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

[NPU] bugfix: resolve modelslim load weights bug (#19472)

Browse Source
This commit is contained in:
Hexq0210
2026-02-28 16:22:45 +08:00
committed by GitHub
parent 53c767d224
commit 4ebe9e1e2f
1 changed files with 0 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/modelslim/modelslim.py
-9
View File
@@ -239,15 +239,6 @@ class ModelSlimConfig(QuantizationConfig):
): ):
# adapted from vllm.model_executor.layers.quantization.utils.quant_utils.is_layer_skipped # adapted from vllm.model_executor.layers.quantization.utils.quant_utils.is_layer_skipped
proj_name = prefix.split(".")[-1] proj_name = prefix.split(".")[-1]
if not hasattr(self, "_quant_description_normalized"):
quant_description = {}
for prefix_, value in self.quant_description.items():
prefix_ = prefix_.replace("language_model.", "")
if "visual" in prefix_:
prefix_ = prefix_.replace("model.", "")
quant_description[prefix_] = value
self.quant_description = quant_description
self._quant_description_normalized = True
if proj_name in fused_mapping: if proj_name in fused_mapping:
shard_prefixes = [ shard_prefixes = [
prefix.replace(proj_name, shard_proj_name) prefix.replace(proj_name, shard_proj_name)