unique_toolkit 1.17.0__py3-none-any.whl → 1.17.1__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 unique_toolkit might be problematic. Click here for more details.
- unique_toolkit/agentic/evaluation/hallucination/hallucination_evaluation.py +11 -2
- {unique_toolkit-1.17.0.dist-info → unique_toolkit-1.17.1.dist-info}/METADATA +3 -1
- {unique_toolkit-1.17.0.dist-info → unique_toolkit-1.17.1.dist-info}/RECORD +5 -5
- {unique_toolkit-1.17.0.dist-info → unique_toolkit-1.17.1.dist-info}/LICENSE +0 -0
- {unique_toolkit-1.17.0.dist-info → unique_toolkit-1.17.1.dist-info}/WHEEL +0 -0
|
@@ -1,3 +1,5 @@
|
|
|
1
|
+
import regex as re
|
|
2
|
+
|
|
1
3
|
from unique_toolkit.agentic.evaluation.evaluation_manager import Evaluation
|
|
2
4
|
from unique_toolkit.agentic.evaluation.hallucination.constants import (
|
|
3
5
|
HallucinationConfig,
|
|
@@ -40,13 +42,20 @@ class HallucinationEvaluation(Evaluation):
|
|
|
40
42
|
async def run(
|
|
41
43
|
self, loop_response: LanguageModelStreamResponse
|
|
42
44
|
) -> EvaluationMetricResult: # type: ignore
|
|
43
|
-
|
|
45
|
+
all_chunks = self._reference_manager.get_chunks()
|
|
46
|
+
# source numbers from original text
|
|
47
|
+
ref_pattern = r"\[source(\d+)\]"
|
|
48
|
+
original_text = loop_response.message.original_text
|
|
49
|
+
source_number_matches = re.findall(ref_pattern, original_text)
|
|
50
|
+
source_numbers = {int(num) for num in source_number_matches}
|
|
51
|
+
|
|
52
|
+
referenced_chunks = [all_chunks[idx] for idx in source_numbers]
|
|
44
53
|
|
|
45
54
|
evaluation_result: EvaluationMetricResult = await check_hallucination(
|
|
46
55
|
company_id=self._company_id,
|
|
47
56
|
input=EvaluationMetricInput(
|
|
48
57
|
input_text=self._user_message,
|
|
49
|
-
context_texts=[context.text for context in
|
|
58
|
+
context_texts=[context.text for context in referenced_chunks],
|
|
50
59
|
history_messages=[], # TODO include loop_history messages
|
|
51
60
|
output_text=loop_response.message.text,
|
|
52
61
|
),
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.1
|
|
2
2
|
Name: unique_toolkit
|
|
3
|
-
Version: 1.17.
|
|
3
|
+
Version: 1.17.1
|
|
4
4
|
Summary:
|
|
5
5
|
License: Proprietary
|
|
6
6
|
Author: Cedric Klinkert
|
|
@@ -118,6 +118,8 @@ All notable changes to this project will be documented in this file.
|
|
|
118
118
|
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
|
|
119
119
|
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
|
|
120
120
|
|
|
121
|
+
## [1.17.1] - 2025-10-23
|
|
122
|
+
- Fix hallucination check input with all cited reference chunks.
|
|
121
123
|
## [1.17.0] - 2025-10-22
|
|
122
124
|
- Add more options to display sub agent answers in the chat.
|
|
123
125
|
|
|
@@ -35,7 +35,7 @@ unique_toolkit/agentic/evaluation/context_relevancy/service.py,sha256=2NM1_PCP6f
|
|
|
35
35
|
unique_toolkit/agentic/evaluation/evaluation_manager.py,sha256=IPx4BVUgkjFOP1BGLi0BlB6UujpXlZ0KGuSXDRemQhY,8143
|
|
36
36
|
unique_toolkit/agentic/evaluation/exception.py,sha256=7lcVbCyoN4Md1chNJDFxpUYyWbVrcr9dcc3TxWykJTc,115
|
|
37
37
|
unique_toolkit/agentic/evaluation/hallucination/constants.py,sha256=SoGmoYti2J33tSmmOC1BSF6Pkh8DQvbQAU9xIZFQZRs,2070
|
|
38
|
-
unique_toolkit/agentic/evaluation/hallucination/hallucination_evaluation.py,sha256=
|
|
38
|
+
unique_toolkit/agentic/evaluation/hallucination/hallucination_evaluation.py,sha256=JPSBr9JvufeKpIyiscaWCUqpJB5E7Z8ftKUPdbdp6XE,3829
|
|
39
39
|
unique_toolkit/agentic/evaluation/hallucination/prompts.py,sha256=O3Hi_rOzZlujvnO2wn2jhoPmrYLjzVtRWwxn5Q81m9Y,3405
|
|
40
40
|
unique_toolkit/agentic/evaluation/hallucination/service.py,sha256=Ut-f768HY4E9zEhfMoKYnGTFRZVkxWGiSTGOpgfZWYM,2447
|
|
41
41
|
unique_toolkit/agentic/evaluation/hallucination/utils.py,sha256=QLsYvgAyQ5XnKEzn7ko7bXfzePD4De99TWnMKglMpds,8178
|
|
@@ -165,7 +165,7 @@ unique_toolkit/short_term_memory/service.py,sha256=5PeVBu1ZCAfyDb2HLVvlmqSbyzBBu
|
|
|
165
165
|
unique_toolkit/smart_rules/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
166
166
|
unique_toolkit/smart_rules/compile.py,sha256=Ozhh70qCn2yOzRWr9d8WmJeTo7AQurwd3tStgBMPFLA,1246
|
|
167
167
|
unique_toolkit/test_utilities/events.py,sha256=_mwV2bs5iLjxS1ynDCjaIq-gjjKhXYCK-iy3dRfvO3g,6410
|
|
168
|
-
unique_toolkit-1.17.
|
|
169
|
-
unique_toolkit-1.17.
|
|
170
|
-
unique_toolkit-1.17.
|
|
171
|
-
unique_toolkit-1.17.
|
|
168
|
+
unique_toolkit-1.17.1.dist-info/LICENSE,sha256=GlN8wHNdh53xwOPg44URnwag6TEolCjoq3YD_KrWgss,193
|
|
169
|
+
unique_toolkit-1.17.1.dist-info/METADATA,sha256=D7xtrTsj4TmyJ0beu7DX3LgK1mPLQG2Iyr0aOVO_Fh4,38019
|
|
170
|
+
unique_toolkit-1.17.1.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
|
|
171
|
+
unique_toolkit-1.17.1.dist-info/RECORD,,
|
|
File without changes
|
|
File without changes
|