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 16, 15:45 UTC
[doc] improve installation doc (#8550)
Co-authored-by: Andy Dai <76841985+Imss27@users.noreply.github.com>
youkaichao committed
Sep 17, 2024 at 16:24 UTC
fa0c114fad4e2b807503e78d5110558cfee92ba4
2 files changed
+6
-1
docs/source/getting_started/installation.rst
+2
@@ -95,6 +95,8 @@ You can also build and install vLLM from source:
95
$ export MAX_JOBS=6
96
$ pip install -e .
97
98
+ This is especially useful when you are building on less powerful machines. For example, when you use WSL, it only `gives you half of the memory by default <https://learn.microsoft.com/en-us/windows/wsl/wsl-config>`_, and you'd better use ``export MAX_JOBS=1`` to avoid compiling multiple files simultaneously and running out of memory. The side effect is that the build process will be much slower. If you only touch the Python code, slow compilation is okay, as you are building in an editable mode: you can just change the code and run the Python script without any re-compilation or re-installation.
99
+
100
.. tip::
101
If you have trouble building vLLM, we recommend using the NVIDIA PyTorch Docker image.
102
tests/compile/test_full_graph.py
+4
-1
@@ -28,7 +28,10 @@ def test_full_graph(model, tp_size):
28
"The future of AI is",
29
]
30
sampling_params = SamplingParams(temperature=0)
31
- llm = LLM(model=model, enforce_eager=True, tensor_parallel_size=tp_size)
31
+ llm = LLM(model=model,
32
+ enforce_eager=True,
33
+ tensor_parallel_size=tp_size,
34
+ disable_custom_all_reduce=True)
35
36
outputs = llm.generate(prompts, sampling_params)
37