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

[Tiny]Remove progress bar for fp8 ue8m0 quant when unneeded (#16177)

Browse Source
This commit is contained in:
Baizhou Zhang
2026-01-03 10:54:12 +08:00
committed by GitHub
parent 8b869e326c
commit bb23a8fe77
1 changed files with 11 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
python/sglang/srt/models/deepseek_v2.py
View File

@@ -3997,16 +3997,17 @@ class DeepseekV2ForCausalLM(nn.Module):
f"model.layers.{nextn_layer_id}.mlp.{expert_sub_name}.{stem}"
)
for partial_name in tqdm.tqdm(
partial_names,
desc="quant weights to fp8 ue8m0",
):
original_weight = weights_dict[f"{partial_name}.weight"]
out_w, out_s = quant_weight_ue8m0(
original_weight, weight_block_size=weight_block_size
)
weights_dict[f"{partial_name}.weight"] = out_w
weights_dict[f"{partial_name}.weight_scale_inv"] = out_s
if len(partial_names) > 0:
for partial_name in tqdm.tqdm(
partial_names,
desc="quant weights to fp8 ue8m0",
):
original_weight = weights_dict[f"{partial_name}.weight"]
out_w, out_s = quant_weight_ue8m0(
original_weight, weight_block_size=weight_block_size
)
weights_dict[f"{partial_name}.weight"] = out_w
weights_dict[f"{partial_name}.weight_scale_inv"] = out_s
if is_nextn and enable_nextn_moe_bf16_cast_to_fp8(self.quant_config):
self._mark_nextn_moe_weights_as_ue8m0()