
|
+
result = prompty(input_a=1, input_b=2)
|
|
122
|
+
"""
|
|
123
|
+
|
|
124
|
+
def __init__(
|
|
125
|
+
self,
|
|
126
|
+
path: Union[str, PathLike],
|
|
127
|
+
**kwargs: Any,
|
|
128
|
+
):
|
|
129
|
+
path = Path(path)
|
|
130
|
+
configs, self._template = self._parse_prompty(path)
|
|
131
|
+
configs = resolve_references(configs, base_path=path.parent)
|
|
132
|
+
configs = update_dict_recursively(configs, resolve_references(kwargs, base_path=path.parent))
|
|
133
|
+
|
|
134
|
+
if configs["model"].get("api") == "completion":
|
|
135
|
+
raise InvalidInputError(
|
|
136
|
+
"Prompty does not support the completion API. Please use the 'chat' completions API instead."
|
|
137
|
+
)
|
|
138
|
+
|
|
139
|
+
self._data = configs
|
|
140
|
+
self._path = path
|
|
141
|
+
self._model = dataclass_from_dict(PromptyModelConfiguration, configs["model"])
|
|
142
|
+
self._inputs: Dict[str, Any] = configs.get("inputs", {})
|
|
143
|
+
self._outputs: Dict[str, Any] = configs.get("outputs", {})
|
|
144
|
+
self._name: str = configs.get("name", path.stem)
|
|
145
|
+
|
|
146
|
+
@property
|
|
147
|
+
def path(self) -> Path:
|
|
148
|
+
"""Path of the prompty file.
|
|
149
|
+
|
|
150
|
+
:return: The path of the prompty file.
|
|
151
|
+
:rtype: Path
|
|
152
|
+
"""
|
|
153
|
+
return self._path
|
|
154
|
+
|
|
155
|
+
@property
|
|
156
|
+
def name(self) -> str:
|
|
157
|
+
"""Name of the prompty.
|
|
158
|
+
|
|
159
|
+
:return: The name of the prompty.
|
|
160
|
+
:rtype: str
|
|
161
|
+
"""
|
|
162
|
+
return self._name
|
|
163
|
+
|
|
164
|
+
@property
|
|
165
|
+
def description(self) -> Optional[str]:
|
|
166
|
+
"""Description of the prompty.
|
|
167
|
+
|
|
168
|
+
:return: The description of the prompty.
|
|
169
|
+
:rtype: str
|
|
170
|
+
"""
|
|
171
|
+
return self._data.get("description")
|
|
172
|
+
|
|
173
|
+
@classmethod
|
|
174
|
+
def load(
|
|
175
|
+
cls,
|
|
176
|
+
source: Union[str, PathLike],
|
|
177
|
+
**kwargs,
|
|
178
|
+
) -> "AsyncPrompty":
|
|
179
|
+
"""
|
|
180
|
+
Loads the prompty file.
|
|
181
|
+
|
|
182
|
+
:param source: The local prompty file. Must be a path to a local file.
|
|
183
|
+
An exception is raised if the file does not exist.
|
|
184
|
+
:type source: Union[PathLike, str]
|
|
185
|
+
:return: A Prompty object
|
|
186
|
+
:rtype: Prompty
|
|
187
|
+
"""
|
|
188
|
+
source_path = Path(source)
|
|
189
|
+
if not source_path.exists():
|
|
190
|
+
raise PromptyException(f"Source {source_path.absolute().as_posix()} does not exist")
|
|
191
|
+
|
|
192
|
+
if source_path.suffix != PROMPTY_EXTENSION:
|
|
193
|
+
raise PromptyException("Source must be a file with .prompty extension.")
|
|
194
|
+
|
|
195
|
+
return cls(path=source_path, **kwargs)
|
|
196
|
+
|
|
197
|
+
@staticmethod
|
|
198
|
+
def _parse_prompty(path) -> Tuple[Dict[str, Any], str]:
|
|
199
|
+
with open(path, "r", encoding=DefaultOpenEncoding.READ) as f:
|
|
200
|
+
prompty_content = f.read()
|
|
201
|
+
pattern = r"-{3,}\n(.*)-{3,}\n(.*)"
|
|
202
|
+
result = re.search(pattern, prompty_content, re.DOTALL)
|
|
203
|
+
if not result:
|
|
204
|
+
raise PromptyException(
|
|
205
|
+
"Illegal formatting of prompty. The prompt file is in markdown format and can be divided into two "
|
|
206
|
+
"parts, the first part is in YAML format and contains connection and model information. The second "
|
|
207
|
+
"part is the prompt template."
|
|
208
|
+
)
|
|
209
|
+
config_content, prompt_template = result.groups()
|
|
210
|
+
configs = load_yaml_string(config_content)
|
|
211
|
+
return configs, prompt_template
|
|
212
|
+
|
|
213
|
+
def _resolve_inputs(self, input_values: Dict[str, Any]) -> Mapping[str, Any]:
|
|
214
|
+
"""
|
|
215
|
+
Resolve prompty inputs. If not provide input_values, sample data will be regarded as input value.
|
|
216
|
+
For inputs are not provided, the default value in the input signature will be used.
|
|
217
|
+
|
|
218
|
+
:param Dict[str, Any] input_values: The input values provided by the user.
|
|
219
|
+
:return: The resolved inputs.
