hygon.text 1.28 KB
Newer Older
1
2
3
4
5
# Hygon DCU dependencies (based on ROCm/HIP)
# Use VLLM_USE_HYGON=1 to enable this file during build.
#
# Keep the lock set minimal:
# - pin numpy explicitly to avoid ABI-breaking upgrades
6
# - keep Hygon custom packages on base versions here
7
8
9
10
-r common.txt

# Pin numpy to 1.26.x: compatible with both custom torch ABI and
# modern PyPI packages (pandas, opencv, etc.)
11
12
#numpy == 1.26.4
# numpy == 1.26.4 Pinning a specific version of this package will cause dependency conflicts with VLLM, making it impossible to install.
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44

# Replace heavy extras dependency chains with the explicit versions
# already validated in the Hygon Docker image.
fastapi == 0.135.3
uvicorn == 0.42.0
python-multipart == 0.0.22
watchfiles == 1.1.1
mistral_common == 1.11.0

numba == 0.61.2 # Required for N-gram speculative decoding

# Dependencies for Hygon DCU
datasets
ray[cgraph]>=2.48.0 # Ray Compiled Graph, required for pipeline parallelism in V1.
peft
pytest-asyncio
tensorizer==2.10.1
packaging>=24.2
setuptools>=77.0.3,<80.0.0
setuptools-scm>=8
runai-model-streamer[s3,gcs]==0.15.3
timm>=1.0.17
grpcio-tools>=1.76.0

numa
quart
fastrlock==0.8.3

# Hygon DCU hardware-bound packages (base versions)
torch == 2.9.0
triton == 3.3.0
flash_attn == 2.6.1
45
46
#flash_mla == 1.2.0
flash_mla == 1.0.0
47
48
49
lightop == 0.6.0
lmslim == 0.3.1