mirror of
https://github.com/radixark/miles.git
synced 2026-11-07 08:54:19 +08:00
XiaoxinWangandwangxiaoxin-sherie
d1d156b177
Co-authored-by: wangxiaoxin-sherie <wangxiaoxin7@huawei.com>
59 lines
2.5 KiB
Diff
59 lines
2.5 KiB
Diff
diff --git a/python/sglang/srt/configs/model_config.py b/python/sglang/srt/configs/model_config.py
|
|
index 691f06411..0f5964410 100644
|
|
--- a/python/sglang/srt/configs/model_config.py
|
|
+++ b/python/sglang/srt/configs/model_config.py
|
|
@@ -297,6 +297,11 @@ class ModelConfig:
|
|
"GlmMoeDsaForCausalLM",
|
|
]:
|
|
self.hf_config.architectures[0] = "DeepseekV3ForCausalLMNextN"
|
|
+ if (
|
|
+ is_draft_model
|
|
+ and self.hf_config.architectures[0] == "DeepseekV32ForCausalLM"
|
|
+ ):
|
|
+ self.hf_config.architectures[0] = "DeepseekV3ForCausalLMNextN"
|
|
|
|
if is_draft_model and self.hf_config.architectures[0] in [
|
|
"Glm4MoeForCausalLM",
|
|
diff --git a/python/sglang/srt/entrypoints/engine.py b/python/sglang/srt/entrypoints/engine.py
|
|
index fcc4af301..be9de5460 100644
|
|
--- a/python/sglang/srt/entrypoints/engine.py
|
|
+++ b/python/sglang/srt/entrypoints/engine.py
|
|
@@ -68,6 +68,7 @@ from sglang.srt.managers.io_struct import (
|
|
LoadLoRAAdapterFromTensorsReqInput,
|
|
LoadLoRAAdapterReqInput,
|
|
MultimodalDataInputFormat,
|
|
+ PostProcessWeightsReqInput,
|
|
OpenSessionReqInput,
|
|
PostProcessWeightsReqInput,
|
|
ReleaseMemoryOccupationReqInput,
|
|
diff --git a/python/sglang/srt/entrypoints/http_server.py b/python/sglang/srt/entrypoints/http_server.py
|
|
index 288f2a24c..89959ff30 100644
|
|
--- a/python/sglang/srt/entrypoints/http_server.py
|
|
+++ b/python/sglang/srt/entrypoints/http_server.py
|
|
@@ -124,6 +124,7 @@ from sglang.srt.managers.io_struct import (
|
|
InitWeightsUpdateGroupReqInput,
|
|
LoadLoRAAdapterFromTensorsReqInput,
|
|
LoadLoRAAdapterReqInput,
|
|
+ PostProcessWeightsReqInput,
|
|
OpenSessionReqInput,
|
|
ParseFunctionCallReq,
|
|
PauseGenerationReqInput,
|
|
diff --git a/python/sglang/srt/models/transformers.py b/python/sglang/srt/models/transformers.py
|
|
index d928870b9..06436aa4a 100644
|
|
--- a/python/sglang/srt/models/transformers.py
|
|
+++ b/python/sglang/srt/models/transformers.py
|
|
@@ -266,10 +266,10 @@ def replace_rms_norm_class(rms_norm: nn.Module, hidden_size: int) -> nn.Module:
|
|
)
|
|
else:
|
|
kwargs["has_weight"] = getattr(rms_norm, "with_scale", True)
|
|
- if weight_meta is not None:
|
|
- kwargs["weight_dtype"] = weight_meta.dtype
|
|
- else:
|
|
- kwargs["has_weight"] = False
|
|
+ # if weight_meta is not None:
|
|
+ # kwargs["weight_dtype"] = weight_meta.dtype
|
|
+ # else:
|
|
+ # kwargs["has_weight"] = False
|
|
base_cls = RMSNorm
|
|
norm = base_cls(**kwargs)
|