judgeval 0.1.0__tar.gz → 0.2.0__tar.gz

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.
Files changed (111) hide show
  1. {judgeval-0.1.0 → judgeval-0.2.0}/PKG-INFO +1 -1
  2. {judgeval-0.1.0 → judgeval-0.2.0}/pyproject.toml +1 -1
  3. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/run_evaluation.py +7 -0
  4. {judgeval-0.1.0 → judgeval-0.2.0}/.github/ISSUE_TEMPLATE/bug_report.md +0 -0
  5. {judgeval-0.1.0 → judgeval-0.2.0}/.github/ISSUE_TEMPLATE/config.yml +0 -0
  6. {judgeval-0.1.0 → judgeval-0.2.0}/.github/ISSUE_TEMPLATE/feature_request.md +0 -0
  7. {judgeval-0.1.0 → judgeval-0.2.0}/.github/pull_request_template.md +0 -0
  8. {judgeval-0.1.0 → judgeval-0.2.0}/.github/workflows/blocked-pr.yaml +0 -0
  9. {judgeval-0.1.0 → judgeval-0.2.0}/.github/workflows/ci.yaml +0 -0
  10. {judgeval-0.1.0 → judgeval-0.2.0}/.github/workflows/lint.yaml +0 -0
  11. {judgeval-0.1.0 → judgeval-0.2.0}/.github/workflows/merge-branch-check.yaml +0 -0
  12. {judgeval-0.1.0 → judgeval-0.2.0}/.github/workflows/release.yaml +0 -0
  13. {judgeval-0.1.0 → judgeval-0.2.0}/.github/workflows/validate-branch.yaml +0 -0
  14. {judgeval-0.1.0 → judgeval-0.2.0}/.gitignore +0 -0
  15. {judgeval-0.1.0 → judgeval-0.2.0}/.pre-commit-config.yaml +0 -0
  16. {judgeval-0.1.0 → judgeval-0.2.0}/LICENSE.md +0 -0
  17. {judgeval-0.1.0 → judgeval-0.2.0}/README.md +0 -0
  18. {judgeval-0.1.0 → judgeval-0.2.0}/assets/Screenshot 2025-05-17 at 8.14.27/342/200/257PM.png" +0 -0
  19. {judgeval-0.1.0 → judgeval-0.2.0}/assets/agent.gif +0 -0
  20. {judgeval-0.1.0 → judgeval-0.2.0}/assets/data.gif +0 -0
  21. {judgeval-0.1.0 → judgeval-0.2.0}/assets/dataset_clustering_screenshot.png +0 -0
  22. {judgeval-0.1.0 → judgeval-0.2.0}/assets/dataset_clustering_screenshot_dm.png +0 -0
  23. {judgeval-0.1.0 → judgeval-0.2.0}/assets/datasets_preview_screenshot.png +0 -0
  24. {judgeval-0.1.0 → judgeval-0.2.0}/assets/document.gif +0 -0
  25. {judgeval-0.1.0 → judgeval-0.2.0}/assets/error_analysis_dashboard.png +0 -0
  26. {judgeval-0.1.0 → judgeval-0.2.0}/assets/experiments_dashboard_screenshot.png +0 -0
  27. {judgeval-0.1.0 → judgeval-0.2.0}/assets/experiments_page.png +0 -0
  28. {judgeval-0.1.0 → judgeval-0.2.0}/assets/experiments_pagev2.png +0 -0
  29. {judgeval-0.1.0 → judgeval-0.2.0}/assets/logo-dark.svg +0 -0
  30. {judgeval-0.1.0 → judgeval-0.2.0}/assets/logo-light.svg +0 -0
  31. {judgeval-0.1.0 → judgeval-0.2.0}/assets/monitoring_screenshot.png +0 -0
  32. {judgeval-0.1.0 → judgeval-0.2.0}/assets/new_darkmode.svg +0 -0
  33. {judgeval-0.1.0 → judgeval-0.2.0}/assets/new_lightmode.svg +0 -0
  34. {judgeval-0.1.0 → judgeval-0.2.0}/assets/product_shot.png +0 -0
  35. {judgeval-0.1.0 → judgeval-0.2.0}/assets/trace.gif +0 -0
  36. {judgeval-0.1.0 → judgeval-0.2.0}/assets/trace_demo.png +0 -0
  37. {judgeval-0.1.0 → judgeval-0.2.0}/assets/trace_screenshot.png +0 -0
  38. {judgeval-0.1.0 → judgeval-0.2.0}/assets/trace_screenshot_old.png +0 -0
  39. {judgeval-0.1.0 → judgeval-0.2.0}/pytest.ini +0 -0
  40. {judgeval-0.1.0 → judgeval-0.2.0}/src/.coveragerc +0 -0
  41. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/__init__.py +0 -0
  42. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/clients.py +0 -0
  43. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/__init__.py +0 -0
  44. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/api/__init__.py +0 -0
  45. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/api/api.py +0 -0
  46. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/api/constants.py +0 -0
  47. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/exceptions.py +0 -0
  48. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/logger.py +0 -0
  49. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/storage/__init__.py +0 -0
  50. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/storage/s3_storage.py +0 -0
  51. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/tracer/__init__.py +0 -0
  52. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/tracer/constants.py +0 -0
  53. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/tracer/core.py +0 -0
  54. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/tracer/otel_exporter.py +0 -0
  55. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/tracer/otel_span_processor.py +0 -0
  56. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/tracer/span_processor.py +0 -0
  57. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/tracer/span_transformer.py +0 -0
  58. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/tracer/trace_manager.py +0 -0
  59. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/common/utils.py +0 -0
  60. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/constants.py +0 -0
  61. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/__init__.py +0 -0
  62. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/datasets/__init__.py +0 -0
  63. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/datasets/dataset.py +0 -0
  64. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/datasets/eval_dataset_client.py +0 -0
  65. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/example.py +0 -0
  66. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/judgment_types.py +0 -0
  67. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/result.py +0 -0
  68. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/scorer_data.py +0 -0
  69. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/scripts/fix_default_factory.py +0 -0
  70. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/scripts/openapi_transform.py +0 -0
  71. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/tool.py +0 -0
