A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 09:45 UTC
[Doc] Fix invalid JSON in example args (#18527)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
May 22, 2025 at 15:11 UTC
23b67b37b246fb7cecb3815c6873fa3d18c3c0e7
1 file changed
+7
-3
docs/source/design/v1/torch_compile.md
+7
-3
@@ -99,7 +99,9 @@ This time, Inductor compilation is completely bypassed, and we will load from di
99
100
The above example just uses Inductor to compile for a general shape (i.e. symbolic shape). We can also use Inductor to compile for some of the specific shapes, for example:
101
102
-`vllm serve meta-llama/Llama-3.2-1B --compilation_config "{'compile_sizes': [1, 2, 4, 8]}"`
102
+```
103
+vllm serve meta-llama/Llama-3.2-1B --compilation_config '{"compile_sizes": [1, 2, 4, 8]}'
104
+```
105
106
Then it will also compile a specific kernel just for batch size `1, 2, 4, 8`. At this time, all of the shapes in the computation graph are static and known, and we will turn on auto-tuning to tune for max performance. This can be slow when you run it for the first time, but the next time you run it, we can directly bypass the tuning and run the tuned kernel.
107
@@ -134,12 +136,14 @@ The cudagraphs are captured and managed by the compiler backend, and replayed wh
136
137
By default, vLLM will try to determine a set of sizes to capture cudagraph. You can also override it using the config `cudagraph_capture_sizes`:
138
137
-`vllm serve meta-llama/Llama-3.2-1B --compilation-config "{'cudagraph_capture_sizes': [1, 2, 4, 8]}"`
139
+```
140
+vllm serve meta-llama/Llama-3.2-1B --compilation-config '{"cudagraph_capture_sizes": [1, 2, 4, 8]}'
141
+```
142
143
Then it will only capture cudagraph for the specified sizes. It can be useful to have fine-grained control over the cudagraph capture.
144
145
### Full Cudagraph capture
146
143
-It is possible to include attention as part of the cudagraph if using an attention backend that is cudagraph compatible. This can improve performance in some cases such as decode speed for smaller models. Enable this using `--compilation-config "{'full_cuda_graph': True}"`
147
+It is possible to include attention as part of the cudagraph if using an attention backend that is cudagraph compatible. This can improve performance in some cases such as decode speed for smaller models. Enable this using `--compilation-config '{"full_cuda_graph": true}'`.
148
149
Currently only FlashAttention 3 is compatible, and only when cascade attention is disabled.