Skip to content
This repository was archived by the owner on Sep 10, 2025. It is now read-only.
Run parallel prefill

Add SDPA backend tests and refactor generate.py #2794

Sign in to view logs

Add SDPA backend tests and refactor generate.py

Add SDPA backend tests and refactor generate.py #2794

Workflow file for this run

.github/workflows/more-tests.yml at e3933b2
name: Run parallel prefill
on:
pull_request:
push:
branches:
- main
workflow_dispatch:
jobs:
test-cuda:
permissions:
id-token: write
contents: read
uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
with:
runner: linux.g5.4xlarge.nvidia.gpu
gpu-arch-type: cuda
gpu-arch-version: "12.4"
timeout: 60
script: |
set -xeou pipefail
echo "::group::Print machine info"
uname -a
echo "::endgroup::"
echo "::group::Download checkpoints"
# Install requirements
./install/install_requirements.sh cuda
pip3 list
python3 -c 'import torch;print(f"torch: {torch.__version__, torch.version.git_version}")'
echo "::endgroup::"
echo "::group::Download checkpoints"
mkdir -p checkpoints/stories15M
pushd checkpoints/stories15M
wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories15M.pt
wget https://github.com/karpathy/llama2.c/raw/master/tokenizer.model
popd
echo "::endgroup::"
echo "::group::Run inference"
export MODEL_PATH=checkpoints/stories15M/stories15M.pt
export MODEL_NAME=stories15M
export MODEL_DIR=/tmp
for DTYPE in bfloat16 float16 float32; do
###################################################################
# group with different temperatures
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 0
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 0.9
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 1.0
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 100
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 200
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 500
###################################################################
# group with different temperatures and prefill, and compile
# and prefill compile
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 0 --compile --compile-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 0.9 --compile --compile-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 1.0 --compile --compile-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 100 --compile --compile-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 200 --compile --compile-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 500 --compile --compile-prefill
###################################################################
# group with different temperatures and sequential prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 0 --sequential-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 0.9 --sequential-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 1.0 --sequential-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 100 --sequential-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 200 --sequential-prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 500 --sequential-prefill
###################################################################
# group with different temperatures and prefill, and compile
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 0 --sequential-prefill --compile
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 0.9 --sequential-prefill --compile
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --temperature 1.0 --sequential-prefill --compile
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 100 --sequential-prefill --compile
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 200 --sequential-prefill --compile
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --device cpu --dtype ${DTYPE} --top-k 500 --sequential-prefill --compile
done
echo "tests complete"
echo "******************************************"
echo "::endgroup::"
test-sdpa-backends:
permissions:
id-token: write
contents: read
uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
with:
runner: linux.g5.4xlarge.nvidia.gpu
gpu-arch-type: cuda
gpu-arch-version: "12.4"
timeout: 60
script: |
set -xeou pipefail
echo "::group::Print machine info"
uname -a
echo "::endgroup::"
echo "::group::Download checkpoints"
# Install requirements
./install/install_requirements.sh cuda
pip3 list
python3 -c 'import torch;print(f"torch: {torch.__version__, torch.version.git_version}")'
echo "::endgroup::"
echo "::group::Download checkpoints"
mkdir -p checkpoints/stories15M
pushd checkpoints/stories15M
wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories15M.pt
wget https://github.com/karpathy/llama2.c/raw/master/tokenizer.model
popd
echo "::endgroup::"
echo "::group::Run inference"
export MODEL_PATH=checkpoints/stories15M/stories15M.pt
export MODEL_NAME=stories15M
export MODEL_DIR=/tmp
for DEVICE in cpu cuda; do
for DTYPE in bfloat16 float16 float32; do
for SDPA in 'math' 'flash_attention' 'efficient_attention' 'cudnn_attention'; do
echo "******************************************************************"
echo "******* $DEVICE $DTYPE $SDPA "
###################################################################
# Python execution interpreted vanilla
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --attention-backend ${SDPA} --device ${DEVICE} --dtype ${DTYPE} --temperature 0
###################################################################
# prefill, and compile and prefill compile
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --attention-backend ${SDPA} --device ${DEVICE} --dtype ${DTYPE} --temperature 0 --compile --compile-prefill
###################################################################
# sequential prefill
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --attention-backend ${SDPA} --device ${DEVICE} --dtype ${DTYPE} --temperature 0 --sequential-prefill
###################################################################
# prefill, and compile
python torchchat.py generate --checkpoint-path ${MODEL_PATH} --attention-backend ${SDPA} --device ${DEVICE} --dtype ${DTYPE} --temperature 0 --sequential-prefill --compile
done
done
done
echo "tests complete"
echo "******************************************"
echo "::endgroup::"