Feat: strip sampling params when modelDetails.shouldSkipTemperature is true#74

Merged

cosminacho merged 14 commits into

mainfrom

fix/skip-disabled-sampling-params

Apr 23, 2026

Commits on Apr 23, 2026

Feat: strip sampling params at invoke time when modelDetails.shouldSkipTemperature is true

cosminacho
and
claude
committed
Refactor: populate model_details in model_post_init instead of lazy on first invoke

cosminacho
and
claude
committed
Refactor: extract sampling-strip helpers + move model_details resolution to UiPathBaseLLMClient

cosminacho
and
claude
committed
Refactor: move sampling helpers to core utils + resolve model_details via field_validator

cosminacho
and
claude
committed
Nit: tighten the model_details field_validator

cosminacho
and
claude
committed
Revert to @model_validator for model_details + inline strip_disabled_sampling_kwargs calls

cosminacho
and
claude
committed
Drop sampling + model_family re-exports from langchain utils.py; import from core directly

cosminacho
and
claude
committed
Chore: drop from __future__ import annotations across the repo

cosminacho
and
claude
committed
Generalize via disabled_params + mode="before" validator (init-time clearing, langchain-openai compatible)

cosminacho
and
claude
committed
Switch disabled_params validator from mode=before to mode=after

cosminacho
and
claude
committed
Trim validator to metadata resolution only; merge user-provided disabled_params

cosminacho
and
claude
committed
Rename metadata hook to setup_model_info

cosminacho
and
claude
committed
Add OpenAI / Azure interop tests for disabled_params merging

cosminacho
and
claude
committed
Nit: use settings= (pydantic alias) instead of client_settings= in new tests

cosminacho
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Feat: strip sampling params when modelDetails.shouldSkipTemperature is true#74

Feat: strip sampling params when modelDetails.shouldSkipTemperature is true#74
cosminacho merged 14 commits into
mainfrom
fix/skip-disabled-sampling-params

Commits on Apr 23, 2026

Feat: strip sampling params at invoke time when modelDetails.shouldSkipTemperature is true

Refactor: populate model_details in model_post_init instead of lazy on first invoke

Refactor: extract sampling-strip helpers + move model_details resolution to UiPathBaseLLMClient

Refactor: move sampling helpers to core utils + resolve model_details via field_validator

Nit: tighten the model_details field_validator

Revert to @model_validator for model_details + inline strip_disabled_sampling_kwargs calls

Drop sampling + model_family re-exports from langchain utils.py; import from core directly

Chore: drop from future import annotations across the repo

Generalize via disabled_params + mode="before" validator (init-time clearing, langchain-openai compatible)

Switch disabled_params validator from mode=before to mode=after

Trim validator to metadata resolution only; merge user-provided disabled_params

Rename metadata hook to setup_model_info

Add OpenAI / Azure interop tests for disabled_params merging

Nit: use settings= (pydantic alias) instead of client_settings= in new tests