chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Auto Sync] Rename is_hybrid to is_hybrid_swa (#14252)

Browse Source 
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Hanming Lu <69857889+hanming-lu@users.noreply.github.com>
Co-authored-by: Hanming Lu <hanming@x.ai>
This commit is contained in:
Lianmin Zheng
2025-12-01 23:24:24 -08:00
committed by GitHub
co-authored by github-actions[bot] <github-actions[bot]@users.noreply.github.com> Hanming Lu Hanming Lu
parent 63b9300f00
commit 64092c8b55
15 changed files with 79 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/model_executor/model_runner.py
+9 -10
View File
@@ -315,8 +315,7 @@ class ModelRunner:
self.page_size = server_args.page_size
self.req_to_token_pool = req_to_token_pool
self.token_to_kv_pool_allocator = token_to_kv_pool_allocator
self.is_hybrid = model_config.is_hybrid
self.is_hybrid_swa = self.is_hybrid
self.is_hybrid_swa = model_config.is_hybrid_swa
self.use_mla_backend = self.model_config.attention_arch == AttentionArch.MLA
self.attention_chunk_size = model_config.attention_chunk_size
self.forward_pass_id = 0
@@ -444,7 +443,7 @@ class ModelRunner:
):
architectures = self.model_config.hf_config.architectures
if architectures and not any("Llama4" in arch for arch in architectures):
self.is_hybrid = self.model_config.is_hybrid = True
self.is_hybrid_swa = self.model_config.is_hybrid_swa = True
if config := self.mamba2_config:
class_name = config.__class__.__name__
@@ -1530,8 +1529,8 @@ class ModelRunner:
in self.model_config.hf_config.architectures
):
temp_ratio = (
(1 - self.is_hybrid)
+ self.is_hybrid
(1 - self.is_hybrid_swa)
+ self.is_hybrid_swa
* self.attention_chunk_size
/ self.model_config.context_len
)
@@ -1567,7 +1566,7 @@ class ModelRunner:
try:
layers = self.model.language_model.layers
except:
self.is_hybrid = False
self.is_hybrid_swa = False
return
for layer in layers:
@@ -1743,7 +1742,7 @@ class ModelRunner:
self.max_total_num_tokens = tensor.item()
# create token size for hybrid cache
if self.is_hybrid:
if self.is_hybrid_swa:
self.set_num_token_hybrid()
if self.max_total_num_tokens <= 0:
@@ -1900,7 +1899,7 @@ class ModelRunner:
end_layer=self.end_layer,
)
else:
if self.is_hybrid:
if self.is_hybrid_swa:
self.token_to_kv_pool = SWAKVPool(
size=self.full_max_total_num_tokens,
size_swa=self.swa_max_total_num_tokens,
@@ -1997,7 +1996,7 @@ class ModelRunner:
)
else:
if self.page_size == 1:
if self.is_hybrid:
if self.is_hybrid_swa:
self.token_to_kv_pool_allocator = SWATokenToKVPoolAllocator(
self.full_max_total_num_tokens,
self.swa_max_total_num_tokens,
@@ -2015,7 +2014,7 @@ class ModelRunner:
need_sort=need_sort,
)
else:
assert not self.is_hybrid
assert not self.is_hybrid_swa
self.token_to_kv_pool_allocator = PagedTokenToKVPoolAllocator(
self.max_total_num_tokens,
page_size=self.page_size,