xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-06 11:07:08 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix duplicate torch operation registrations in tests/compile

Browse Source 
Co-authored-by: ProExpertProg <11367180+ProExpertProg@users.noreply.github.com>
This commit is contained in:
copilot-swe-agent[bot] 2025-08-20 13:31:41 +00:00
parent e263eccfae
commit 9d6f0372e5
3 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tests/compile/piecewise/test_multiple_graphs.py
View File

@ -19,7 +19,7 @@ from vllm.forward_context import set_forward_context
from vllm.utils import direct_register_custom_op
# create a library to hold the custom op
silly_lib = Library("silly", "FRAGMENT") # noqa
silly_lib = Library("silly_multiple", "FRAGMENT") # noqa
BATCH_SIZE = 32
MLP_SIZE = 128
@ -188,7 +188,7 @@ def test_ignore_torch_compile_decorator():
def forward(self, x: torch.Tensor) -> torch.Tensor:
x = x + x
attn_output = torch.empty_like(x)
torch.ops.silly.attention(x, x, x, attn_output)
torch.ops.silly_multiple.attention(x, x, x, attn_output)
x = attn_output
x = x * 3
return x

6
tests/compile/piecewise/test_simple.py
View File

@ -20,7 +20,7 @@ from vllm.utils import direct_register_custom_op
global_counter = 0
# create a library to hold the custom op
silly_lib = Library("silly", "FRAGMENT") # noqa
silly_lib = Library("silly_simple", "FRAGMENT") # noqa
def silly_attention(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor,
@ -66,12 +66,12 @@ class SillyModel(nn.Module):
x = x + 1
x = x + 2
out = torch.empty_like(x)
torch.ops.silly.attention(x, x, x, out)
torch.ops.silly_simple.attention(x, x, x, out)
x = out
x = x - 2
x = x - 1
out = torch.empty_like(x)
torch.ops.silly.attention(x, x, x, out)
torch.ops.silly_simple.attention(x, x, x, out)
x = out
x = x + 1
return x

4
tests/compile/piecewise/test_toy_llama.py
View File

@ -24,7 +24,7 @@ from vllm.forward_context import BatchDescriptor, set_forward_context
from vllm.utils import direct_register_custom_op
# create a library to hold the custom op
silly_lib = Library("silly", "FRAGMENT") # noqa
silly_lib = Library("silly_toy_llama", "FRAGMENT") # noqa
def silly_attention(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor,
@ -160,7 +160,7 @@ class LlamaAttention(nn.Module):
k = k + positions.unsqueeze(1)
attn_output = torch.empty_like(q)
torch.ops.silly.attention(q, k, v, attn_output)
torch.ops.silly_toy_llama.attention(q, k, v, attn_output)
output = self.output_projection(attn_output)
return output