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 16, 17:00 UTC
[CI/Build] Enabling kernels tests for AMD, ignoring some of then that fail (#8130)
Alexey Kondratiev(AMD) committed
Sep 10, 2024 at 14:51 UTC
f421f3cefb58d968767536d745fcc6e9ac342df5
2 files changed
+24
-1
.buildkite/run-amd-test.sh
+23
-1
@@ -71,13 +71,35 @@ mkdir -p ${HF_CACHE}
71
HF_MOUNT="/root/.cache/huggingface"
72
73
commands=$@
74
+echo "Commands:$commands"
75
+#ignore certain kernels tests
76
+if [[ $commands == *" kernels "* ]]; then
77
+ commands="${commands} \
78
+ --ignore=kernels/test_attention.py \
79
+ --ignore=kernels/test_attention_selector.py \
80
+ --ignore=kernels/test_blocksparse_attention.py \
81
+ --ignore=kernels/test_causal_conv1d.py \
82
+ --ignore=kernels/test_cutlass.py \
83
+ --ignore=kernels/test_encoder_decoder_attn.py \
84
+ --ignore=kernels/test_flash_attn.py \
85
+ --ignore=kernels/test_flashinfer.py \
86
+ --ignore=kernels/test_int8_quant.py \
87
+ --ignore=kernels/test_machete_gemm.py \
88
+ --ignore=kernels/test_mamba_ssm.py \
89
+ --ignore=kernels/test_marlin_gemm.py \
90
+ --ignore=kernels/test_prefix_prefill.py \
91
+ --ignore=kernels/test_rand.py \
92
+ --ignore=kernels/test_sampler.py"
93
+fi
94
+
95
PARALLEL_JOB_COUNT=8
96
# check if the command contains shard flag, we will run all shards in parallel because the host have 8 GPUs.
97
if [[ $commands == *"--shard-id="* ]]; then
98
for GPU in $(seq 0 $(($PARALLEL_JOB_COUNT-1))); do
99
#replace shard arguments
79
- commands=${@//"--shard-id= "/"--shard-id=${GPU} "}
100
+ commands=${commands//"--shard-id= "/"--shard-id=${GPU} "}
101
commands=${commands//"--num-shards= "/"--num-shards=${PARALLEL_JOB_COUNT} "}
102
+ echo "Shard ${GPU} commands:$commands"
103
docker run \
104
--device /dev/kfd --device /dev/dri \
105
--network host \
.buildkite/test-pipeline.yaml
+1
@@ -228,6 +228,7 @@ steps:
228
parallelism: 4
229
230
- label: Kernels Test %N # 30min each
231
+ mirror_hardwares: [amd]
232
source_file_dependencies:
233
- csrc/
234
- vllm/attention