2.37 MB
68 files
Updated 15 days ago
Name
Size
A100_decode_speedup_speedup_BitNet-2B-INT8.csv177 Bytes
xet
A100_decode_speedup_speedup_Llama3.2-1B-INT8.csv177 Bytes
xet
A100_decode_speedup_speedup_Llama3.2-3B-INT8.csv177 Bytes
xet
A100_decode_speedup_speedup_Qwen2.5-14B-INT8.csv177 Bytes
xet
A100_decode_speedup_speedup_Qwen2.5-7B-INT8.csv177 Bytes
xet
A100_kernel_INT8_total_Qwen2.5-7B.csv579 Bytes
xet
A100_prefill_absolute_Qwen2.5-7B-INT8.csv459 Bytes
xet
A100_prefill_speedup_Qwen2.5-7B-INT8.csv284 Bytes
xet
B200_decode_speedup_speedup_BitNet-2B-FP8.csv177 Bytes
xet
B200_decode_speedup_speedup_BitNet-2B-INT8.csv177 Bytes
xet
B200_decode_speedup_speedup_Llama3.2-1B-FP8.csv177 Bytes
xet
B200_decode_speedup_speedup_Llama3.2-1B-INT8.csv177 Bytes
xet
B200_decode_speedup_speedup_Llama3.2-3B-FP8.csv177 Bytes
xet
B200_decode_speedup_speedup_Llama3.2-3B-INT8.csv177 Bytes
xet
B200_decode_speedup_speedup_Qwen2.5-14B-FP8.csv177 Bytes
xet
B200_decode_speedup_speedup_Qwen2.5-14B-INT8.csv177 Bytes
xet
B200_decode_speedup_speedup_Qwen2.5-7B-FP8.csv177 Bytes
xet
B200_decode_speedup_speedup_Qwen2.5-7B-INT8.csv177 Bytes
xet
H100_kernel_FP8_total_Qwen2.5-7B.csv565 Bytes
xet
README.md456 Bytes
xet
RTX4090_decode_speedup_speedup_BitNet-2B-FP8.csv177 Bytes
xet
RTX4090_decode_speedup_speedup_BitNet-2B-INT8.csv177 Bytes
xet
RTX4090_decode_speedup_speedup_Llama3.2-1B-FP8.csv177 Bytes
xet
RTX4090_decode_speedup_speedup_Llama3.2-1B-INT8.csv177 Bytes
xet
RTX4090_decode_speedup_speedup_Llama3.2-3B-FP8.csv177 Bytes
xet
RTX4090_decode_speedup_speedup_Llama3.2-3B-INT8.csv177 Bytes
xet
RTX4090_decode_speedup_speedup_Qwen2.5-14B-FP8.csv161 Bytes
xet
RTX4090_decode_speedup_speedup_Qwen2.5-14B-INT8.csv161 Bytes
xet
RTX4090_decode_speedup_speedup_Qwen2.5-7B-FP8.csv177 Bytes
xet
RTX4090_decode_speedup_speedup_Qwen2.5-7B-INT8.csv177 Bytes
xet
RTX4090_kernel_FP8_total_Qwen2.5-7B.csv603 Bytes
xet
reported_Qwen3_reasoning_accuracy.csv128 Bytes
xet
README.md

Source data provenance

The CSV files in this directory are exact snapshots from the public SlideSparse repository at commit 92a8ca495b1c8cfe9ae5b1fc31badeb48a47ea65 (retrieved 2026-07-17), except reported_Qwen3_reasoning_accuracy.csv, which transcribes Figure 2 of arXiv:2603.05232v1. The reproduction runner recomputes ratios and averages from these snapshots rather than trusting the paper's prose.

Upstream: https://github.com/bcacdwk/vllmbench

Total size
2.37 MB
Files
68
Last updated
Jul 17
Pre-warmed CDN
US EU US EU

Contributors