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 17, 05:30 UTC
[CI] Ensure documentation build is checked in CI (#2842)
Simon Mo committed
Feb 12, 2024 at 22:53 UTC
f964493274c3c839b2e27453cb70f179090cd027
5 files changed
+14
-1
.buildkite/test-pipeline.yaml
+7
@@ -49,3 +49,10 @@ steps:
49
commands:
50
- pip install aiohttp
51
- bash run-benchmarks.sh
52
+
53
+- label: Documentation Build
54
+ working_dir: "/vllm-workspace/docs"
55
+ no_gpu: True
56
+ commands:
57
+ - pip install -r requirements-docs.txt
58
+ - SPHINXOPTS=\"-W\" make html
.buildkite/test-template.j2
+3
-1
@@ -35,13 +35,15 @@ steps:
35
- image: "{{ docker_image }}"
36
command: ["bash"]
37
args:
38
- - "-c"
38
+ - '-c'
39
- "'cd {{ (step.working_dir or default_working_dir) | safe }} && {{ step.command or (step.commands | join(' && ')) | safe }}'"
40
+ {% if not step.no_gpu %}
41
resources:
42
requests:
43
nvidia.com/gpu: "{{ step.num_gpus or default_num_gpu }}"
44
limits:
45
nvidia.com/gpu: "{{ step.num_gpus or default_num_gpu }}"
46
+ {% endif %}
47
env:
48
- name: HF_TOKEN
49
valueFrom:
docs/source/conf.py
+2
@@ -94,3 +94,5 @@ class MockedClassDocumenter(autodoc.ClassDocumenter):
94
95
96
autodoc.ClassDocumenter = MockedClassDocumenter
97
+
98
+navigation_with_keys = False
docs/source/index.rst
+1
@@ -89,6 +89,7 @@ Documentation
89
:caption: Quantization
90
91
quantization/auto_awq
92
+ quantization/fp8_e5m2_kv_cache
93
94
.. toctree::
95
:maxdepth: 2
docs/source/quantization/fp8_e5m2_kv_cache.rst
+1
@@ -9,6 +9,7 @@ The FP8 data format retains 2~3 mantissa bits and can convert float/fp16/bflaot1
9
Here is an example of how to enable this feature:
10
11
.. code-block:: python
12
+
13
from vllm import LLM, SamplingParams
14
# Sample prompts.
15
prompts = [