judgeval 0.3.0__py3-none-any.whl → 0.3.2__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.
@@ -142,7 +142,7 @@ class DatasetStatsPayload(TypedDict):
142
142
 
143
143
 
144
144
  # Projects API
145
- JUDGMENT_PROJECT_DELETE_API_URL = f"{ROOT_API}/projects/delete_from_judgeval"
145
+ JUDGMENT_PROJECT_DELETE_API_URL = f"{ROOT_API}/projects/delete_from_judgeval/"
146
146
  JUDGMENT_PROJECT_CREATE_API_URL = f"{ROOT_API}/projects/add/"
147
147
 
148
148
 
@@ -54,7 +54,6 @@ class TraceManagerClient:
54
54
  Returns:
55
55
  dict: Server response containing UI URL and other metadata
56
56
  """
57
- server_response = self.api_client.upsert_trace(trace_data)
58
57
 
59
58
  if self.tracer and self.tracer.use_s3 and final_save:
60
59
  try:
@@ -67,6 +66,11 @@ class TraceManagerClient:
67
66
  except Exception as e:
68
67
  judgeval_logger.warning(f"Failed to save trace to S3: {str(e)}")
69
68
 
69
+ trace_data.pop("trace_spans", None)
70
+ trace_data.pop("evaluation_runs", None)
71
+
72
+ server_response = self.api_client.upsert_trace(trace_data)
73
+
70
74
  if not offline_mode and show_link and "ui_results_url" in server_response:
71
75
  pretty_str = f"\n🔍 You can view your trace data here: [rgb(106,0,255)][link={server_response['ui_results_url']}]View Trace[/link]\n"
72
76
  rprint(pretty_str)
judgeval/constants.py CHANGED
@@ -104,6 +104,8 @@ TOGETHER_SUPPORTED_MODELS = [
104
104
  "mistralai/Mistral-7B-Instruct-v0.1",
105
105
  ]
106
106
 
107
+ DEFAULT_TOGETHER_MODEL = "meta-llama/Meta-Llama-3-8B-Instruct-Lite"
108
+
107
109
  JUDGMENT_SUPPORTED_MODELS = {"osiris-large", "osiris-mini", "osiris"}
108
110
 
109
111
  ACCEPTABLE_MODELS = (
@@ -1,6 +1,6 @@
1
1
  # generated by datamodel-codegen:
2
2
  # filename: openapi_new.json
3
- # timestamp: 2025-07-26T00:14:40+00:00
3
+ # timestamp: 2025-07-29T18:13:07+00:00
4
4
 
5
5
  from __future__ import annotations
6
6
 
@@ -135,7 +135,6 @@ class TraceJudgmentType(BaseModel):
135
135
  created_at: Annotated[str, Field(title="Created At")]
136
136
  duration: Annotated[float, Field(title="Duration")]
137
137
  trace_spans: Annotated[List[TraceSpanJudgmentType], Field(title="Trace Spans")]
138
- overwrite: Annotated[Optional[bool], Field(title="Overwrite")] = False
139
138
  offline_mode: Annotated[Optional[bool], Field(title="Offline Mode")] = False
140
139
  rules: Annotated[Optional[Dict[str, Any]], Field(title="Rules")] = Field(
141
140
  default_factory=dict
@@ -11,6 +11,7 @@ from judgeval.common.utils import (
11
11
  afetch_together_api_response,
12
12
  )
13
13
  from judgeval.common.logger import judgeval_logger
14
+ from judgeval.constants import DEFAULT_TOGETHER_MODEL
14
15
 
15
16
  BASE_CONVERSATION = [
16
17
  {"role": "system", "content": "You are a helpful assistant."},
@@ -18,7 +19,7 @@ BASE_CONVERSATION = [
18
19
 
