[Model] add GLM-5.3-Flash support (#53906)
This commit is contained in:
+12
-1
@@ -333,8 +333,19 @@ class EngineCore:
|
||||
vllm_config.cache_config.num_gpu_blocks = scheduler_kv_cache_config.num_blocks
|
||||
kv_cache_groups = scheduler_kv_cache_config.kv_cache_groups
|
||||
if kv_cache_groups:
|
||||
# Exclude groups that opt out of prefix caching (e.g. GLM-5.3-Flash
|
||||
# kpool tail, a 1-block/req scratch buffer with block_size=kpool):
|
||||
# their small block_size would otherwise drag the global block_size
|
||||
# below the real allocator block size and desync it from mamba.
|
||||
participating = [
|
||||
g.kv_cache_spec.block_size
|
||||
for g in kv_cache_groups
|
||||
if g.kv_cache_spec.prefix_cacheable
|
||||
]
|
||||
vllm_config.cache_config.block_size = min(
|
||||
g.kv_cache_spec.block_size for g in kv_cache_groups
|
||||
participating
|
||||
if participating
|
||||
else [g.kv_cache_spec.block_size for g in kv_cache_groups]
|
||||
)
|
||||
update_kv_cache_capacity(vllm_config, scheduler_kv_cache_config)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user