EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Update GLM-4.5 Model Doc (#11017)

Browse Source
This commit is contained in:
Yuxuan Zhang
2025-09-29 02:21:27 +08:00
committed by GitHub
parent 07440f5f34
commit abb6781573
5 changed files with 11 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/srt/function_call/glm4_moe_detector.py
View File

@@ -39,7 +39,7 @@ def parse_arguments(json_value):
class Glm4MoeDetector(BaseFormatDetector):
"""
Detector for GLM-4.5 models.
Detector for GLM-4.5 and GLM-4.6 models.
Assumes function call format:
<tool_call>get_weather\n<arg_key>city</arg_key>\n<arg_value>北京</arg_value>\n<arg_key>date</arg_key>\n<arg_value>2024-06-27</arg_value>\n</tool_call>\n<tool_call>get_weather\n<arg_key>city</arg_key>\n<arg_value>上海</arg_value>\n<arg_key>date</arg_key>\n<arg_value>2024-06-27</arg_value>\n</tool_call>
"""
@@ -53,7 +53,7 @@ class Glm4MoeDetector(BaseFormatDetector):
self.func_arg_regex = r"<arg_key>(.*?)</arg_key>\s*<arg_value>(.*?)</arg_value>"
def has_tool_call(self, text: str) -> bool:
"""Check if the text contains a glm-4.5 format tool call."""
"""Check if the text contains a glm-4.5 / glm-4.6 format tool call."""
return self.bot_token in text
def detect_and_parse(self, text: str, tools: List[Tool]) -> StreamingParseResult:
@@ -102,7 +102,7 @@ class Glm4MoeDetector(BaseFormatDetector):
self, new_text: str, tools: List[Tool]
) -> StreamingParseResult:
"""
Streaming incremental parsing tool calls for GLM-4.5 format.
Streaming incremental parsing tool calls for GLM-4.5 and GLM-4.6 format.
"""
self._buffer += new_text
current_text = self._buffer

6
python/sglang/srt/models/glm4_moe.py
View File

@@ -12,7 +12,7 @@
# limitations under the License.
# ==============================================================================
"""Inference-only GLM-4.5 model compatible with HuggingFace weights"""
"""Inference-only GLM-4.5, GLM-4.6 model compatible with HuggingFace weights"""
import logging
from typing import Any, Dict, Iterable, Optional, Tuple
@@ -785,9 +785,9 @@ class Glm4MoeForCausalLM(DeepseekV2ForCausalLM):
or self.config.architectures[0] != architecture
or self.config.n_shared_experts != 1
):
disable_reason = "Only GLM-4.5 on NV-platform with capability >= 80 can use shared experts fusion optimization."
disable_reason = "Only GLM-4.5 or GLM-4.6 on NV-platform with capability >= 80 can use shared experts fusion optimization."
elif get_moe_expert_parallel_world_size() > 1:
disable_reason = "Deepseek and GLM-4.5 can not use shared experts fusion optimization under expert parallelism."
disable_reason = "Deepseek and GLM-4.5 or GLM-4.6 can not use shared experts fusion optimization under expert parallelism."
if disable_reason is not None:
global_server_args_dict["disable_shared_experts_fusion"] = True

4
python/sglang/srt/models/glm4_moe_nextn.py
View File

@@ -12,7 +12,7 @@
# limitations under the License.
# ==============================================================================
"""Inference-only GLM-4.5 NextN Speculative Decoding."""
"""Inference-only GLM-4.5, GLM-4.6 NextN Speculative Decoding."""
import logging
from typing import Iterable, Optional, Tuple
@@ -48,7 +48,7 @@ class Glm4MoeModelNextN(nn.Module):
super().__init__()
if quant_config is not None and quant_config.get_name() == "modelopt_fp4":
logger.warning(
"Overriding Glm4MoeForCausalLMNextN quant config for modelopt_fp4 GLM-4.5 model."
"Overriding Glm4MoeForCausalLMNextN quant config for modelopt_fp4 GLM-4.5 / GLM-4.6 model."
)
quant_config = None