  72. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/trace.py +0 -0
  73. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/data/trace_run.py +0 -0
  74. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/evaluation_run.py +0 -0
  75. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/integrations/langgraph.py +0 -0
  76. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/judges/__init__.py +0 -0
  77. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/judges/base_judge.py +0 -0
  78. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/judges/litellm_judge.py +0 -0
  79. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/judges/mixture_of_judges.py +0 -0
  80. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/judges/together_judge.py +0 -0
  81. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/judges/utils.py +0 -0
  82. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/judgment_client.py +0 -0
  83. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/rules.py +0 -0
  84. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/__init__.py +0 -0
  85. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/agent_scorer.py +0 -0
  86. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/api_scorer.py +0 -0
  87. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/base_scorer.py +0 -0
  88. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/example_scorer.py +0 -0
  89. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/exceptions.py +0 -0
  90. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/__init__.py +0 -0
  91. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/__init__.py +0 -0
  92. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/answer_correctness.py +0 -0
  93. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/answer_relevancy.py +0 -0
  94. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/derailment_scorer.py +0 -0
  95. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/execution_order.py +0 -0
  96. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/faithfulness.py +0 -0
  97. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/hallucination.py +0 -0
  98. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/instruction_adherence.py +0 -0
  99. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/prompt_scorer.py +0 -0
  100. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/tool_dependency.py +0 -0
  101. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/judgeval_scorers/api_scorers/tool_order.py +0 -0
  102. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/score.py +0 -0
  103. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/scorers/utils.py +0 -0
  104. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/tracer/__init__.py +0 -0
  105. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/utils/alerts.py +0 -0
  106. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/utils/file_utils.py +0 -0
  107. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/utils/requests.py +0 -0
  108. {judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/version_check.py +0 -0
  109. {judgeval-0.1.0 → judgeval-0.2.0}/src/update_types.sh +0 -0
  110. {judgeval-0.1.0 → judgeval-0.2.0}/update_version.py +0 -0
  111. {judgeval-0.1.0 → judgeval-0.2.0}/uv.lock +0 -0
{judgeval-0.1.0 → judgeval-0.2.0}/PKG-INFO RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: judgeval
3
- Version: 0.1.0
3
+ Version: 0.2.0
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
{judgeval-0.1.0 → judgeval-0.2.0}/pyproject.toml RENAMED
@@ -1,6 +1,6 @@
1
1
  [project]
2
2
  name = "judgeval"
3
- version = "0.1.0"
3
+ version = "0.2.0"
4
4
  authors = [
5
5
  { name="Andrew Li", email="andrew@judgmentlabs.ai" },
6
6
  { name="Alex Shan", email="alex@judgmentlabs.ai" },
{judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/run_evaluation.py RENAMED
@@ -292,6 +292,13 @@ def run_trace_eval(
292
292
  # This is a callback handler, get the underlying tracer
293
293
  actual_tracer = tracer.tracer
294
294
 
295
+ if trace_run.project_name != actual_tracer.project_name:
296
+ raise ValueError(
297
+ f"Project name mismatch between run_trace_eval and tracer. "
298
+ f"Trace run: {trace_run.project_name}, "
299
+ f"Tracer: {actual_tracer.project_name}"
300
+ )
301
+
295
302
  actual_tracer.offline_mode = True
296
303
  actual_tracer.traces = []
297
304
  judgeval_logger.info("Running agent function: ")
{judgeval-0.1.0 → judgeval-0.2.0}/.gitignore RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/LICENSE.md RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/README.md RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/assets/agent.gif RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/assets/data.gif RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/assets/document.gif RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/assets/logo-dark.svg RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/assets/logo-light.svg RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/assets/trace.gif RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/assets/trace_demo.png RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/pytest.ini RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/src/.coveragerc RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/src/judgeval/rules.py RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/src/update_types.sh RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/update_version.py RENAMED
File without changes
{judgeval-0.1.0 → judgeval-0.2.0}/uv.lock RENAMED
File without changes