[CI] Split (H200 MIG/MI300) Basic Correctness into named jobs (#57054)

Signed-off-by: Thang Nguyen <thangnguyenvn647@gmail.com>
Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Andreas Karatzas <Andreas.Karatzas@amd.com>
Co-authored-by: Codex <noreply@openai.com>
This commit is contained in:
Thang Nguyen
2026-09-25 17:05:33 -07:00
committed by GitHub
co-authored by Claude Opus 5.5 Andreas Karatzas Codex
parent 9bf44c4013
commit da1aaec31f
18 changed files with 461 additions and 330 deletions
@@ -0,0 +1,33 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Test prefetch offloading correctness with Llama model."""
from ...utils import compare_two_settings
def test_prefetch_offload_llama():
"""Test prefetch CPU offloading with Llama-3.2-1B-Instruct.
Compares outputs between:
1. Baseline (no offloading)
2. Prefetch offloading (group_size=8, num_in_group=2, prefetch_step=1)
This tests prefetching-based offloading on a dense model.
"""
compare_two_settings(
"meta-llama/Llama-3.2-1B-Instruct",
[
# Prefetch offloading configuration
"--offload-group-size",
"8",
"--offload-num-in-group",
"2",
"--offload-prefetch-step",
"1",
# Selective offloading: only MLP weights
"--offload-params",
"gate_up_proj",
"down_proj",
],
[], # Baseline: no offloading
)