github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-10 15:38:29 -05:00
Code Issues Packages Projects Releases Wiki Activity

add winograd hlb_cifar10 back to tinybox benchmark (#3300)

Browse Source 
* add winograd hlb_cifar10 back to tinybox benchmark

* LATEWINO

* use wino for the full run to save benchmark time
This commit is contained in:
chenyu
2024-02-02 04:29:56 -05:00
committed by GitHub
parent aebaab011f
commit 3a7c1eb383
1 changed files with 6 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
.github/workflows/benchmark.yml vendored
View File

@@ -138,17 +138,14 @@ jobs:
HIP=1 JIT=0 python3 examples/gpt2.py --prompt "Hello." --count 10 --temperature 0 --timing | tee gpt2_unjitted.txt
HIP=1 JIT=1 python3 examples/gpt2.py --prompt "Hello." --count 10 --temperature 0 --timing | tee gpt2_jitted.txt
- name: Run 10 CIFAR training steps
run: STEPS=10 python3 examples/hlb_cifar10.py | tee train_cifar.txt
run: HIP=1 STEPS=10 python3 examples/hlb_cifar10.py | tee train_cifar.txt
- name: Run 10 CIFAR training steps w HALF
run: HIP=1 STEPS=10 HALF=1 python3 examples/hlb_cifar10.py | tee train_cifar_half.txt
# # TODO: enable this. it took 3 minutes in CI and made the full training one more than 5 minutes
# - name: Run 10 CIFAR training steps w HALF and 6 GPUS
# - name: Run 10 CIFAR training steps w 6 GPUS
# run: time HALF=1 STEPS=10 BS=1536 GPUS=6 python3 examples/hlb_cifar10.py
- name: Run full CIFAR training w HALF
run: time HALF=1 STEPS=1000 python3 examples/hlb_cifar10.py | tee train_cifar_half.txt
# # TODO: make wino faster so we can enable both
# - name: Run 10 CIFAR training steps w winograd
# run: WINO=1 STEPS=10 python3 examples/hlb_cifar10.py | tee train_cifar_wino.txt
# - name: Run 10 CIFAR training steps w WINO/HALF/HIP
# run: HALF=1 WINO=1 STEPS=10 python3 examples/hlb_cifar10.py | tee train_cifar_wino_half_hip.txt
- name: Run full CIFAR training
run: time HIP=1 HALF=1 LATEWINO=1 STEPS=1000 python3 examples/hlb_cifar10.py | tee train_cifar_one_gpu.py
- uses: actions/upload-artifact@v4
with:
name: Speed (AMD)