judgeval 0.0.26__py3-none-any.whl → 0.0.28__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.
- judgeval/common/tracer.py +515 -193
- judgeval/constants.py +4 -2
- judgeval/data/__init__.py +0 -3
- judgeval/data/{api_example.py → custom_api_example.py} +12 -19
- judgeval/data/datasets/eval_dataset_client.py +59 -20
- judgeval/data/result.py +34 -56
- judgeval/evaluation_run.py +1 -0
- judgeval/judgment_client.py +47 -15
- judgeval/run_evaluation.py +20 -36
- judgeval/scorers/score.py +9 -11
- {judgeval-0.0.26.dist-info → judgeval-0.0.28.dist-info}/METADATA +1 -1
- {judgeval-0.0.26.dist-info → judgeval-0.0.28.dist-info}/RECORD +14 -14
- {judgeval-0.0.26.dist-info → judgeval-0.0.28.dist-info}/WHEEL +0 -0
- {judgeval-0.0.26.dist-info → judgeval-0.0.28.dist-info}/licenses/LICENSE.md +0 -0
judgeval/scorers/score.py
CHANGED
@@ -13,7 +13,6 @@ from judgeval.data import (
|
|
13
13
|
Example,
|
14
14
|
ScoringResult,
|
15
15
|
generate_scoring_result,
|
16
|
-
create_process_example,
|
17
16
|
create_scorer_data,
|
18
17
|
)
|
19
18
|
from judgeval.scorers import JudgevalScorer
|
@@ -400,7 +399,6 @@ async def a_eval_examples_helper(
|
|
400
399
|
scorer.error = None # Reset scorer error
|
401
400
|
|
402
401
|
# scoring the Example
|
403
|
-
process_example = create_process_example(example) # Creates process example to track progress
|
404
402
|
scoring_start_time = time.perf_counter()
|
405
403
|
await score_with_indicator(
|
406
404
|
scorers=scorers,
|
@@ -411,22 +409,22 @@ async def a_eval_examples_helper(
|
|
411
409
|
) # execute the scoring functions of each scorer on the example
|
412
410
|
|
413
411
|
# Now that all the scoring functions of each scorer have executed, we collect
|
414
|
-
# the results and update the
|
412
|
+
# the results and update the ScoringResult with the scorer data
|
413
|
+
success = True
|
414
|
+
scorer_data_list = []
|
415
415
|
for scorer in scorers:
|
416
416
|
# At this point, the scorer has been executed and already contains data.
|
417
417
|
if getattr(scorer, 'skipped', False):
|
418
418
|
continue
|
419
419
|
scorer_data = create_scorer_data(scorer) # Fetch scorer data from completed scorer evaluation
|
420
|
-
|
420
|
+
success = success and scorer_data.success
|
421
|
+
scorer_data_list.append(scorer_data)
|
421
422
|
|
422
|
-
|
423
|
-
run_duration =
|
423
|
+
scoring_end_time = time.perf_counter()
|
424
|
+
run_duration = scoring_end_time - scoring_start_time
|
424
425
|
|
425
|
-
|
426
|
-
|
427
|
-
# Generate the scoring result and store it safely (to avoid race conditions)
|
428
|
-
result = generate_scoring_result(process_example)
|
429
|
-
scoring_results[score_index] = result
|
426
|
+
scoring_result = generate_scoring_result(example, scorer_data_list, run_duration, success)
|
427
|
+
scoring_results[score_index] = scoring_result
|
430
428
|
|
431
429
|
if pbar is not None:
|
432
430
|
pbar.update(1)
|
@@ -1,23 +1,23 @@
|
|
1
1
|
judgeval/__init__.py,sha256=dtXxsCmI4eEsZdGSUMy8P_pA0bc2-OSGAgb2C__yJoA,252
|
2
2
|
judgeval/clients.py,sha256=6VQmEqmfCngUdS2MuPBIpHvtDFqOENm8-_BmMvjLyRQ,944
|
3
|
-
judgeval/constants.py,sha256=
|
4
|
-
judgeval/evaluation_run.py,sha256=
|
5
|
-
judgeval/judgment_client.py,sha256=
|
3
|
+
judgeval/constants.py,sha256=ksAXhAXovzJKH0uHOdQtREs168uCJRG79PooHNmEbYQ,5313
|
4
|
+
judgeval/evaluation_run.py,sha256=6Kft3wZDWkdBDZoMwOhWf7zSAOF4naI7Pcg_YlZaZY4,6394
|
5
|
+
judgeval/judgment_client.py,sha256=uf0V1-eu3qnFTwrQ_Ckcv8IiWRVv7dbvou4P4KjU6hM,26794
|
6
6
|
judgeval/rules.py,sha256=B0ZL0pn72D4Jnlr0zMQ6CPHi7D8AQQRariXCVsiCMiI,20542
|
7
|
-
judgeval/run_evaluation.py,sha256=
|
7
