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

Remove unused row_idx in token_dispatcher (#3442)

Browse Source 
### What this PR does / why we need it?
The `row_idx` parameter is no longer used since
PR[#2689](https://github.com/vllm-project/vllm-ascend/pull/2689), so
remove it across multiple files to remove unnecessary calculations and
parameter passing.

### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
accuracy test passed for Qwen3 235B and DeepSeek V3 671B after this PR.


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

---------

Signed-off-by: CaranLic <740821011@qq.com>
This commit is contained in:
CaranLic
2025-10-15 09:08:31 +08:00
committed by GitHub
parent 3642b64afc
commit 15b2e5c995
11 changed files with 37 additions and 88 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
tests/ut/quantization/test_w8a8.py
View File

@@ -777,12 +777,12 @@ class TestSelectExperts(TestBase):
-1).permute(1,
0).contiguous())
weights, ids, _ = select_experts(hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,
use_grouped_topk=False,
renormalize=False,
scoring_func="softmax")
weights, ids = select_experts(hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,
use_grouped_topk=False,
renormalize=False,
scoring_func="softmax")
self.assertEqual(weights.shape, (self.num_tokens, self.top_k))
self.assertEqual(ids.shape, (self.num_tokens, self.top_k))
@@ -790,12 +790,12 @@ class TestSelectExperts(TestBase):
def test_sigmoid_scoring(self):
"""Test sigmoid scoring function"""
weights, ids, _ = select_experts(hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,
use_grouped_topk=False,
renormalize=False,
scoring_func="sigmoid")
weights, ids = select_experts(hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,
use_grouped_topk=False,
renormalize=False,
scoring_func="sigmoid")
self.assertEqual(weights.shape, (self.num_tokens, self.top_k))
self.assertEqual(ids.shape, (self.num_tokens, self.top_k))
@@ -818,13 +818,13 @@ class TestSelectExperts(TestBase):
self.top_k,
dtype=torch.long))
weights, ids, _ = select_experts(hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,
use_grouped_topk=True,
renormalize=False,
topk_group=4,
num_expert_group=2)
weights, ids = select_experts(hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,
use_grouped_topk=True,
renormalize=False,
topk_group=4,
num_expert_group=2)
mock_topk.assert_called()
self.assertEqual(weights.shape, (self.num_tokens, self.top_k))
@@ -838,7 +838,7 @@ class TestSelectExperts(TestBase):
self.num_experts)
e_score_correction_bias = torch.randn(self.num_experts)
weights, ids, _ = select_experts(
weights, ids = select_experts(
hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,
@@ -861,7 +861,7 @@ class TestSelectExperts(TestBase):
self.top_k,
dtype=torch.int32))
weights, ids, _ = select_experts(
weights, ids = select_experts(
hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,
@@ -888,7 +888,7 @@ class TestSelectExperts(TestBase):
-1).permute(1,
0).contiguous())
weights, ids, _ = select_experts(
weights, ids = select_experts(
hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,
@@ -914,7 +914,7 @@ class TestSelectExperts(TestBase):
-1).permute(1,
0).contiguous())
weights, ids, _ = select_experts(
weights, ids = select_experts(
hidden_states=self.hidden_states,
router_logits=self.router_logits,
top_k=self.top_k,