19
20
 
20
21
  class TogetherJudge(JudgevalJudge):
21
- def __init__(self, model: str = "Qwen/Qwen2.5-72B-Instruct-Turbo", **kwargs):
22
+ def __init__(self, model: str = DEFAULT_TOGETHER_MODEL, **kwargs):
22
23
  self.model = model
23
24
  self.kwargs = kwargs
24
25
  super().__init__(model_name=model)
@@ -414,7 +414,7 @@ def _poll_evaluation_until_complete(
414
414
  expected_scorer_data_count: int,
415
415
  poll_interval_seconds: float = 5,
416
416
  max_failures: int = 5,
417
- max_poll_count: int = 24, # This should be equivalent to 120 seconds
417
+ max_poll_count: int = 60, # This should be equivalent to 5 minutes
418
418
  ) -> Tuple[List[ScoringResult], str]:
419
419
  """
420
420
  Polls until the evaluation is complete and returns the results.
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: judgeval
3
- Version: 0.3.0
3
+ Version: 0.3.2
4
4
  Summary: Judgeval Package
5
5
  Project-URL: Homepage, https://github.com/JudgmentLabs/judgeval
6
6
  Project-URL: Issues, https://github.com/JudgmentLabs/judgeval/issues
@@ -1,11 +1,11 @@
1
1
  judgeval/__init__.py,sha256=HM1M8hmqRum6G554QKkXhB4DF4f5eh_xtYo0Kf-t3kw,332
2
2
  judgeval/clients.py,sha256=JnB8n90GyXiYaGmSEYaA67mdJSnr3SIrzArao7NGebw,980
3
- judgeval/constants.py,sha256=rfl4gW9_4irxgamjTC-jvDj2ATSUrjEu0UAHZ4pLLtY,4081
3
+ judgeval/constants.py,sha256=hWed25HwGUJy-tePbtoUZ0_Zg0X_MkAH84KiH-OHHFI,4150
4
4
  judgeval/dataset.py,sha256=rjV54XNTslNNtf-Uu2ndDIh602ZwSCFhPg2NuckDJ-w,6081
5
5
  judgeval/evaluation_run.py,sha256=edNpO444Fwt2ykWsflIzlYdDJUlUfbpXHHQSKfFS4y0,2876
6
6
  judgeval/judgment_client.py,sha256=vPoxbmxAlhbG5rXXqxWjMbyEqOI044BaQanr1fev2CE,11723
7
7
  judgeval/rules.py,sha256=CoQjqmP8daEXewMkplmA-7urubDtweOr5O6z8klVwLI,20031
8
- judgeval/run_evaluation.py,sha256=hOKDdymH8VPQ8LPXJOVku41nwDDYLSCtHQIlIzFNYaE,27654
8
+ judgeval/run_evaluation.py,sha256=7J6FHhWhB-IDPMSOcWkrjTpSNm2v3s_KBq8Np3y2pys,27652
9
9
  judgeval/version_check.py,sha256=FoLEtpCjDw2HuDQdpw5yT29UtwumSc6ZZN6AV_c9Mnw,1057
10
10
  judgeval/common/__init__.py,sha256=KH-QJyWtQ60R6yFIBDYS3WGRiNpEu1guynpxivZvpBQ,309
11
11
  judgeval/common/exceptions.py,sha256=OkgDznu2wpBQZMXiZarLJYNk1HIcC8qYW7VypDC3Ook,556
@@ -13,7 +13,7 @@ judgeval/common/logger.py,sha256=514eFLYWS_UL8VY-zAR2ePUlpQe4rbYlleLASFllLE4,151
13
13
  judgeval/common/utils.py,sha256=oxGDRVWOICKWeyGgsoc36_yAyHSYF4XtH842Mkznwis,34739
14
14
  judgeval/common/api/__init__.py,sha256=-E7lpZz1fG8puR_aYUMfPmQ-Vyhd0bgzoaU5EhIuFjQ,114
15
15
  judgeval/common/api/api.py,sha256=wty02HYANeOYlM8fHOLc33ux5bu9Ieq7iRqCr-UP0ng,14157
16
- judgeval/common/api/constants.py,sha256=9B7y-oaqAspWZLHEf8wkfWf0OzpDyu3AKD1sfq4JsB0,4566
16
+ judgeval/common/api/constants.py,sha256=vAW94pbyTS6rv1TKpt7z6xxMJvTaAxFiy1D4kzuLHeg,4567
17
17
  judgeval/common/storage/__init__.py,sha256=a-PI7OL-ydyzugGUKmJKRBASnK-Q-gs82L9K9rSyJP8,90
18
18
  judgeval/common/storage/s3_storage.py,sha256=0-bNKheqJJyBZ92KGrzQtd1zocIRWBlfn_58L4a-Ay0,3719
19
19
  judgeval/common/tracer/__init__.py,sha256=tJCJsmVmrL89Phv88gNCJ-j0ITPez6lh8vhMAAlLNSc,795
