[Core] Remove deprecated mamba_cache_mode "all" (#58997)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Thomas Parnell <tpa@zurich.ibm.com>
This commit is contained in:
co-authored by
Claude Opus 5.5
Thomas Parnell
parent
08d77cad7b
commit
cb6458c5ce
@@ -646,20 +646,9 @@ class MambaModelConfig(VerifyAndUpdateConfig):
|
||||
cache_config.mamba_cache_mode,
|
||||
model_config.architecture,
|
||||
)
|
||||
if (
|
||||
cache_config.mamba_cache_mode == "all"
|
||||
and not model_config.supports_mamba_prefix_caching
|
||||
):
|
||||
cache_config.mamba_cache_mode = "align"
|
||||
logger.warning(
|
||||
"Hybrid or mamba-based model detected without support "
|
||||
"for prefix caching with Mamba cache 'all' mode: "
|
||||
"falling back to 'align' mode."
|
||||
)
|
||||
if cache_config.mamba_cache_mode == "align":
|
||||
assert vllm_config.scheduler_config.enable_chunked_prefill, (
|
||||
"Chunked prefill is required for mamba cache mode 'align'."
|
||||
)
|
||||
assert vllm_config.scheduler_config.enable_chunked_prefill, (
|
||||
"Chunked prefill is required for mamba cache mode 'align'."
|
||||
)
|
||||
# By default, mamba block size will be set to max_model_len (see
|
||||
# below). When enabling prefix caching, we align mamba block size
|
||||
# to the block size as the basic granularity for prefix caching.
|
||||
|
||||
@@ -418,12 +418,6 @@ class InternS2MobiusForCausalLM(Qwen3_5ForCausalLMBase):
|
||||
self.quant_config = vllm_config.quant_config
|
||||
self.scheduler_config = vllm_config.scheduler_config
|
||||
|
||||
if vllm_config.cache_config.mamba_cache_mode == "all":
|
||||
raise NotImplementedError(
|
||||
"Intern-S2-Mobius does not support 'all' prefix caching; "
|
||||
"use '--mamba-cache-mode=align' instead."
|
||||
)
|
||||
|
||||
nn.Module.__init__(self)
|
||||
self.config = config
|
||||
self.model = InternS2MobiusModel(
|
||||
|
||||
@@ -458,13 +458,6 @@ class Lfm2ForCausalLM(
|
||||
def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
|
||||
config = vllm_config.model_config.hf_config
|
||||
quant_config = vllm_config.quant_config
|
||||
cache_config = vllm_config.cache_config
|
||||
if cache_config.mamba_cache_mode == "all":
|
||||
raise NotImplementedError(
|
||||
"Lfm2 currently does not support 'all' prefix caching, "
|
||||
"please use '--mamba-cache-mode=align' instead"
|
||||
)
|
||||
|
||||
super().__init__()
|
||||
self.config = config
|
||||
self.model = Lfm2Model(
|
||||
|
||||
@@ -573,13 +573,6 @@ class Lfm2MoeForCausalLM(
|
||||
def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
|
||||
config = vllm_config.model_config.hf_config
|
||||
quant_config = vllm_config.quant_config
|
||||
cache_config = vllm_config.cache_config
|
||||
|
||||
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
|
||||
|
||||
@@ -329,14 +329,8 @@ class Qwen3_5ForCausalLMBase(
|
||||
config = vllm_config.model_config.hf_text_config
|
||||
self.vllm_config = vllm_config
|
||||
self.model_config = vllm_config.model_config
|
||||
cache_config = vllm_config.cache_config
|
||||
|
||||
scheduler_config = vllm_config.scheduler_config
|
||||
if cache_config.mamba_cache_mode == "all":
|
||||
raise NotImplementedError(
|
||||
"Qwen3.5 currently does not support 'all' prefix caching, "
|
||||
"please use '--mamba-cache-mode=align' instead"
|
||||
)
|
||||
self.quant_config = vllm_config.quant_config
|
||||
|
||||
super().__init__()
|
||||
|
||||
@@ -227,13 +227,6 @@ class Qwen3_5MTP(LocalArgmaxMixin, nn.Module, SupportsMultiModal, SupportsPP):
|
||||
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
|
||||
config = vllm_config.model_config.hf_text_config
|
||||
self.vllm_config = vllm_config
|
||||
cache_config = vllm_config.cache_config
|
||||
if cache_config.mamba_cache_mode == "all":
|
||||
raise NotImplementedError(
|
||||
"Qwen3_5MTP currently does not support 'all' prefix caching, "
|
||||
"please use '--mamba-cache-mode=align' instead"
|
||||
)
|
||||
|
||||
self.quant_config = vllm_config.quant_config
|
||||
|
||||
super().__init__()
|
||||
|
||||
@@ -828,14 +828,8 @@ class Qwen3NextForCausalLM(
|
||||
config = vllm_config.model_config.hf_text_config
|
||||
self.vllm_config = vllm_config
|
||||
self.model_config = vllm_config.model_config
|
||||
cache_config = vllm_config.cache_config
|
||||
|
||||
scheduler_config = vllm_config.scheduler_config
|
||||
if cache_config.mamba_cache_mode == "all":
|
||||
raise NotImplementedError(
|
||||
"Qwen3Next currently does not support 'all' prefix caching, "
|
||||
"please use '--mamba-cache-mode=align' instead"
|
||||
)
|
||||
self.quant_config = vllm_config.quant_config
|
||||
|
||||
super().__init__()
|
||||
|
||||
@@ -187,13 +187,6 @@ class Qwen3NextMTP(nn.Module, QwenNextMixtureOfExperts):
|
||||
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
|
||||
config = vllm_config.model_config.hf_config
|
||||
self.vllm_config = vllm_config
|
||||
cache_config = vllm_config.cache_config
|
||||
if cache_config.mamba_cache_mode == "all":
|
||||
raise NotImplementedError(
|
||||
"Qwen3NextMTP currently does not support 'all' prefix caching, "
|
||||
"please use '--mamba-cache-mode=align' instead"
|
||||
)
|
||||
|
||||
self.quant_config = vllm_config.quant_config
|
||||
|
||||
super().__init__()
|
||||
|
||||
Reference in New Issue
Block a user