Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
OpenDAS
vllm_cscc
Commits
b8b0859b
Unverified
Commit
b8b0859b
authored
May 02, 2025
by
Yang Wang
Committed by
GitHub
May 02, 2025
Browse files
add more pytorch related tests for torch nightly (#17422)
Signed-off-by:
Yang Wang
<
elainewy@meta.com
>
parent
d7543862
Changes
3
Hide whitespace changes
Inline
Side-by-side
Showing
3 changed files
with
14 additions
and
4 deletions
+14
-4
.buildkite/test-pipeline.yaml
.buildkite/test-pipeline.yaml
+4
-0
requirements/nightly_torch_test.txt
requirements/nightly_torch_test.txt
+7
-1
vllm/sampling_params.py
vllm/sampling_params.py
+3
-3
No files found.
.buildkite/test-pipeline.yaml
View file @
b8b0859b
...
@@ -293,6 +293,7 @@ steps:
...
@@ -293,6 +293,7 @@ steps:
parallelism
:
4
parallelism
:
4
-
label
:
PyTorch Compilation Unit Tests
-
label
:
PyTorch Compilation Unit Tests
torch_nightly
:
true
source_file_dependencies
:
source_file_dependencies
:
-
vllm/
-
vllm/
-
tests/compile
-
tests/compile
...
@@ -302,6 +303,7 @@ steps:
...
@@ -302,6 +303,7 @@ steps:
-
pytest -v -s compile/test_sequence_parallelism.py
-
pytest -v -s compile/test_sequence_parallelism.py
-
label
:
PyTorch Fullgraph Smoke Test
# 9min
-
label
:
PyTorch Fullgraph Smoke Test
# 9min
torch_nightly
:
true
source_file_dependencies
:
source_file_dependencies
:
-
vllm/
-
vllm/
-
tests/compile
-
tests/compile
...
@@ -312,6 +314,7 @@ steps:
...
@@ -312,6 +314,7 @@ steps:
-
pytest -v -s compile/piecewise/test_toy_llama.py
-
pytest -v -s compile/piecewise/test_toy_llama.py
-
label
:
PyTorch Fullgraph Test
# 18min
-
label
:
PyTorch Fullgraph Test
# 18min
torch_nightly
:
true
source_file_dependencies
:
source_file_dependencies
:
-
vllm/
-
vllm/
-
tests/compile
-
tests/compile
...
@@ -436,6 +439,7 @@ steps:
...
@@ -436,6 +439,7 @@ steps:
##### models test #####
##### models test #####
-
label
:
Basic Models Test
# 24min
-
label
:
Basic Models Test
# 24min
torch_nightly
:
true
source_file_dependencies
:
source_file_dependencies
:
-
vllm/
-
vllm/
-
tests/models
-
tests/models
...
...
requirements/nightly_torch_test.txt
View file @
b8b0859b
...
@@ -23,5 +23,11 @@ runai-model-streamer-s3==0.11.0
...
@@ -23,5 +23,11 @@ runai-model-streamer-s3==0.11.0
tensorizer>=2.9.0
tensorizer>=2.9.0
lm-eval==0.4.8
lm-eval==0.4.8
buildkite-test-collector==0.1.9
buildkite-test-collector==0.1.9
lm-eval[api]==0.4.8 # required for model evaluation test
lm-eval[api]==0.4.8 # required for model evaluation test
# required for quantization test
bitsandbytes>=0.45.3
# required for minicpmo_26 test
vector_quantize_pytorch
vocos
vllm/sampling_params.py
View file @
b8b0859b
...
@@ -186,9 +186,9 @@ class SamplingParams(
...
@@ -186,9 +186,9 @@ class SamplingParams(
logits_processors: list of functions that modify logits based on
logits_processors: list of functions that modify logits based on
previously generated tokens, and optionally prompt tokens as
previously generated tokens, and optionally prompt tokens as
a first argument.
a first argument.
truncate_prompt_tokens: If set to -1, will use the truncation size
truncate_prompt_tokens: If set to -1, will use the truncation size
supported by the model. If set to an integer k, will use only
supported by the model. If set to an integer k, will use only
the last k tokens from the prompt (i.e., left truncation).
the last k tokens from the prompt (i.e., left truncation).
Defaults to None (i.e., no truncation).
Defaults to None (i.e., no truncation).
guided_decoding: If provided, the engine will construct a guided
guided_decoding: If provided, the engine will construct a guided
decoding logits processor from these parameters. Defaults to None.
decoding logits processor from these parameters. Defaults to None.
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment