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

cosminacho · claude · cosminacho · commit a2c7c6e612ab · 2026-04-23T17:49:17.000+03:00
- New src/uipath/llm_client/utils/sampling.py in the core package exports
  DISABLED_SAMPLING_PARAMS, should_skip_sampling, and
  strip_disabled_sampling_kwargs. These are framework-agnostic and fit the
  existing core utils pattern (one file per concern).
- Langchain's utils.py re-exports them, so the public import path
  uipath_langchain_client.utils.strip_disabled_sampling_kwargs is preserved.
- model_details resolution is now a @field_validator("model_details",
  mode="after") collocated with the field declaration. It reads
  already-validated client_settings and model_name off info.data and calls
  get_model_info, instead of living in a separate @model_validator method.
- Core version 1.9.9 -&gt; 1.10.0 with changelog entry; langchain's core-dep
  floor bumped to &gt;=1.10.0.

Co-Authored-By: Claude Opus 4.7 (1M context) &lt;noreply@anthropic.com&gt;
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -2,6 +2,11 @@
 
 All notable changes to `uipath_llm_client` (core package) will be documented in this file.
 
+## [1.10.0] - 2026-04-23
+
+### Added
+- `uipath.llm_client.utils.sampling` module exposing `DISABLED_SAMPLING_PARAMS`, `should_skip_sampling(model_details)`, and `strip_disabled_sampling_kwargs(...)`. Centralizes the gateway's rule that `modelDetails.shouldSkipTemperature=True` implies the full sampling set is rejected (temperature, top_p, top_k, frequency/presence penalty, seed, logit_bias, logprobs, top_logprobs). Framework-agnostic helpers intended for reuse by any wrapper layer.
+
 ## [1.9.9] - 2026-04-23
 
 ### Changed
diff --git a/packages/uipath_langchain_client/CHANGELOG.md b/packages/uipath_langchain_client/CHANGELOG.md
@@ -5,8 +5,11 @@ All notable changes to `uipath_langchain_client` will be documented in this file
 ## [1.10.0] - 2026-04-23
 
 ### Added
-- `UiPathBaseChatModel` now strips sampling kwargs (`temperature`, `top_p`, `top_k`, `frequency_penalty`, `presence_penalty`, `seed`, `logit_bias`, `logprobs`, `top_logprobs`) at invocation time when the model's `modelDetails.shouldSkipTemperature` is true. Fixes `anthropic.claude-opus-4-7` rejecting any sampling parameter passed to `.invoke()` / `.ainvoke()` / streams.
-- `model_details` field on `UiPathBaseLLMClient`, populated eagerly: `get_chat_model` forwards it from the discovery response it already fetches; direct instantiation resolves it in `model_post_init` via `client_settings.get_model_info` (backed by the class-cached discovery response, so at most one network call per process). Each strip logs a warning via `self.logger` when one is configured.
+- `UiPathBaseChatModel` now strips sampling kwargs (`temperature`, `top_p`, `top_k`, `frequency_penalty`, `presence_penalty`, `seed`, `logit_bias`, `logprobs`, `top_logprobs`) at invocation time when the model's `modelDetails.shouldSkipTemperature` is true. Fixes `anthropic.claude-opus-4-7` rejecting any sampling parameter passed to `.invoke()` / `.ainvoke()` / streams. The shared helpers live in the core package at `uipath.llm_client.utils.sampling` and are re-exported from `uipath_langchain_client.utils`.
+- `model_details` field on `UiPathBaseLLMClient`, populated eagerly via a `@field_validator("model_details", mode="after")`: `get_chat_model` forwards it from the discovery response it already fetches; direct instantiation resolves it via `client_settings.get_model_info` (backed by the class-cached discovery response, so at most one network call per process). Each strip logs a warning via `self.logger` when one is configured.
+
+### Changed
+- Bumped `uipath-llm-client` floor to `>=1.10.0` to match the release that adds `uipath.llm_client.utils.sampling`.
 
