Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[MLPerf Llama2] update llama2 reference fp32 accuracy #1583

Merged
merged 3 commits into from
Jan 26, 2024
Merged

[MLPerf Llama2] update llama2 reference fp32 accuracy #1583

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
fb12049
#1558 update llama2 reference fp32 accuracy
nvzhihanj Jan 25, 2024
4240ff6
Merge branch 'master' into fix-zhihanj-llama2acc
mrmhodak Jan 26, 2024
3bcb5e4
Merge branch 'master' into fix-zhihanj-llama2acc
mrmhodak Jan 26, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 6 additions & 6 deletions language/llama2-70b/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -195,15 +195,15 @@ if [ -e ${ACCURACY_LOG_FILE} ]; then
fi
```

The ServerSUT was not tested for GPU runs. You can try setting `--device cuda:0`, but YMMV.
The ServerSUT was not tested for GPU runs.


## Accuracy Target
Running the GPU implementation in FP32 precision resulted in the following FP32 accuracy targets (normalized to a 0-100
scale from a 0.0-1.0 scale):
- Rouge1: 43.88
- Rouge2: 21.7108
- RougeL: 28.2502
- RougeLsum: 41.4821
- Rouge1: 44.4312
- Rouge2: 22.0352
- RougeL: 28.6162
- Tokens per sample: 294.45

This was run an 8xH100 node. Total runtime was ~4.5 days.
This was run on a DGX-H100 node. Total runtime was ~4.5 days.
4 changes: 2 additions & 2 deletions tools/submission/submission_checker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1124,8 +1124,8 @@
"3d-unet-99.9": ("DICE", 0.86170 * 0.999),
"gptj-99" : ("ROUGE1", 42.9865 * 0.99, "ROUGE2", 20.1235 * 0.99, "ROUGEL", 29.9881 * 0.99, "GEN_LEN", 4016878*0.9),
"gptj-99.9" : ("ROUGE1", 42.9865 * 0.999, "ROUGE2", 20.1235 * 0.999, "ROUGEL", 29.9881 * 0.999, "GEN_LEN", 4016878*0.9),
"llama2-70b-99" : ("ROUGE1", 43.88 * 0.99, "ROUGE2", 21.7108 * 0.99, "ROUGEL", 28.2502 * 0.99, "TOKENS_PER_SAMPLE", 293.3*0.9),
"llama2-70b-99.9" : ("ROUGE1", 43.88 * 0.999, "ROUGE2", 21.7108 * 0.999, "ROUGEL", 28.2502 * 0.999, "TOKENS_PER_SAMPLE", 293.3*0.9),
"llama2-70b-99" : ("ROUGE1", 44.4312 * 0.99, "ROUGE2", 22.0352 * 0.99, "ROUGEL", 28.6162 * 0.99, "TOKENS_PER_SAMPLE", 294.45*0.9),
"llama2-70b-99.9" : ("ROUGE1", 44.4312 * 0.999, "ROUGE2", 22.0352 * 0.999, "ROUGEL", 28.6162 * 0.999, "TOKENS_PER_SAMPLE", 294.45*0.9),
"stable-diffusion-xl": ("CLIP_SCORE", 31.68631873, "FID_SCORE", 23.01085758)
},
"accuracy-upper-limit": {
Expand Down
Loading