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, 18:15 UTC
[CI/Build][ROCm] Enabling tensorizer tests for ROCm (#7237)
alexeykondrat committed
Aug 27, 2024 at 13:09 UTC
42e932c7d4c9f0c36227d7eb68fe7b71318bfce6
3 files changed
+4
-1
.buildkite/run-amd-test.sh
+1
@@ -75,6 +75,7 @@ docker run \
75
--network host \
76
--shm-size=16gb \
77
--rm \
78
+ -e HIP_VISIBLE_DEVICES=0 \
79
-e HF_TOKEN \
80
-v ${HF_CACHE}:${HF_MOUNT} \
81
-e HF_HOME=${HF_MOUNT} \
.buildkite/test-pipeline.yaml
+2
-1
@@ -233,12 +233,13 @@ steps:
233
parallelism: 4
234
235
- label: Tensorizer Test # 11min
236
+ mirror_hardwares: [amd]
237
soft_fail: true
238
source_file_dependencies:
239
- vllm/model_executor/model_loader
240
- tests/tensorizer_loader
241
commands:
241
- - apt-get install -y curl libsodium23
242
+ - apt-get update && apt-get install -y curl libsodium23
243
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
244
- pytest -v -s tensorizer_loader
245
requirements-rocm.txt
+1
@@ -8,3 +8,4 @@ botocore
8
ray >= 2.10.0
9
peft
10
pytest-asyncio
11
+tensorizer>=2.9.0
\ No newline at end of file