|
|
220
|
+
:rtype: Mapping[str, Any]
|
|
221
|
+
"""
|
|
222
|
+
|
|
223
|
+
resolved_inputs: Dict[str, Any] = {}
|
|
224
|
+
missing_inputs: List[str] = []
|
|
225
|
+
for input_name, value in self._inputs.items():
|
|
226
|
+
if input_name not in input_values and "default" not in value:
|
|
227
|
+
missing_inputs.append(input_name)
|
|
228
|
+
continue
|
|
229
|
+
|
|
230
|
+
resolved_inputs[input_name] = input_values.get(input_name, value.get("default", None))
|
|
231
|
+
|
|
232
|
+
if missing_inputs:
|
|
233
|
+
raise MissingRequiredInputError(f"Missing required inputs: {missing_inputs}")
|
|
234
|
+
|
|
235
|
+
return resolved_inputs
|
|
236
|
+
|
|
237
|
+
# TODO ralphe: error handling
|
|
238
|
+
# @trace
|
|
239
|
+
# @handle_openai_error()
|
|
240
|
+
async def __call__( # pylint: disable=docstring-keyword-should-match-keyword-only
|
|
241
|
+
self,
|
|
242
|
+
**kwargs: Any,
|
|
243
|
+
) -> Union[OpenAIChatResponseType, AsyncGenerator[str, None], str, Mapping[str, Any]]:
|
|
244
|
+
"""Calling prompty as a function in async, the inputs should be provided with key word arguments.
|
|
245
|
+
Returns the output of the prompty.
|
|
246
|
+
|
|
247
|
+
The function call throws PromptyException if the Prompty file is not valid or the inputs are not valid.
|
|
248
|
+
|
|
249
|
+
:keyword kwargs: Additional keyword arguments passed to the parent class.
|
|
250
|
+
:paramtype kwargs: Any
|
|
251
|
+
:return: The output of the prompty.
|
|
252
|
+
:rtype: ChatCompletion | AsyncStream[ChatCompletionChunk] | AsyncGenerator[str] | str | Mapping[str, Any]
|
|
253
|
+
"""
|
|
254
|
+
|
|
255
|
+
inputs = self._resolve_inputs(kwargs)
|
|
256
|
+
connection = Connection.parse_from_config(self._model.configuration)
|
|
257
|
+
messages = build_messages(prompt=self._template, working_dir=self.path.parent, **inputs)
|
|
258
|
+
params = prepare_open_ai_request_params(self._model, messages)
|
|
259
|
+
|
|
260
|
+
timeout: Union[NotGiven, float] = NotGiven()
|
|
261
|
+
if timeout_val := cast(Any, kwargs.get("timeout", None)):
|
|
262
|
+
timeout = float(timeout_val)
|
|
263
|
+
|
|
264
|
+
# disable OpenAI's built-in retry mechanism by using our own retry
|
|
265
|
+
# for better debugging and real-time status updates.
|
|
266
|
+
max_retries = 0
|
|
267
|
+
|
|
268
|
+
api_client: Union[AsyncAzureOpenAI, AsyncOpenAI]
|
|
269
|
+
if isinstance(connection, AzureOpenAIConnection):
|
|
270
|
+
api_client = AsyncAzureOpenAI(
|
|
271
|
+
azure_endpoint=connection.azure_endpoint,
|
|
272
|
+
api_key=connection.api_key,
|
|
273
|
+
azure_deployment=connection.azure_deployment,
|
|
274
|
+
api_version=connection.api_version,
|
|
275
|
+
max_retries=max_retries,
|
|
276
|
+
)
|
|
277
|
+
elif isinstance(connection, OpenAIConnection):
|
|
278
|
+
api_client = AsyncOpenAI(
|
|
279
|
+
base_url=connection.base_url,
|
|
280
|
+
api_key=connection.api_key,
|
|
281
|
+
organization=connection.organization,
|
|
282
|
+
max_retries=max_retries,
|
|
283
|
+
)
|
|
284
|
+
else:
|
|
285
|
+
raise NotSupportedError(
|
|
286
|
+
f"'{type(connection).__name__}' is not a supported connection type.", target=ErrorTarget.EVAL_RUN
|
|
287
|
+
)
|
|
288
|
+
|
|
289
|
+
response: OpenAIChatResponseType = await api_client.with_options(timeout=timeout).chat.completions.create(
|
|
290
|
+
**params
|
|
291
|
+
)
|
|
292
|
+
|
|
293
|
+
return await format_llm_response(
|
|
294
|
+
response=response,
|
|
295
|
+
is_first_choice=self._data.get("model", {}).get("response", "first").lower() == "first",
|
|
296
|
+
response_format=params.get("response_format", {}),
|
|
297
|
+
outputs=self._outputs,
|
|
298
|
+
)
|
|
299
|
+
|
|
300
|
+
def render( # pylint: disable=docstring-keyword-should-match-keyword-only
|
|
301
|
+
self, **kwargs: Any
|
|
302
|
+
) -> Sequence[Mapping[str, Any]]:
|
|
303
|
+
"""Render the prompt content.
|
|
304
|
+
|
|
305
|
+
:keyword kwargs: Additional keyword arguments passed to the parent class.
|
|
306
|
+
:paramtype kwargs: Any
|
|
307
|
+
:return: Prompt content
|
|
308
|
+
:rtype: Sequence[Mapping[str, Any]]
|
|
309
|
+
"""
|
|
310
|
+
|
|
311
|
+
inputs = self._resolve_inputs(kwargs)
|
|
312
|
+
messages = build_messages(prompt=self._template, working_dir=self.path.parent, **inputs)
|
|
313
|
+
return messages
|