xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-15 13:55:38 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Respect num-gpu-blocks-override in v1 (#19503)

Browse Source 
Signed-off-by: Jon Swenson <jmswen@gmail.com>
This commit is contained in:
jmswen 2025-06-12 04:00:23 -07:00 committed by GitHub
parent af09b3f0a0
commit c9280e6346
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 17 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
tests/v1/core/test_kv_cache_utils.py
View File

@ -900,3 +900,19 @@ def test_get_kv_cache_config():
with pytest.raises(NotImplementedError): with pytest.raises(NotImplementedError):
get_kv_cache_config(vllm_config, kv_cache_specs_hybrid, get_kv_cache_config(vllm_config, kv_cache_specs_hybrid,
mem_per_block_per_layer * 2 * 32) mem_per_block_per_layer * 2 * 32)
# Test num_gpu_blocks_override
vllm_config.cache_config.num_gpu_blocks_override = 16
kv_cache_config_override_blocks = get_kv_cache_config(
vllm_config, kv_cache_specs_full, mem_per_block_per_layer * 2 * 32)
assert kv_cache_config_override_blocks == KVCacheConfig(
num_blocks=16,
kv_cache_tensors=[
KVCacheTensor(size=mem_per_block_per_layer * 16,
shared_by=["layer_1"]),
KVCacheTensor(size=mem_per_block_per_layer * 16,
shared_by=["layer_2"]),
],
kv_cache_groups=[
KVCacheGroupSpec(["layer_1", "layer_2"], new_kv_cache_spec())
])

1
vllm/v1/core/kv_cache_utils.py
View File

@ -660,6 +660,7 @@ def get_num_blocks(vllm_config: VllmConfig, num_layers: int,
logger.info( logger.info(
"Overriding num_gpu_blocks=%d with " "Overriding num_gpu_blocks=%d with "
"num_gpu_blocks_override=%d", num_blocks, num_gpu_blocks_override) "num_gpu_blocks_override=%d", num_blocks, num_gpu_blocks_override)
num_blocks = num_gpu_blocks_override
return num_blocks return num_blocks