|
Download docs/cards/decode_budget.svg from PYTHAI/bankml: direct link, hf CLI and curl.
- Browser
- Download file 2.29 kB
-
https://huggingface.co/spaces/PYTHAI/bankml/resolve/0ab4e0b6dd5a713718f7d8e1409d62164949e896/docs/cards/decode_budget.svg
- Command line
-
hf download hf://spaces/PYTHAI/bankml@0ab4e0b6dd5a713718f7d8e1409d62164949e896/docs/cards/decode_budget.svg
-
curl -L -o decode_budget.svg https://huggingface.co/spaces/PYTHAI/bankml/resolve/0ab4e0b6dd5a713718f7d8e1409d62164949e896/docs/cards/decode_budget.svg
2.29 kB
| <svg xmlns='http://www.w3.org/2000/svg' width='720' height='182' viewBox='0 0 720 182' role='img' aria-label='One ternary token: all 253 matmuls, 3 threads'><rect x='0.5' y='0.5' width='719' height='181' rx='10' fill='#0d1117' stroke='#30363d'/><text x='20' y='32' font-family='ui-monospace,SFMono-Regular,Menlo,Consolas,monospace' font-size='15' font-weight='700' text-anchor='start' fill='#e6edf3'>One ternary token: all 253 matmuls, 3 threads</text><text x='20' y='70' font-family='ui-monospace,SFMono-Regular,Menlo,Consolas,monospace' font-size='12' font-weight='400' text-anchor='start' fill='#8b949e'>253 matmuls, 1 token</text><rect x='200' y='68' width='29.1' height='9' rx='2' fill='#D9A23A'/><text x='237.1' y='76' font-family='ui-monospace,SFMono-Regular,Menlo,Consolas,monospace' font-size='10.5' font-weight='400' text-anchor='start' fill='#e6edf3'>0.222 s 路 bankml</text><rect x='200' y='80' width='280.0' height='9' rx='2' fill='#5AD1FF'/><text x='488.0' y='88' font-family='ui-monospace,SFMono-Regular,Menlo,Consolas,monospace' font-size='10.5' font-weight='400' text-anchor='start' fill='#e6edf3'>2.139 s 路 llama.cpp</text><text x='700' y='70' font-family='ui-monospace,SFMono-Regular,Menlo,Consolas,monospace' font-size='12' font-weight='700' text-anchor='end' fill='#56D364'>9.64脳</text><text x='20' y='116' font-family='ui-monospace,SFMono-Regular,Menlo,Consolas,monospace' font-size='10.5' font-weight='400' text-anchor='start' fill='#8b949e'>fastest of the runs (min) 路 median 0.229 s vs 2.184 s (9.54脳) 路 AMD Ryzen 3 3200U</text><text x='20' y='134' font-family='ui-monospace,SFMono-Regular,Menlo,Consolas,monospace' font-size='10.5' font-weight='400' text-anchor='start' fill='#8b949e'>the model stayed resident in memory</text><text x='20' y='152' font-family='ui-monospace,SFMono-Regular,Menlo,Consolas,monospace' font-size='10.5' font-weight='400' text-anchor='start' fill='#8b949e'>matmul ceiling 4.50 tok/s vs 0.47 路 memory floor 0.122 s/token (2.13 GB read)</text><text x='20' y='170' font-family='ui-monospace,SFMono-Regular,Menlo,Consolas,monospace' font-size='10.5' font-weight='400' text-anchor='start' fill='#8b949e'>gate record testing/results/0.3.0.txt 路 matmuls only; since 0.3.0 bankML serves the whole model itself, token-identical</text></svg> | |