A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 7, 06:30 UTC
[ROCm][CI] Added MI325 mirrors (stage C) (#35239)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Mar 4, 2026 at 16:48 UTC
6c21a0c2d75a716fa0b8bcf90b68dd46d2bc7265
4 files changed
+35
.buildkite/test_areas/entrypoints.yaml
+15
@@ -41,6 +41,11 @@ steps:
41
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
42
- pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/test_oot_registration.py --ignore=entrypoints/openai/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses
43
- pytest -v -s entrypoints/test_chat_utils.py
44
+ mirror:
45
+ amd:
46
+ device: mi325_1
47
+ depends_on:
48
+ - image-build-amd
49
50
- label: Entrypoints Integration (API Server 2)
51
timeout_in_minutes: 130
@@ -55,6 +60,11 @@ steps:
60
- pytest -v -s entrypoints/instrumentator
61
- PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/rpc
62
- pytest -v -s tool_use
63
+ mirror:
64
+ amd:
65
+ device: mi325_1
66
+ depends_on:
67
+ - image-build-amd
68
69
- label: Entrypoints Integration (Pooling)
70
timeout_in_minutes: 50
@@ -87,6 +97,11 @@ steps:
97
- tests/v1
98
commands:
99
- pytest -v -s v1/entrypoints
100
+ mirror:
101
+ amd:
102
+ device: mi325_1
103
+ depends_on:
104
+ - image-build-amd
105
106
- label: OpenAI API Correctness
107
timeout_in_minutes: 30
.buildkite/test_areas/misc.yaml
+5
@@ -87,6 +87,11 @@ steps:
87
- python3 offline_inference/spec_decode.py --test --method eagle --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 2048
88
# https://github.com/vllm-project/vllm/pull/26682 uses slightly more memory in PyTorch 2.9+ causing this test to OOM in 1xL4 GPU
89
- python3 offline_inference/spec_decode.py --test --method eagle3 --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 1536
90
+ mirror:
91
+ amd:
92
+ device: mi325_1
93
+ depends_on:
94
+ - image-build-amd
95
96
- label: Metrics, Tracing (2 GPUs)
97
timeout_in_minutes: 20
.buildkite/test_areas/models_multimodal.yaml
+10
@@ -12,6 +12,11 @@ steps:
12
- pip freeze | grep -E 'torch'
13
- pytest -v -s models/multimodal -m core_model --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/processing
14
- cd .. && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest -v -s tests/models/multimodal/generation/test_whisper.py -m core_model # Otherwise, mp_method="spawn" doesn't work
15
+ mirror:
16
+ amd:
17
+ device: mi325_1
18
+ depends_on:
19
+ - image-build-amd
20
21
- label: Multi-Modal Processor Test (CPU)
22
depends_on:
@@ -54,6 +59,11 @@ steps:
59
commands:
60
- pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
61
- pytest -v -s models/multimodal -m 'not core_model' --ignore models/multimodal/generation/test_common.py --ignore models/multimodal/processing
62
+ mirror:
63
+ amd:
64
+ device: mi325_1
65
+ depends_on:
66
+ - image-build-amd
67
68
- label: Multi-Modal Models (Extended) 2
69
optional: true
.buildkite/test_areas/plugins.yaml
+5
@@ -39,3 +39,8 @@ steps:
39
- pytest -v -s entrypoints/openai/test_oot_registration.py # it needs a clean process
40
- pytest -v -s models/test_oot_registration.py # it needs a clean process
41
- pytest -v -s plugins/lora_resolvers # unit tests for in-tree lora resolver plugins
42
+ mirror:
43
+ amd:
44
+ device: mi325_2
45
+ depends_on:
46
+ - image-build-amd