EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[BugFix] Fix torchair+mtp bug after deleting deepseek_mtp. (#3590)

Browse Source 
This is a missing bug fix introduced by PR #3561

- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

---------

Signed-off-by: whx-sjtu <2952154980@qq.com>
This commit is contained in:
whx
2025-10-21 22:23:52 +08:00
committed by GitHub
parent 0c83eee9b1
commit bd11c0054f
3 changed files with 7 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm_ascend/spec_decode/mtp_proposer.py
View File

@@ -175,7 +175,7 @@ class MtpProposer(Proposer):
torchair_compiled_model(
input_ids=input_ids,
positions=positions,
previous_hidden_states=previous_hidden_states,
hidden_states=previous_hidden_states,
inputs_embeds=None,
intermediate_tensors=None,
attn_metadata=attn_metadata,
@@ -460,7 +460,7 @@ class MtpProposer(Proposer):
hidden_states = torchair_compiled_model(
input_ids=self.input_ids[:num_input_tokens],
positions=self.positions[:num_input_tokens],
previous_hidden_states=self.
hidden_states=self.
hidden_states[:num_input_tokens],
inputs_embeds=None,
intermediate_tensors=None,