|
+
judgeval/run_evaluation.py,sha256=N2ppmEE5WoSReChKjr_n0NcdAUlUR6Nua7M1C_3zHQ8,24949
|
8
8
|
judgeval/common/__init__.py,sha256=7d24BRxtncpMj3AAJCj8RS7TqgjXmW777HVZH6-3sBs,289
|
9
9
|
judgeval/common/exceptions.py,sha256=U-TxHLn7oVMezsMuoYouNDb2XuS8RCggfntYf5_6u4E,565
|
10
10
|
judgeval/common/logger.py,sha256=KO75wWXCxhUHUMvLaTU31ZzOk6tkZBa7heQ7y0f-zFE,6062
|
11
|
-
judgeval/common/tracer.py,sha256=
|
11
|
+
judgeval/common/tracer.py,sha256=Qpn2m6LCpRq1OOWRd1z16JtmeS7ITIWaQNJOddmAfQY,58178
|
12
12
|
judgeval/common/utils.py,sha256=LUQV5JfDr6wj7xHAJoNq-gofNZ6mjXbeKrGKzBME1KM,33533
|
13
|
-
judgeval/data/__init__.py,sha256=
|
14
|
-
judgeval/data/
|
13
|
+
judgeval/data/__init__.py,sha256=dG5ytBOeOWCTd5o0KP7IblqtW4G1EBaGreLWepM3jas,345
|
14
|
+
judgeval/data/custom_api_example.py,sha256=uW_ZBzkDLWumtudmfRHAJQkVYpm2qWgcDf7vBNLpS-o,3444
|
15
15
|
judgeval/data/example.py,sha256=BhGBhamFWgH6wtvrRYM8dGtDfXh-cDxDhtNL5Gbdz_M,5892
|
16
|
-
judgeval/data/result.py,sha256=
|
16
|
+
judgeval/data/result.py,sha256=BT4f2FF5EFuiRjOmS4vuIXsrEwSlG16Vw3QaWi6PZzc,3122
|
17
17
|
judgeval/data/scorer_data.py,sha256=JVlaTx1EP2jw2gh3Vgx1CSEsvIFABAN26IquKyxwiJQ,3273
|
18
18
|
judgeval/data/datasets/__init__.py,sha256=IdNKhQv9yYZ_op0rdBacrFaFVmiiYQ3JTzXzxOTsEVQ,176
|
19
19
|
judgeval/data/datasets/dataset.py,sha256=AFYjksV_wXx5CqFYJsl3aN8yZ6hC50O1myRuOJ8s8_E,12867
|
20
|
-
judgeval/data/datasets/eval_dataset_client.py,sha256=
|
20
|
+
judgeval/data/datasets/eval_dataset_client.py,sha256=P9fEmcNrjPPaiYbbLiEiBziZrIexA39HN9qzClt6uPE,12691
|
21
21
|
judgeval/integrations/langgraph.py,sha256=fGDZOTlVbxTO4ErC-m9OSg3h-RkOIIWXCfhjgkKRh4E,11187
|
22
22
|
judgeval/judges/__init__.py,sha256=6X7VSwrwsdxGBNxCyapVRWGghhKOy3MVxFNMQ62kCXM,308
|
23
23
|
judgeval/judges/base_judge.py,sha256=ch_S7uBB7lyv44Lf1d7mIGFpveOO58zOkkpImKgd9_4,994
|
@@ -31,7 +31,7 @@ judgeval/scorers/base_scorer.py,sha256=xdUlY3CnLdCQ1Z5iUeY22Bim5v-OQruZmaVF_4Y1m
|
|
31
31
|
judgeval/scorers/exceptions.py,sha256=eGW5CuJgZ5YJBFrE4FHDSF651PO1dKAZ379mJ8gOsfo,178
|
32
32
|
judgeval/scorers/judgeval_scorer.py,sha256=jq_rzfTG0XBTuLCaa6TlaK4YcT-LlgsO1LEm6hpOYdg,6601
|
33
33
|
judgeval/scorers/prompt_scorer.py,sha256=PaAs2qRolw1P3_I061Xvk9qzvF4O-JR8g_39RqXnHcM,17728
|
34
|
-
judgeval/scorers/score.py,sha256=
|
34
|
+
judgeval/scorers/score.py,sha256=ObFAlMbNRcGrfBpH4WW_6OA3CjrneC539xSWhGH60GQ,18578
|
35
35
|
judgeval/scorers/utils.py,sha256=iHQVTlIANbmCTXz9kTeSdOytgUZ_T74Re61ajqsk_WQ,6827
|
36
36
|
judgeval/scorers/judgeval_scorers/__init__.py,sha256=xFRb62sp4JmBUSeuAB_pC_7kEGp-lGdqCRIu9--Bbdg,5992
|
37
37
|
judgeval/scorers/judgeval_scorers/api_scorers/__init__.py,sha256=mZ6b_5Dl04k3PaG24ICBajB_j43ody1II1OJhO1DkXo,1648
|
@@ -87,7 +87,7 @@ judgeval/scorers/judgeval_scorers/local_implementations/summarization/prompts.py
|
|
87
87
|
judgeval/scorers/judgeval_scorers/local_implementations/summarization/summarization_scorer.py,sha256=Qk7lwHgRPYeGoxTOyclAh1VfGItfvHJ6l1t7Nk3SWFM,20927
|
88
88
|
judgeval/tracer/__init__.py,sha256=wy3DYpH8U_z0GO_K_gOSkK0tTTD-u5eLDo0T5xIBoAc,147
|
89
89
|
judgeval/utils/alerts.py,sha256=O19Xj7DA0YVjl8PWiuH4zfdZeu3yiLVvHfY8ah2wG0g,2759
|
90
|
-
judgeval-0.0.
|
91
|
-
judgeval-0.0.
|
92
|
-
judgeval-0.0.
|
93
|
-
judgeval-0.0.
|
90
|
+
judgeval-0.0.28.dist-info/METADATA,sha256=GSGf7_cb7FkKdQ_PFPf4nw9hlMrKyD3Tv6X8m2uo3EY,5418
|
91
|
+
judgeval-0.0.28.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
|
92
|
+
judgeval-0.0.28.dist-info/licenses/LICENSE.md,sha256=tKmCg7k5QOmxPK19XMfzim04QiQJPmgIm0pAn55IJwk,11352
|
93
|
+
judgeval-0.0.28.dist-info/RECORD,,
|
File without changes
|
File without changes
|