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

fix: Support PP for Mistral Small 3.1 (#14254)

Browse Source
This commit is contained in:
Kevin Li
2025-12-02 13:04:14 +08:00
committed by GitHub
parent 106df4eac5
commit c9e2090101
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/models/llava.py
+3 -1
View File
@@ -41,7 +41,7 @@ from sglang.srt.managers.schedule_batch import (
MultimodalDataItem, MultimodalDataItem,
MultimodalInputs, MultimodalInputs,
) )
from sglang.srt.model_executor.forward_batch_info import ForwardBatch from sglang.srt.model_executor.forward_batch_info import ForwardBatch, PPProxyTensors
from sglang.srt.model_loader.weight_utils import default_weight_loader from sglang.srt.model_loader.weight_utils import default_weight_loader
from sglang.srt.models.llama import LlamaForCausalLM from sglang.srt.models.llama import LlamaForCausalLM
from sglang.srt.models.mistral import MistralForCausalLM from sglang.srt.models.mistral import MistralForCausalLM
@@ -785,6 +785,7 @@ class LlavaForConditionalGeneration(LlavaBaseForCausalLM):
positions: torch.Tensor, positions: torch.Tensor,
forward_batch: ForwardBatch, forward_batch: ForwardBatch,
get_embedding: bool = False, get_embedding: bool = False,
pp_proxy_tensors: Optional[PPProxyTensors] = None,
): ):
hidden_states = general_mm_embed_routine( hidden_states = general_mm_embed_routine(
input_ids=input_ids, input_ids=input_ids,
@@ -796,6 +797,7 @@ class LlavaForConditionalGeneration(LlavaBaseForCausalLM):
}, },
placeholder_tokens=None, # using mm_item.pad_value placeholder_tokens=None, # using mm_item.pad_value
positions=positions, positions=positions,
pp_proxy_tensors=pp_proxy_tensors,
) )
return hidden_states return hidden_states