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

[CI] Fix Nemotron nightly mixed precision checkpoints test (#27284)

Browse Source 
Co-authored-by: Brayden Zhong <brayden.zhong@radixark.ai>
This commit is contained in:
Brayden Zhong
2026-06-05 13:51:26 -07:00
committed by GitHub
co-authored by Brayden Zhong
parent 3b62286fca
commit d381ec7997
1 changed files with 8 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/model_executor/model_runner.py
+9 -7
View File
@@ -1419,14 +1419,16 @@ class ModelRunner(ModelRunnerKVCacheMixin):
) )
# TODO: Make sure all models have `quant_config` attribute, and all online quantization methods register which layers they actually quantize. # TODO: Make sure all models have `quant_config` attribute, and all online quantization methods register which layers they actually quantize.
if ( # TODO: Move this online-quantization reporting out of ModelRunner.
hasattr(self.model, "quant_config") quantized_layers = getattr(
and hasattr(self.model.quant_config, "quantized_layers") getattr(self.model, "quant_config", None), "quantized_layers", None
and self.server_args.quantization is not None
):
layer_types, quantized_layers_count = (
self.model.quant_config.quantized_layers
) )
if (
self.server_args.quantization is not None
and isinstance(quantized_layers, tuple)
and len(quantized_layers) == 2
):
layer_types, quantized_layers_count = quantized_layers
logger.info( logger.info(
f"Online {self.server_args.quantization} quantization: quantized {quantized_layers_count} layers of types: {layer_types}" f"Online {self.server_args.quantization} quantization: quantized {quantized_layers_count} layers of types: {layer_types}"
) )