Mirrors
/
hf_text-generation-inference
mirror of https://github.com/huggingface/text-generation-inference.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Damn inflated sparse tflops.

This commit is contained in:
Nicolas Patry 2024-11-10 15:32:34 +01:00
parent 748dce60cd
commit 3a53e8c288
No known key found for this signature in database
GPG Key ID: D2920555C90F704C
1 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
launcher/src/main.rs
View File

@ -33,7 +33,7 @@ fn compute_optimal(config: Option<&Config>, compute: Option<&ComputeType>) -> Op
if let (Some(config), Some(compute)) = (config, compute) {
if let (Some(f16_max_compute), Some(model_compute)) = (compute.f16_flop(), config.flop()) {
tracing::debug!("MAx compute {f16_max_compute} model compute {model_compute}");
let optimal_size = (f16_max_compute / model_compute / 2) as usize;
let optimal_size = (f16_max_compute / model_compute) as usize;
if optimal_size > 100 {
// Ignore calculations that's too low
// Most likely an error
@ -1508,7 +1508,8 @@ impl ComputeType {
// https://www.nvidia.com/en-us/data-center/products/a10-gpu/
"nvidia-a10g" => Some(125 * 10u64.pow(12)),
// https://www.nvidia.com/en-us/data-center/h100/
"nvidia-h100-80gb-hb3" => Some(1900 * 10u64.pow(12)),
// https://www.techpowerup.com/gpu-specs/docs/nvidia-gh100-architecture.pdf
"nvidia-h100-80gb-hb3" => Some(900 * 10u64.pow(12)),
// https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-us-nvidia-1758950-r4-web.pdf
"nvidia-a100" => Some(312 * 10u64.pow(12)),
card => {