feat(catalog): add vertex_ai/gemini-3.5-flash to bundled model catalog#1138
Open
Serhan-Asad wants to merge 1 commit into
Open
feat(catalog): add vertex_ai/gemini-3.5-flash to bundled model catalog#1138Serhan-Asad wants to merge 1 commit into
Serhan-Asad wants to merge 1 commit into
Conversation
#1136) pdd_cloud runs with LLM_INVOKE_DEFAULT_MODEL=vertex_ai/gemini-3.5-flash (commit 6e08eb488b in pdd_cloud). Without this row in the bundled CSV, PDD_MODEL_DEFAULT=vertex_ai/gemini-3.5-flash falls through to the surrogate-base path. Under the provider lock from PR #1115, the surrogate becomes vertex_ai/claude-opus-4-7 (the first Google Vertex AI row) — wrong model, wrong pricing, wrong ELO for strength interpolation. Changes: - pdd/data/llm_model.csv: add Google Vertex AI,vertex_ai/gemini-3.5-flash row. Pricing $1.50/$9.00 matches Google's published Vertex rate and the pdd_cloud worker catalog. ELO 1442 matches the pdd_cloud catalog. Preview row (vertex_ai/gemini-3-flash-preview) is kept for users who pin it. - pdd/generate_model_catalog.py: add vertex_ai/gemini-3.5-flash to _MANDATORY_MODEL_ROWS so catalog regeneration preserves the row even if LiteLLM's model_cost dedup removes it; add "gemini-3.5-flash": 1442 to STATIC_ELO_FALLBACK so the mandatory-row ELO resolver passes the ELO_CUTOFF guard. - tests/test_generate_model_catalog.py: pin both the committed-CSV assertion and the build_rows() seeding path (verified against fake litellm with no gemini-3.5-flash entry). - tests/test_llm_invoke.py: regression that vertex_ai/gemini-3.5-flash resolves directly to the GA row (not the claude-opus-4-7 surrogate trap); pin that the model is in the Gemini 3 temperature-clamp family. Closes #1136. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Google Vertex AI,vertex_ai/gemini-3.5-flashtopdd/data/llm_model.csv. Pricing$1.50/$9.00matches Google's published Vertex rate and the pdd_cloud worker catalog (commit6e08eb488b). ELO 1442 matches pdd_cloud. Preview row (vertex_ai/gemini-3-flash-preview) is kept for backward compat.vertex_ai/gemini-3.5-flashto_MANDATORY_MODEL_ROWSinpdd/generate_model_catalog.pyso catalog regeneration preserves it even if litellm dedup removes it; adds"gemini-3.5-flash": 1442toSTATIC_ELO_FALLBACKto pass theELO_CUTOFFguard in the mandatory-row seeder.Why
pdd_cloudruns withLLM_INVOKE_DEFAULT_MODEL=vertex_ai/gemini-3.5-flash(pdd_cloudorigin/main). Without this row, a localPDD_MODEL_DEFAULT=vertex_ai/gemini-3.5-flashrun falls through to the surrogate-base path. Under the provider lock from PR #1115, the surrogate becomesvertex_ai/claude-opus-4-7(first Google Vertex AI row) — wrong model, wrong pricing, wrong ELO for strength interpolation, wrong temperature-clamp behavior.Test plan
test_committed_csv_includes_vertex_gemini_3_5_flash_ga_default— CSV substring assertiontest_build_rows_includes_vertex_gemini_3_5_flash_ga_default— seeded via_MANDATORY_MODEL_ROWSwith fake litellm that lacks the rowTestSelectModelCandidates::test_issue_1136_vertex_gemini_3_5_flash_resolves_directly— resolves to GA row, notvertex_ai/claude-opus-4-7surrogate trapTestSelectModelCandidates::test_issue_1136_gemini_3_5_flash_in_gemini_3_family_clamp— temperature clamp fires for GA idpytest tests/test_generate_model_catalog.py tests/test_llm_invoke.py tests/test_llm_invoke_csv_model_registration.py tests/test_update_model_costs.py— 344 passedCloses #1136.
🤖 Generated with Claude Code