 ## [1.9.9] - 2026-04-23
 
diff --git a/packages/uipath_langchain_client/pyproject.toml b/packages/uipath_langchain_client/pyproject.toml
@@ -6,7 +6,7 @@ readme = "README.md"
 requires-python = ">=3.11"
 dependencies = [
     "langchain>=1.2.15,<2.0.0",
-    "uipath-llm-client>=1.9.9,<2.0.0",
+    "uipath-llm-client>=1.10.0,<2.0.0",
 ]
 
 [project.optional-dependencies]
diff --git a/packages/uipath_langchain_client/src/uipath_langchain_client/base_client.py b/packages/uipath_langchain_client/src/uipath_langchain_client/base_client.py
@@ -27,7 +27,7 @@
 from abc import ABC
 from collections.abc import AsyncGenerator, Generator, Mapping, Sequence
 from functools import cached_property
-from typing import Any, ClassVar, Literal, Self
+from typing import Any, ClassVar, Literal
 
 from httpx import URL, Response
 from langchain_core.callbacks import (
@@ -38,7 +38,14 @@
 from langchain_core.language_models.chat_models import BaseChatModel
 from langchain_core.messages import BaseMessage
 from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
-from pydantic import AliasChoices, BaseModel, ConfigDict, Field, model_validator
+from pydantic import (
+    AliasChoices,
+    BaseModel,
+    ConfigDict,
+    Field,
+    ValidationInfo,
+    field_validator,
+)
 
 from uipath.llm_client.httpx_client import (
     UiPathHttpxAsyncClient,
@@ -49,13 +56,12 @@
     get_captured_response_headers,
     set_captured_response_headers,
 )
-from uipath_langchain_client._sampling import strip_disabled_sampling_kwargs
 from uipath_langchain_client.settings import (
     UiPathAPIConfig,
     UiPathBaseSettings,
     get_default_client_settings,
 )
-from uipath_langchain_client.utils import RetryConfig
+from uipath_langchain_client.utils import RetryConfig, strip_disabled_sampling_kwargs
 
 
 class UiPathBaseLLMClient(BaseModel, ABC):
@@ -112,10 +118,36 @@ class UiPathBaseLLMClient(BaseModel, ABC):
     model_details: dict[str, Any] | None = Field(
         default=None,
         description="Per-model capability flags sourced from the discovery endpoint "
-        "(e.g. {'shouldSkipTemperature': True}). The factory forwards it; direct "
-        "instantiation lazy-resolves it from client_settings on first construction.",
+        "(e.g. {'shouldSkipTemperature': True}). The factory forwards it; when absent, "
+        "the field validator below eagerly resolves it from client_settings.",
     )
 
+    @field_validator("model_details", mode="after")
+    @classmethod
+    def _resolve_model_details(
+        cls, value: dict[str, Any] | None, info: ValidationInfo
+    ) -> dict[str, Any]:
+        # Fields validate in declaration order, so by the time this runs both
+        # ``client_settings`` and ``model_name`` are already in ``info.data``.
+        # Eager resolution here keeps direct instantiation and the factory
+        # path consistent. ``get_available_models`` is class-cached inside
+        # the settings layer, so at most one discovery HTTP call fires per
+        # process regardless of how many chat/embedding models are built.
+        if value is not None:
+            return value
+        settings = info.data.get("client_settings")
+        model_name = info.data.get("model_name")
+        if settings is None or not model_name:
+            return {}
+        try:
+            model_info = settings.get_model_info(
+                model_name,
+                byo_connection_id=info.data.get("byo_connection_id"),
+            )
+            return model_info.get("modelDetails") or {}
+        except Exception:
+            return {}
+
     default_headers: Mapping[str, str] | None = Field(
         default=None,
         description="Caller-supplied request headers. Merged on top of `class_default_headers`; "
@@ -148,25 +180,6 @@ class UiPathBaseLLMClient(BaseModel, ABC):
         description="Logger for request/response logging",
     )
 
