Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Move the two highest-fire workflows off the ubuntu-latest label onto the same image (#10918)

Daniel Hanfa16bf759dd7
diff --git a/.github/workflows/consolidated-tests-ci.yml b/.github/workflows/consolidated-tests-ci.yml
index 98a538e0326..78b15d4d4af 100644
--- a/.github/workflows/consolidated-tests-ci.yml
+++ b/.github/workflows/consolidated-tests-ci.yml
@@ -30,6 +30,13 @@
# things; if they regress we want a red cell, not a green-with-fail-prints
# cell.
+#
+# runs-on: ubuntu-24.04, not ubuntu-latest. Same image, so the jobs are unchanged; different
+# queue, because nearly all of the org's ~871 queued runs ask for ubuntu-latest. Identical
+# no-op jobs in one matrix: ubuntu-latest still queued past 40 min, every other label under
+# 2 min (#10914). Cost: the image no longer tracks the fleet default, so bump it for 26.04.
+#
+
name: Core
on:
@@ -104,7 +111,7 @@ jobs:
transformers_spec: "__from_pyproject__"
trl_spec: "__from_pyproject__"
name: "Core (${{ matrix.combo.label }})"
- runs-on: ubuntu-latest
+ runs-on: ubuntu-24.04
timeout-minutes: 35
# No job-level or per-step `continue-on-error`. Earlier iterations
# masked real test failures behind green check icons; that lie is
@@ -2356,7 +2363,7 @@ jobs:
# builds the LLAMA_CPP_TARGETS list; the result is independent of the
# HF stack version. Run once, gate the PR.
name: llama.cpp build + smoke
- runs-on: ubuntu-latest
+ runs-on: ubuntu-24.04
timeout-minutes: 25
env:
UNSLOTH_ZOO_REF: ${{ inputs.unsloth_zoo_ref || 'main' }}
diff --git a/.github/workflows/studio-backend-ci.yml b/.github/workflows/studio-backend-ci.yml
index 0762b724a2e..43c30625f3d 100644
--- a/.github/workflows/studio-backend-ci.yml
+++ b/.github/workflows/studio-backend-ci.yml
@@ -16,6 +16,12 @@
# - pytest on 3.13 over studio/backend/tests, with the floor checked statically
# - repo-cpu-tests: auto-discovered tests/ + state-isolated spoof files
#
+#
+# runs-on: ubuntu-24.04, not ubuntu-latest. Same image, so the jobs are unchanged; different
+# queue, because nearly all of the org's ~871 queued runs ask for ubuntu-latest. Identical
+# no-op jobs in one matrix: ubuntu-latest still queued past 40 min, every other label under
+# 2 min (#10914). Cost: the image no longer tracks the fleet default, so bump it for 26.04.
+#
# Whole-repo Python lint (syntax + ruff + debugger-leftover scan)
# moved to the dedicated `Lint CI` workflow (.github/workflows/lint-ci.yml)
# so it fires on every PR rather than only on studio/unsloth/tests
@@ -87,7 +93,7 @@ env:
jobs:
pytest:
name: (Python ${{ matrix.python }})
- runs-on: ubuntu-latest
+ runs-on: ubuntu-24.04
# 45 guards against a hang; it is not a performance budget. The comment this replaces
# claimed 14.1 to 15.0 minutes against a limit of 30, but this step now measures 22.5 to
# 29.2 minutes across the matrix, and it has been cancelling at the limit: across the six
@@ -353,7 +359,7 @@ jobs:
# mocked torch.cuda.is_available so the unsloth import chain
# succeeds on CPU.
name: Repo tests (CPU, ${{ matrix.shard }})
- runs-on: ubuntu-latest
+ runs-on: ubuntu-24.04
# Same reason as the matrix above, and now the same number. The comment this replaces
# claimed 13.4 minutes. Across the last 40 runs of this job on main the pytest step
# alone measured 852s to 1683s (14.2 to 28.1 minutes), and the install plus the four
@@ -638,7 +644,7 @@ jobs:
# No pip install here at all -- the runner's own python3 and bash are what these scripts
# use -- so the job is checkout plus the loop, and it finishes long before the shards do.
name: Shell installer tests
- runs-on: ubuntu-latest
+ runs-on: ubuntu-24.04
# 165s measured, and a hung mock would otherwise sit here until the workflow's own limit.
timeout-minutes: 15
steps: