feat: add Granite 4.2 model defaults - #1587

Open

planetf1 wants to merge 24 commits into

generative-computing:mainfrom

planetf1:feat/granite-4-2-defaults

Commits on Aug 28, 2026

test: point Granite 4.2 eval/CI test fixtures at ibm-granite/granite-4.2-*
planetf1
committed
a8b236e
fix: use published Granite 4.2 Ollama tags
planetf1
committed
7c72ad0
fix: create vLLM environments with Python 3.12
planetf1
committed
8394342
fix: isolate Ollama and vLLM test lanes
planetf1
committed
9279d26
fix: reuse Ollama test URL resolver
planetf1
committed
64e76ff
fix: preserve custom pytest arguments in test runner
planetf1
committed
e7b7083
fix: provision compatible Ollama test models
planetf1
committed
04fc32a
fix: make Ollama telemetry test failures retryable and diagnosable in CI
planetf1
committed
127d36f
fix: capture the whole-run request story from the Ollama server log
planetf1
committed
c3f4962
fix: bound live telemetry test output and retry the stream-guard stall shape
planetf1
committed
21c41ce
fix: bound total streaming request time in live telemetry tests
planetf1
committed
a7aead4
fix: disable SDK auto-retries and bound the live OpenAI test client
planetf1
committed
413f44a
fix: send the live-test output cap as max_tokens and bound avalue()
planetf1
committed
8fba775
docs: use single backticks in the new regression docstring
planetf1
committed
91f7826
fix: constrain granite4.2:3b context in CI to stop Ollama stalls
planetf1
committed
23cdf5d
fix: bump CI Ollama pin 0.32.2 -> 0.33.1 to chase the 4.2 stall
planetf1
committed
ebd5d08
fix: disable granite4.2 thinking on /v1 via reasoning_effort=none
planetf1
committed
c426fe9
test: instrument CI Ollama to diagnose the sustained 300 s stalls
planetf1
committed
162f1a4
test: fix stateless-context telemetry tests and align 4.2 num_ctx
planetf1
committed
e35910c
test: make streaming chunk-timing assertion robust to short replies
planetf1
committed
2ce9ea3
fix: scope reasoning_effort=none to non-OpenAI targets, fix review findings
planetf1
committed
b667812

Navigation Menu

Uh oh!

feat: add Granite 4.2 model defaults - #1587

feat: add Granite 4.2 model defaults#1587
planetf1 wants to merge 24 commits into
generative-computing:mainfrom
planetf1:feat/granite-4-2-defaults

Commits on Aug 28, 2026

test: point Granite 4.2 eval/CI test fixtures at ibm-granite/granite-4.2-*

fix: use published Granite 4.2 Ollama tags

fix: create vLLM environments with Python 3.12

fix: isolate Ollama and vLLM test lanes

fix: reuse Ollama test URL resolver

fix: preserve custom pytest arguments in test runner

fix: provision compatible Ollama test models

fix: make Ollama telemetry test failures retryable and diagnosable in CI

fix: capture the whole-run request story from the Ollama server log

fix: bound live telemetry test output and retry the stream-guard stall shape

fix: bound total streaming request time in live telemetry tests

fix: disable SDK auto-retries and bound the live OpenAI test client

fix: send the live-test output cap as max_tokens and bound avalue()

docs: use single backticks in the new regression docstring

fix: constrain granite4.2:3b context in CI to stop Ollama stalls

fix: bump CI Ollama pin 0.32.2 -> 0.33.1 to chase the 4.2 stall

fix: disable granite4.2 thinking on /v1 via reasoning_effort=none

test: instrument CI Ollama to diagnose the sustained 300 s stalls

test: fix stateless-context telemetry tests and align 4.2 num_ctx

test: make streaming chunk-timing assertion robust to short replies

fix: scope reasoning_effort=none to non-OpenAI targets, fix review findings

Commits on Aug 29, 2026

docs: trim verbose CI/backend comments to load-bearing rationale

docs: trim verbose CI-diagnostics comments in conftest.py

refactor(backends): dedup THINKING-to-param mapping in OpenAIBackend