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 1, 23:45 UTC
[ROCm] Bump fastsafetensors to v0.3.2 from PyPI, remove git source build (#43625)
Signed-off-by: Turner Jabbour <doubleujabbour@gmail.com>
Turner Jabbour committed
Jun 4, 2026 at 15:30 UTC
0c96dd64fb6a8005b1ba5e2ef335fe4f75d9eb90
8 files changed
+14
-9
requirements/cuda.txt
+1
-1
@@ -18,7 +18,7 @@ tilelang==0.1.9
18
nvidia-cudnn-frontend>=1.13.0,<1.19.0
19
20
# Required for faster safetensors model loading
21
-fastsafetensors >= 0.2.2
21
+fastsafetensors >= 0.3.2
22
23
# QuACK and Cutlass DSL for FA4 (cute-DSL implementation)
24
nvidia-cutlass-dsl[cu13]==4.5.2
requirements/rocm.txt
+4
-1
@@ -19,7 +19,10 @@ setuptools-rust>=1.9.0
19
runai-model-streamer[s3,gcs,azure]==0.15.7
20
conch-triton-kernels==1.2.1
21
timm>=1.0.17
22
-# amd-quark: required for Quark quantization on ROCm
22
+# amd-quark: required for Quark quantization on ROCm
23
# To be consistent with test_quark.py
24
amd-quark>=0.8.99
25
tilelang==0.1.10
26
+
27
+# Required for faster safetensors model loading
28
+fastsafetensors >= 0.3.2
requirements/test/cuda.in
+1
-1
@@ -57,7 +57,7 @@ arctic-inference == 0.1.1; platform_machine == "x86_64" # Required for suffix de
57
numba == 0.65.0 # Required for N-gram speculative decoding
58
numpy
59
runai-model-streamer[s3,gcs,azure]==0.15.7
60
-fastsafetensors>=0.2.2; platform_machine == "x86_64" # 0.2.2 contains important fixes for multi-GPU mem usage
60
+fastsafetensors>=0.3.2
61
instanttensor>=0.1.5; platform_machine == "x86_64"
62
pydantic>=2.12 # 2.11 leads to error on python 3.13
63
decord==0.6.0; platform_machine == "x86_64"
requirements/test/cuda.txt
+1
-1
@@ -191,7 +191,7 @@ fastparquet==2024.11.0
191
# via genai-perf
192
fastrlock==0.8.2
193
# via cupy-cuda12x
194
-fastsafetensors==0.2.2
194
+fastsafetensors==0.3.2
195
# via
196
# -c requirements/cuda.txt
197
# -r requirements/test/cuda.in
requirements/test/nightly-torch.txt
+1
-1
@@ -43,6 +43,6 @@ tritonclient>=2.51.0
43
numba == 0.65.0 # Required for N-gram speculative decoding
44
numpy
45
runai-model-streamer[s3,gcs,azure]==0.15.7
46
-fastsafetensors>=0.2.2
46
+fastsafetensors>=0.3.2
47
instanttensor>=0.1.5
48
pydantic>=2.12 # 2.11 leads to error on python 3.13
requirements/test/rocm.in
+1
-1
@@ -56,7 +56,7 @@ arctic-inference==0.1.1 # Required for suffix decoding test
56
numba==0.65.0 # Required for N-gram speculative decoding
57
numpy
58
runai-model-streamer[s3,gcs,azure]==0.15.7
59
-fastsafetensors @ git+https://github.com/foundation-model-stack/fastsafetensors.git@0.2.2 # PyPI only ships CUDA wheels
59
+fastsafetensors>=0.3.2
60
instanttensor>=0.1.5
61
pydantic>=2.12 # 2.11 leads to error on python 3.13
62
decord==0.6.0
requirements/test/rocm.txt
+4
-2
@@ -240,8 +240,10 @@ fastar==0.10.0
240
# via fastapi-cloud-cli
241
fastparquet==2026.3.0
242
# via genai-perf
243
-fastsafetensors @ git+https://github.com/foundation-model-stack/fastsafetensors.git@65d80088fca7a8f567fba30415fbcc80f7d2259c
244
- # via -r requirements/test/rocm.in
243
+fastsafetensors==0.3.2
244
+ # via
245
+ # -c requirements/rocm.txt
246
+ # -r requirements/test/rocm.in
247
filelock==3.25.2
248
# via
249
# -c requirements/common.txt
setup.py
+1
-1
@@ -1168,7 +1168,7 @@ setup(
1168
"zen": ["zentorch==2.11.0.0"],
1169
"bench": ["pandas", "matplotlib", "seaborn", "datasets", "scipy", "plotly"],
1170
"tensorizer": ["tensorizer==2.10.1"],
1171
- "fastsafetensors": ["fastsafetensors >= 0.2.2"],
1171
+ "fastsafetensors": ["fastsafetensors >= 0.3.2"],
1172
"instanttensor": ["instanttensor >= 0.1.5"],
1173
"runai": ["runai-model-streamer[s3,gcs,azure] >= 0.15.7"],
1174
"audio": [