UN-3709 [FIX] Strip deprecated temperature for Claude Sonnet 5 and post-4.7 models#2165
Open
jaseemjaskp wants to merge 1 commit into
Open
UN-3709 [FIX] Strip deprecated temperature for Claude Sonnet 5 and post-4.7 models#2165jaseemjaskp wants to merge 1 commit into
jaseemjaskp wants to merge 1 commit into
Conversation
…odels Configuring Claude Sonnet 5 via Azure AI Foundry (azure_ai adapter) failed the connection test with "temperature is deprecated for this model". Anthropic removed the sampling params (temperature/top_p/top_k) starting with Opus 4.7, and every model since (Opus 4.8, Sonnet 5, Fable 5, Mythos 5) rejects them too. The SDK1 sampling-param strip only recognized claude-opus-4-7, so for Sonnet 5 the strip never fired and Pydantic's default temperature=0.1 was forwarded to Azure -> Anthropic -> 400. Replace the single detection pattern with a maintained stem list covering all sampling-deprecated Claude models, keeping the trailing-edge boundary that rejects prefix collisions (claude-sonnet-50) and alpha continuations (...-5verbose). Extend tests with positives, boundary negatives, and adapter wiring for Sonnet 5 (including the reported azure_ai_foundry case).
Contributor
WalkthroughThe Claude sampling-parameter detection now uses an allowlist of model stems covering Opus 4.7 and later releases. Tests add positive, collision-negative, and adapter validation coverage for Sonnet 5 and other supported models. ChangesClaude sampling detection
Estimated code review effort: 3 (Moderate) | ~20 minutes 🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
|
| Filename | Overview |
|---|---|
| unstract/sdk1/src/unstract/sdk1/adapters/base1.py | Extends the existing model allowlist without changing normalization, stripping behavior, or adapter integration. |
| unstract/sdk1/tests/test_sampling_strip.py | Adds coverage for new model families, provider identifier formats, boundary collisions, and adapter validation. |
Reviews (1): Last reviewed commit: "[FIX] Strip deprecated temperature for C..." | Re-trigger Greptile
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
🧹 Nitpick comments (1)
unstract/sdk1/tests/test_sampling_strip.py (1)
330-358: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueConsider adding a Bedrock cross-region profile positive case.
The
POST_47_POSITIVESlist covers Bedrock foundation model ids (anthropic.claude-sonnet-5-...) but not cross-region profile variants (us.anthropic.claude-sonnet-5-...,eu.,apac.,global.) that are explicitly documented inbase1.pylines 34-35. The regex would match these (leading text passes through,-delimiter after stem), but an explicit test case would guard against future anchor changes.➕ Suggested addition
"anthropic.claude-sonnet-5-20260101-v1:0", # Bedrock foundation model id + "us.anthropic.claude-sonnet-5-20260101-v1:0", # Bedrock cross-region profile + "global.anthropic.claude-sonnet-5-20260101-v1:0", # Bedrock global profile "vertex_ai/claude-sonnet-5@20260101", # Vertex AI🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@unstract/sdk1/tests/test_sampling_strip.py` around lines 330 - 358, Add a positive Bedrock cross-region profile test case to the existing Sonnet 5 validation coverage, using representative us., eu., apac., or global. model IDs alongside the foundation ID in the relevant POST_47_POSITIVES data. Ensure the case exercises the Bedrock validator and confirms all _DEPRECATED_SAMPLING_PARAMS are stripped, preserving coverage through the existing parameterized test.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Nitpick comments:
In `@unstract/sdk1/tests/test_sampling_strip.py`:
- Around line 330-358: Add a positive Bedrock cross-region profile test case to
the existing Sonnet 5 validation coverage, using representative us., eu., apac.,
or global. model IDs alongside the foundation ID in the relevant
POST_47_POSITIVES data. Ensure the case exercises the Bedrock validator and
confirms all _DEPRECATED_SAMPLING_PARAMS are stripped, preserving coverage
through the existing parameterized test.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 2f93c006-7ce3-4afb-b7ca-e0b6b6307334
📒 Files selected for processing (2)
unstract/sdk1/src/unstract/sdk1/adapters/base1.pyunstract/sdk1/tests/test_sampling_strip.py
Contributor
Unstract test resultsPer-group results
Critical paths
|
jaseemjaskp
changed the title
pk-zipstack
approved these changes
Jul 11, 2026
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
2 participants
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.
What
Extend the SDK1 sampling-parameter strip so it also fires for Claude Sonnet 5 and the other Claude models that deprecated
temperature/top_p/top_k(Opus 4.8, Fable 5, Mythos 5), not just Opus 4.7.Why
Configuring Claude Sonnet 5 via Azure AI Foundry (
azure_aiadapter) failed the connection test with:Anthropic removed the sampling params (
temperature/top_p/top_k) starting with Claude Opus 4.7, and every model released since — Opus 4.8, Sonnet 5, Fable 5, Mythos 5 — rejects them with a 400 (directly and through the Bedrock / Azure AI Foundry / Vertex proxies). Opus 4.6 / Sonnet 4.6 and older still accept them.SDK1 already strips these params via
_strip_deprecated_sampling_params()(wired into every adapter'svalidate()), but the detection allowlist only matchedclaude-opus-4-7. So for Sonnet 5 the strip never fired, and becauseBaseChatCompletionParameters.temperaturedefaults to0.1, Pydantic'smodel_dump()re-emittedtemperature=0.1, which litellm forwarded to Azure → Anthropic → the 400.How
unstract/sdk1/src/unstract/sdk1/adapters/base1.py:_SAMPLING_DEPRECATED_MODEL_PATTERNStuple with a maintained_SAMPLING_DEPRECATED_MODEL_STEMSlist (claude-opus-4-7,claude-opus-4-8,claude-sonnet-5,claude-fable-5,claude-mythos-5), each compiled with the existing trailing-edge boundary(?=$|[-:@/]|v\d)that rejects prefix collisions (claude-sonnet-50) and alpha continuations (...-5verbose).The strip wiring, the per-adapter integration (including
AzureAIFoundryLLMParameters.validate), and theazure_aipath are unchanged — only the detection list grows.unstract/sdk1/tests/test_sampling_strip.py: add positives for the newly-covered models (native, Azure-prefixed, deployment-name, case, separator, version-tag, Bedrock, Vertex forms), boundary negatives (claude-sonnet-50,claude-sonnet-4-5/4-6), and adapter-wiring tests including the exact reported scenario.Can this PR break any existing features. If yes, please list possible items. If no, please explain why. (PS: Admins do not merge the PR without this section filled)
No. The change only adds model stems to a narrow detection allowlist; models that already accept sampling params (Opus 4.6/4.5, Sonnet 4.5/4.6, Haiku, non-Anthropic providers) are unaffected — verified by existing and newly-added negative tests. The trailing-edge boundary is preserved, so no new false positives (
claude-sonnet-50,claude-sonnet-4-5) are introduced.Database Migrations
Env Config
Relevant Docs
Related Issues or PRs
Dependencies Versions
Notes on Testing
pytest tests/test_sampling_strip.py— 87 passed (pre-commit selected-tests hook also green).AzureAIFoundryLLMParameters.validate({"model": "claude-sonnet-5", "temperature": 0.1, ...})returnsmodel == "azure_ai/claude-sonnet-5"withtemperatureabsent.model/deployment string. Azure AI Foundry deployments with an opaque name (e.g.sonnet5-prod) must be named to includeclaude-sonnet-5for the strip to fire.Screenshots
Checklist
I have read and understood the Contribution Guidelines.