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

[BugFix] Fix mlapo accuracy problem related with weight processing. (#3857)

Browse Source 
This PR fixes a mlapo accuracy problem related with weight processing.
Furthermore, modify mlapo related e2e test with quantized deepseek model
to make it effective.

Signed-off-by: whx-sjtu <2952154980@qq.com>
This commit is contained in:
whx
2025-10-30 00:35:50 +08:00
committed by GitHub
parent d9249c968e
commit 211d4b9da4
2 changed files with 2 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py
View File

@@ -111,19 +111,3 @@ def test_mtp2_correctness_full_graph(
model_name: str,
):
mtp_correctness(sampling_config, model_name, 2, CUDAGraphMode.FULL)
@patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_MLAPO": "1"})
def test_mtp_correctness_piecewise_graph_with_mlapo_kernel(
sampling_config: SamplingParams,
model_name: str,
):
mtp_correctness(sampling_config, model_name, 1)
@patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_MLAPO": "1"})
def test_mtp_correctness_full_graph_with_mlapo_kernel(
sampling_config: SamplingParams,
model_name: str,
):
mtp_correctness(sampling_config, model_name, 1, CUDAGraphMode.FULL)

4
vllm_ascend/attention/mla_v1.py
View File

@@ -676,9 +676,9 @@ class AscendMLAImpl(MLAAttentionImpl):
..., self.q_lora_rank:].contiguous()
q_a_proj_wt = self.fused_qkv_a_proj.weight.data[
..., :self.q_lora_rank].contiguous()
kv_a_proj_wt = kv_a_proj_wt.contiguous()
kv_a_proj_wt = kv_a_proj_wt.t().contiguous()
kv_a_proj_wt = trans_rope_weight(kv_a_proj_wt, self.qk_rope_head_dim)
kv_a_proj_wt = kv_a_proj_wt.contiguous()
kv_a_proj_wt = kv_a_proj_wt.t().contiguous()
wd_qkv = torch.cat((kv_a_proj_wt, q_a_proj_wt), dim=-1)
wd_qkv = wd_qkv.t().contiguous()
wd_qkv = transdata(wd_qkv,