youngkingdom/vllm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[torch.compile] PyTorch 2.6 and nightly compatibility (#12393)

Browse Source 
Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao
2025-02-07 01:09:07 +08:00
committed by GitHub
parent 85ac82d228
commit 09b95e36ab
8 changed files with 489 additions and 316 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/compile/piecewise/test_simple.py
View File

@ -92,7 +92,7 @@ def test_simple_piecewise_compile():
num_graphs_seen=1, # one graph for the model
num_piecewise_graphs_seen=5, # 2 * num_layers + 1
num_piecewise_capturable_graphs_seen=3, # 1 + num_layers
num_inductor_compilations=3, # num_piecewise_capturable_graphs_seen
num_backend_compilations=3, # num_piecewise_capturable_graphs_seen
num_cudagraph_caputured=
6, # num_cudagraph_sizes * num_piecewise_capturable_graphs_seen
):

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

@ -322,7 +322,7 @@ def test_toy_llama():
num_graphs_seen=0,
num_piecewise_graphs_seen=0,
num_piecewise_capturable_graphs_seen=0,
num_inductor_compilations=0,
num_backend_compilations=0,
num_cudagraph_caputured=0,
):
outputs.append(run_model(llama_config, use_compile=False))
@ -332,7 +332,7 @@ def test_toy_llama():
num_graphs_seen=1, # one graph for the model
num_piecewise_graphs_seen=1,
num_piecewise_capturable_graphs_seen=1,
num_inductor_compilations=1, # num_piecewise_capturable_graphs_seen
num_backend_compilations=1, # num_piecewise_capturable_graphs_seen
num_cudagraph_caputured=
2, # num_cudagraph_sizes * num_piecewise_capturable_graphs_seen
):
@ -345,7 +345,7 @@ def test_toy_llama():
1, # 2 * num_layers + 1
num_piecewise_capturable_graphs_seen=1 +
llama_config.num_layers, # 1 + num_layers
num_inductor_compilations=1 +
num_backend_compilations=1 +
llama_config.num_layers, # num_piecewise_capturable_graphs_seen
num_cudagraph_caputured=2 *
(1 + llama_config.num_layers