@@ -23,10 +23,10 @@ judgeval/common/tracer/otel_exporter.py,sha256=kZLlOQ6afQE4dmb9H1wgU4P3H5PG1D_zK
23
23
  judgeval/common/tracer/otel_span_processor.py,sha256=W7SM62KnxJ48vC9WllIHRKaLlvxkCwqYoT4KqZLfGNs,6497
24
24
  judgeval/common/tracer/span_processor.py,sha256=eFjTgSWSkM6BWE94CrvgafDg_WkxLsFL_MafwBG-p9M,1145
25
25
  judgeval/common/tracer/span_transformer.py,sha256=nCnwRC52OKfYRFnsOwGdPaqb_U17yn5S_9jfhv1GaLM,7803
26
- judgeval/common/tracer/trace_manager.py,sha256=7KLWBrz5GE_138DHL_eRjhx4-LNfXKz1q_XIDfg6nw8,2992
26
+ judgeval/common/tracer/trace_manager.py,sha256=ltiXcWC-68DRc8uSa28qHiWRSIBf6NpYOPkZYooR8tg,3086
27
27
  judgeval/data/__init__.py,sha256=1QagDcSQtfnJ632t9Dnq8d7XjAqhmY4mInOWt8qH9tM,455
28
28
  judgeval/data/example.py,sha256=kRskIgsjwcvv2Y8jaPwV-PND7zlmMbFsvRVQ_b7SZY0,914
29
- judgeval/data/judgment_types.py,sha256=7dox20cUlAd4gvDN3iaSsRff-J6p-mxT9dwuhE4Ztr0,8411
29
+ judgeval/data/judgment_types.py,sha256=KE1HrFLfSxiu1zutaiZ7B7La9PGXIAsoWpo_5iy645c,8336
30
30
  judgeval/data/result.py,sha256=OtSnBUrdQpjyAqxXRLTW3wC9v9lOm_GqzL14ccRQxrg,2124
31
31
  judgeval/data/scorer_data.py,sha256=5QBHtvOIWOq0Rn9_uPJzAMRYMlWxMB-rXnG_6kV4Z4Y,2955
32
32
  judgeval/data/tool.py,sha256=iWQSdy5uNbIeACu3gQy1DC2oGYxRVYNfkkczWdQMAiA,99
@@ -39,7 +39,7 @@ judgeval/judges/__init__.py,sha256=6X7VSwrwsdxGBNxCyapVRWGghhKOy3MVxFNMQ62kCXM,3
39
39
  judgeval/judges/base_judge.py,sha256=_dz0qWsKRxzXxpRY9l6mrxTRYPSF2FE4ZXkrzhZ4gbY,986
40
40
  judgeval/judges/litellm_judge.py,sha256=yt6QvwKMmxZcrUtjbn3EiO5aVg7CHM2YZkBCSQLS8jk,2509
41
41
  judgeval/judges/mixture_of_judges.py,sha256=cecQ8mRmz2-dDoZl2MGsrhZICkpIvRovGPK3su0kc8s,14889
42
- judgeval/judges/together_judge.py,sha256=e2tr0bODZEkW45RQNW3eE8Z8XUIc8VUTw-cuwT6eYbw,2293
42
+ judgeval/judges/together_judge.py,sha256=5FADUhs6-FN1ZVV_1D3-8_gu9mPbZiG0PYTpme41SfM,2336
43
43
  judgeval/judges/utils.py,sha256=0CF9qtIUQUL3-W-qTGpmTjZbkUUBAM6TslDsrCHnTBU,2725
44
44
  judgeval/scorers/__init__.py,sha256=4H_cinTQ4EogZv59YEV-3U9EOTLppNwgAPTi1-jI9Fw,746
45
45
  judgeval/scorers/agent_scorer.py,sha256=TjwD_YglSywr3EowEojiCyg5qDgCRa5LRGc5nFdmIBc,703
@@ -65,7 +65,7 @@ judgeval/tracer/__init__.py,sha256=wkuXtOGDCrwgPPXlh_sSJmvGuWaAMHyNzk1TzB5f9aI,1
65
65
  judgeval/utils/alerts.py,sha256=3w_AjQrgfmOZvfqCridW8WAnHVxHHXokX9jNzVFyGjA,3297
66
66
  judgeval/utils/file_utils.py,sha256=PWHRs8dUr8iDwpglSSk4Yjd7C6ZhDzUaO-jV3m7riHM,1987
67
67
  judgeval/utils/requests.py,sha256=K3gUKrwL6TvwYKVYO5OeLWdUHn9NiUPmnIXhZEiEaHU,1534
68
- judgeval-0.3.0.dist-info/METADATA,sha256=rENldJCo8cVAjLgiYzmYu1RGByxAmXX63WKLc6owrMo,10348
69
- judgeval-0.3.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
70
- judgeval-0.3.0.dist-info/licenses/LICENSE.md,sha256=tKmCg7k5QOmxPK19XMfzim04QiQJPmgIm0pAn55IJwk,11352
71
- judgeval-0.3.0.dist-info/RECORD,,
68
+ judgeval-0.3.2.dist-info/METADATA,sha256=3VDQzNPY0wKnaIpzmrP2OLsk6eEGJZH_dSl4QcJRVwE,10348
69
+ judgeval-0.3.2.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
70
+ judgeval-0.3.2.dist-info/licenses/LICENSE.md,sha256=tKmCg7k5QOmxPK19XMfzim04QiQJPmgIm0pAn55IJwk,11352
71
+ judgeval-0.3.2.dist-info/RECORD,,