-    @model_validator(mode="after")
-    def _resolve_model_details(self) -> Self:
-        # Populate model_details eagerly so direct instantiation behaves the
-        # same as the factory path. get_available_models is class-cached inside
-        # the settings layer, so at most one discovery HTTP call fires per
-        # process regardless of how many chat/embedding models are built.
-        # Placed on UiPathBaseLLMClient (not just the chat subclass) because
-        # model_details is meaningful for embedding wrappers too.
-        if self.model_details is None:
-            try:
-                info = self.client_settings.get_model_info(
-                    self.model_name,
-                    byo_connection_id=self.byo_connection_id,
-                )
-                self.model_details = info.get("modelDetails") or {}
-            except Exception:
-                self.model_details = {}
-        return self
-
     @cached_property
     def uipath_sync_client(self) -> UiPathHttpxClient:
         """Here we instantiate a synchronous HTTP client with the proper authentication pipeline, retry logic, logging etc."""
diff --git a/packages/uipath_langchain_client/src/uipath_langchain_client/utils.py b/packages/uipath_langchain_client/src/uipath_langchain_client/utils.py
@@ -18,6 +18,11 @@
     is_anthropic_model_name,
 )
 from uipath.llm_client.utils.retry import RetryConfig
+from uipath.llm_client.utils.sampling import (
+    DISABLED_SAMPLING_PARAMS,
+    should_skip_sampling,
+    strip_disabled_sampling_kwargs,
+)
 
 __all__ = [
     "RetryConfig",
@@ -36,4 +41,7 @@
     "UiPathTooManyRequestsError",
     "ANTHROPIC_MODEL_NAME_KEYWORDS",
     "is_anthropic_model_name",
+    "DISABLED_SAMPLING_PARAMS",
+    "should_skip_sampling",
+    "strip_disabled_sampling_kwargs",
 ]
diff --git a/src/uipath/llm_client/__version__.py b/src/uipath/llm_client/__version__.py
@@ -1,3 +1,3 @@
 __title__ = "UiPath LLM Client"
 __description__ = "A Python client for interacting with UiPath's LLM services."
-__version__ = "1.9.9"
+__version__ = "1.10.0"
diff --git a/src/uipath/llm_client/utils/sampling.py b/src/uipath/llm_client/utils/sampling.py
@@ -3,8 +3,9 @@
 Reasoning-style models (e.g. ``anthropic.claude-opus-4-7``) advertise
 ``modelDetails.shouldSkipTemperature: true`` on the discovery endpoint. When
 that flag is set, the gateway rejects the entire sampling set, not just
-``temperature``. The helpers here centralize that knowledge so every chat
-wrapper can reuse them via ``UiPathBaseChatModel``.
+``temperature``. The helpers here centralize that knowledge so every framework
+wrapper (LangChain chat models, future LlamaIndex wrappers, the core
+normalized client, etc.) can reuse the same rule.
 """
 
 from __future__ import annotations
diff --git a/tests/langchain/test_disabled_sampling_params.py b/tests/langchain/test_disabled_sampling_params.py
@@ -14,9 +14,9 @@
 import pytest
 from langchain_core.messages import AIMessage
 from langchain_core.outputs import ChatGeneration, ChatResult
-from uipath_langchain_client._sampling import DISABLED_SAMPLING_PARAMS
 from uipath_langchain_client.clients.normalized.chat_models import UiPathChat
 from uipath_langchain_client.factory import get_chat_model
+from uipath_langchain_client.utils import DISABLED_SAMPLING_PARAMS
 
 from uipath.llm_client.settings import UiPathBaseSettings
 

Original file line number	Diff line number	Diff line change
`@@ -6,7 +6,7 @@ readme = "README.md"`
`6`	`6`	`requires-python = ">=3.11"`
`7`	`7`	`dependencies = [`
`8`	`8`	`"langchain>=1.2.15,<2.0.0",`
`9`		`- "uipath-llm-client>=1.9.9,<2.0.0",`
	`9`	`+ "uipath-llm-client>=1.10.0,<2.0.0",`
`10`	`10`	`]`
`11`	`11`
`12`	`12`	`[project.optional-dependencies]`