抡锤者

生成个中国象棋的游戏，中间干预了两次，一次是让他把双方的颜色标注出来，还有一次是红方炮的初始位置不对，一共用时1个小时，这个正常吗？（而且，电脑还很弱智）

0.00.175.737 W srv llama_server: CORS is set to allow all origins ('*') and no API key is set
0.00.175.739 W srv llama_server: this can be a security risk (cross-origin attacks)
0.00.175.739 W srv llama_server: more info: https://github.com/ggml-org/llama.cpp/pull/25655
0.00.175.739 W srv llama_server: -----------------
0.00.182.354 I srv load_model: loading model 'models\Qwen3.8-27B-Uncensored-Q4_K_M.gguf'
0.01.248.431 W common_fit_params: failed to fit params to free device memory: n_gpu_layers already set by user to 999, abort
0.09.384.504 I cmn init: llama threadpool init, n_threads = 12
0.09.488.935 I common_speculative_init_result: creating MTP draft context against the target model 'models\Qwen3.8-27B-Uncensored-Q4_K_M.gguf'
0.10.493.721 I srv load_model: loaded multimodal model, 'models\mmproj-F16.gguf'
0.10.549.992 I srv load_model: initializing, n_slots = 1, n_ctx_slot = 65536, kv_unified = 'false'
0.10.596.625 I srv llama_server: model loaded
0.10.596.637 I srv llama_server: listening on http://127.0.0.1:8080
0.10.596.638 W srv llama_server: NOTICE: server default port will be changed to :9931 in a future release
0.10.596.638 W srv llama_server: ref: https://github.com/ggml-org/llama.cpp/pull/26508
0.33.561.997 I slot get_availabl: id 0 | task -1 | selected slot by LRU, t_last = -1
0.33.562.071 I slot launch_slot_: id 0 | task 0 | processing task, is_child = 0
0.37.231.323 I slot print_timing: id 0 | task 0 | n_gen = 133, tg = 43.95 t/s, tg_3s = 44.28 t/s
0.40.275.837 I slot print_timing: id 0 | task 0 | n_gen = 255, tg = 42.00 t/s, tg_3s = 40.07 t/s
0.43.296.586 I slot print_timing: id 0 | task 0 | n_gen = 404, tg = 44.44 t/s, tg_3s = 49.33 t/s
0.46.339.374 I slot print_timing: id 0 | task 0 | n_gen = 552, tg = 45.49 t/s, tg_3s = 48.64 t/s
0.47.880.707 I slot print_timing: id 0 | task 0 | prompt eval time = 665.40 ms / 60 tokens ( 11.09 ms per token, 90.17 tokens per second)
0.47.880.728 I slot print_timing: id 0 | task 0 | eval time = 13652.73 ms / 635 tokens ( 21.53 ms per token, 46.44 tokens per second)
0.47.880.732 I slot print_timing: id 0 | task 0 | total time = 14318.13 ms / 695 tokens
0.47.880.734 I slot print_timing: id 0 | task 0 | graphs reused = 282
0.47.880.751 I slot print_timing: id 0 | task 0 | draft acceptance = 0.61796 ( 351 accepted / 568 generated), mean len = 2.24
0.47.880.834 I slot release: id 0 | task 0 | stop processing: n_tokens = 695, truncated = 0

不是我不上传图片是，是总是错误，说我没有权限执行此操作

按照楼上老大的建议，上了MTP，速度立马提升到40多TOKEN/S了

@echo off
chcp 65001 >nul
title Qwen3.8-27B 越狱版 - Q4_K_M

cd /d "%~dp0"

llama-server.exe ^
-m "models\Qwen3.8-27B-Uncensored-Q4_K_M.gguf" ^
--mmproj "models\mmproj-F16.gguf" ^
--no-mmap ^
--image-min-tokens 1024 ^
-c 65536 ^
-ngl 999 ^
-n 16384 ^
-fa on ^
--cache-type-k q8_0 ^
--cache-type-v q8_0 ^
--cont-batching ^
--reasoning-preserve ^
--spec-type draft-mtp ^
--spec-draft-n-max 2 ^
-np 1
--host 127.0.0.1 ^
--port 8080

pause

还有能再提升速度，同时不降低精度的方法了吗？谢谢！

还有那位老大能在不丢失精度的情况下，加快速度的？

33token/s，如何接到HERMES，就只有不到30了，大概是28，29的样子

@echo off
chcp 65001 >nul
title Qwen3.8-27B 越狱版 - Q4_K_M

cd /d "%~dp0"

llama-server.exe ^
-m "models\Qwen3.8-27B-Uncensored-Q4_K_M.gguf" ^
--mmproj "models\mmproj-F16.gguf" ^
--no-mmap ^
--image-min-tokens 1024 ^
-c 65536 ^
-ngl 999 ^
-n 16384 ^
-fa on ^
--cache-type-k q8_0 ^
--cache-type-v q8_0 ^
--cont-batching ^
--reasoning-preserve ^
--host 127.0.0.1 ^
--port 8080

pause