Gate SSU dispatch setup (#40039)
Signed-off-by: Roi Koren <roik@nvidia.com>
This commit is contained in:
@@ -6738,7 +6738,9 @@ class GPUModelRunner(
|
||||
self.may_add_encoder_only_layers_to_kv_cache_config()
|
||||
self.maybe_add_kv_sharing_layers_to_kv_cache_groups(kv_cache_config)
|
||||
self.initialize_attn_backend(kv_cache_config, is_profiling=is_profiling)
|
||||
initialize_mamba_ssu_backend(self.vllm_config.mamba_config)
|
||||
initialize_mamba_ssu_backend(
|
||||
self.vllm_config.mamba_config, self.kv_cache_config
|
||||
)
|
||||
# The kernel block size for all KV cache groups. For example, if
|
||||
# kv_cache_manager uses block_size 256 for a given group, but the attention
|
||||
# backends for that group only supports block_size 64, we will return
|
||||
|
||||
Reference in New Issue
Block a user