-
[OPIK-5845] [BE] fix: fix dataset expansion for test suites and all LLM providers (#6277)
frostbyte_neo 发布于
2026-04-15 15:02:12 +00:00 - [OPIK-5845] [BE] fix: fix dataset expansion for test suites and all LLM providers
- Add maxCompletionTokens (4000) to expansion requests, fixing Anthropic
models that require this field - Re-throw ClientErrorException/ServerErrorException with original
messages instead of wrapping in generic BadRequestException - Skip _generated/_generation_model metadata for test suite expansions
so synthetic fields don't pollute data passed to agents in local
runner scenarios
Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com
- refactor(expansion): use proper HTTP status codes for error handling
- Catch BadRequestException separately to preserve validation errors
- Rethrow ClientErrorException/ServerErrorException from LLM providers
- Use InternalServerErrorException for unexpected failures instead of
mapping everything to 400 BadRequest - Use generic message for 500s, log full details server-side
Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com
- fix(expansion): address PR review comments
- Use lighter getById() instead of findById() to avoid unnecessary
dataset enrichment when only the type is needed - Extract buildDatasetItem() to DRY the duplicated item construction
- Make maxCompletionTokens provider-aware: only set for Anthropic by
default (4000), skip for other providers to avoid impacting results - Allow users to override maxCompletionTokens via the API request
- Inject LlmProviderFactory to resolve provider from model name
Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com
- test(expansion): add unit tests and @Min validation for maxCompletionTokens
Add DatasetExpansionServiceTest covering maxCompletionTokens resolution,
dataset item building (metadata for regular vs test suite), and error
handling. Add @Min(100) validation on maxCompletionTokens field.Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com
Co-authored-by: Claude Opus 4.6 noreply@anthropic.com
下载附件