char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Fathi Boudracommon: migrate the deprecated --mmap/--no-mmap to --load-mode (#26934)0177dcc73

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
2.2 KiB82 linesraw
1#!/usr/bin/env bash
2
3RESULTS="bench-models-results.txt"
4: > "$RESULTS"
5
6ARGS_BB="-c 270336 -npp 512,4096,8192 -npl 1,2,4,8,16,32 -ntg 32"
7ARGS_B="-d 0,4096,8192,16384,32768 -p 2048 -n 32"
8
9QUICK=0
10DIO=0
11while (( "$#" )); do
12    case "$1" in
13        --quick) QUICK=1; shift ;;
14        --dio) DIO=1; shift ;;
15        *) shift ;;
16    esac
17done
18
19if (( QUICK )); then
20    ARGS_BB="-c 20480 -npp 512,4096 -npl 1,2,4 -ntg 32"
21    ARGS_B="-d 0 -p 2048 -n 32"
22fi
23
24if (( DIO )); then
25    ARGS_BB="${ARGS_BB} --load-mode dio"
26    ARGS_B="${ARGS_B} --load-mode dio"
27fi
28
29run_model() {
30    local HFR=$1
31    local HFF=$2
32
33    printf "## ${HFR}\n" | tee -a "$RESULTS"
34    printf "\n" | tee -a "$RESULTS"
35    printf "Model: https://huggingface.co/${HFR}\n" | tee -a "$RESULTS"
36    printf "\n" | tee -a "$RESULTS"
37
38    printf -- "- \`llama-batched-bench\`\n" | tee -a "$RESULTS"
39    printf "\n" | tee -a "$RESULTS"
40
41    ./bin/llama-batched-bench \
42        -hfr "${HFR}" -hff "${HFF}" \
43        -m "${HFF}" -fa 1 -ub 2048 \
44        ${ARGS_BB} | tee -a "$RESULTS"
45
46    printf "\n" | tee -a "$RESULTS"
47
48    printf -- "- \`llama-bench\`\n" | tee -a "$RESULTS"
49    printf "\n" | tee -a "$RESULTS"
50
51    ./bin/llama-bench \
52        -m "${HFF}" -fa 1 -ub 2048 \
53        ${ARGS_B} | tee -a "$RESULTS"
54
55    printf "\n" | tee -a "$RESULTS"
56
57    printf "\n"
58}
59
60run_model "ggml-org/gpt-oss-20b-GGUF"                       "gpt-oss-20b-mxfp4.gguf"
61run_model "ggml-org/gpt-oss-120b-GGUF"                      "gpt-oss-120b-mxfp4-00001-of-00003.gguf"
62run_model "ggml-org/Qwen3-Coder-30B-A3B-Instruct-Q8_0-GGUF" "qwen3-coder-30b-a3b-instruct-q8_0.gguf"
63run_model "ggml-org/Qwen2.5-Coder-7B-Q8_0-GGUF"             "qwen2.5-coder-7b-q8_0.gguf"
64run_model "ggml-org/gemma-3-4b-it-qat-GGUF"                 "gemma-3-4b-it-qat-Q4_0.gguf"
65run_model "ggml-org/GLM-4.7-Flash-GGUF"                     "GLM-4.7-Flash-Q8_0.gguf"
66
67if [[ -f models-extra.txt ]]; then
68    while read -r HFR HFF; do
69        [[ -z "$HFR" ]] && continue
70        run_model "$HFR" "$HFF"
71    done < models-extra.txt
72fi
73
74printf "\n=====================================\n"
75printf "\n"
76
77cat "$RESULTS"
78
79printf "\n"
80printf "Done! Results are written to $RESULTS\n"
81printf "\n"
82