fix: allow LFM2 MoE prefix caching (align) (#33376)
Signed-off-by: Tianshu Yu <tianshuyu.formal@gmail.com>
This commit is contained in:
@@ -651,9 +651,11 @@ class Lfm2MoeForCausalLM(
|
||||
quant_config = vllm_config.quant_config
|
||||
cache_config = vllm_config.cache_config
|
||||
|
||||
assert not cache_config.enable_prefix_caching, (
|
||||
"Lfm2Moe currently does not support prefix caching"
|
||||
)
|
||||
if cache_config.mamba_cache_mode == "all":
|
||||
raise NotImplementedError(
|
||||
"Lfm2Moe currently does not support 'all' prefix caching, "
|
||||
"please use '--mamba-cache-mode=align' instead"
|
||||
)
|
||||
|
||||
super().__init__()
|
||||
self.config = config
|
||||
|
||||
Reference in New Issue
Block a user