{"id":2475,"date":"2026-07-31T11:31:56","date_gmt":"2026-07-31T03:31:56","guid":{"rendered":"https:\/\/www.o6s.net\/?p=2475"},"modified":"2026-07-31T12:08:42","modified_gmt":"2026-07-31T04:08:42","slug":"llama-cpp-%e5%8f%82%e6%95%b0%e5%ae%8c%e6%95%b4%e4%b8%ad%e6%96%87%e7%bf%bb%e8%af%91","status":"publish","type":"post","link":"https:\/\/www.o6s.net\/index.php\/2026\/07\/31\/llama-cpp-%e5%8f%82%e6%95%b0%e5%ae%8c%e6%95%b4%e4%b8%ad%e6%96%87%e7%bf%bb%e8%af%91\/","title":{"rendered":"llama.cpp \u53c2\u6570\u5b8c\u6574\u4e2d\u6587\u7ffb\u8bd1"},"content":{"rendered":"<div>\n<h1><span style=\"font-size: 16px;\">\u6ce8\uff1a\u6587\u540e\u9644llama.cpp\u73af\u5883\u53d8\u91cf\u7f16\u8f91\u5668\u53ca\u6e90\u7801<br \/>\n\u901a\u7528\u53c2\u6570 common params<\/span><\/h1>\n<\/div>\n<div>\n<div>\n<p><code>-h, --help, --usage<\/code><\/p>\n<div><\/div>\n<p>\u6253\u5370\u4f7f\u7528\u5e2e\u52a9\u5e76\u9000\u51fa\u7a0b\u5e8f<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--version<\/code><\/p>\n<div><\/div>\n<p>\u8f93\u51fa\u7a0b\u5e8f\u7248\u672c\u4e0e\u7f16\u8bd1\u6784\u5efa\u4fe1\u606f<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-cl, --cache-list<\/code><\/p>\n<div><\/div>\n<p>\u5217\u51fa\u672c\u5730\u7f13\u5b58\u4e2d\u5df2\u4e0b\u8f7d\u7684\u6a21\u578b<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--completion-bash<\/code><\/p>\n<div><\/div>\n<p>\u8f93\u51fa\u53ef\u76f4\u63a5\u5bfc\u5165\u7684 Bash \u81ea\u52a8\u8865\u5168\u811a\u672c<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-t, --threads N<\/code><\/p>\n<div><\/div>\n<p>\u6587\u672c\u751f\u6210\u9636\u6bb5\u4f7f\u7528\u7684 CPU \u7ebf\u7a0b\u6570\uff08\u9ed8\u8ba4\uff1a-1\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_THREADS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-tb, --threads-batch N<\/code><\/p>\n<div><\/div>\n<p>\u6279\u91cf\u5904\u7406\u3001\u63d0\u793a\u8bcd\u9884\u5904\u7406\u9636\u6bb5\u4f7f\u7528\u7684\u7ebf\u7a0b\u6570\uff08\u9ed8\u8ba4\u4e0e <code>--threads<\/code> \u4e00\u81f4\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-C, --cpu-mask M<\/code><\/p>\n<div><\/div>\n<p>CPU \u4eb2\u548c\u63a9\u7801\uff1a\u4efb\u610f\u957f\u5ea6\u5341\u516d\u8fdb\u5236\u5b57\u7b26\u4e32\uff0c\u4e0e <code>cpu-range<\/code> \u914d\u5408\u4f7f\u7528\uff08\u9ed8\u8ba4\u7a7a\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-Cr, --cpu-range lo-hi<\/code><\/p>\n<div><\/div>\n<p>\u7ed1\u5b9a\u4f7f\u7528\u7684 CPU \u6838\u5fc3\u533a\u95f4\uff0c\u4e0e <code>--cpu-mask<\/code> \u914d\u5408\u4f7f\u7528<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--cpu-strict &lt;0|1&gt;<\/code><\/p>\n<div><\/div>\n<p>\u5f00\u542f\u4e25\u683c CPU \u6838\u5fc3\u7ed1\u5b9a\uff08\u9ed8\u8ba4\uff1a0 \u5173\u95ed\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--prio N<\/code><\/p>\n<div><\/div>\n<p>\u8bbe\u7f6e\u8fdb\u7a0b \/ \u7ebf\u7a0b\u4f18\u5148\u7ea7\uff1a-1 \u4f4e\u30010 \u666e\u901a\u30011 \u4e2d\u7b49\u30012 \u9ad8\u30013 \u5b9e\u65f6\uff08\u9ed8\u8ba4\uff1a0\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--poll &lt;0...100&gt;<\/code><\/p>\n<div><\/div>\n<p>\u5de5\u4f5c\u8f6e\u8be2\u7b49\u5f85\u7b49\u7ea7\uff1b0 = \u4e0d\u542f\u7528\u8f6e\u8be2\uff08\u9ed8\u8ba4\uff1a50\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-Cb, --cpu-mask-batch M<\/code><\/p>\n<div><\/div>\n<p>\u6279\u91cf\u4efb\u52a1\u4e13\u7528 CPU \u4eb2\u548c\u63a9\u7801\uff0c\u914d\u5408 <code>cpu-range-batch<\/code>\uff08\u9ed8\u8ba4\u540c <code>--cpu-mask<\/code>\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-Crb, --cpu-range-batch lo-hi<\/code><\/p>\n<div><\/div>\n<p>\u6279\u91cf\u4efb\u52a1\u7ed1\u5b9a\u7684 CPU \u6838\u5fc3\u533a\u95f4\uff0c\u914d\u5408 <code>--cpu-mask-batch<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--cpu-strict-batch &lt;0|1&gt;<\/code><\/p>\n<div><\/div>\n<p>\u6279\u91cf\u4efb\u52a1\u542f\u7528\u4e25\u683c CPU \u7ed1\u5b9a\uff08\u9ed8\u8ba4\u540c <code>--cpu-strict<\/code>\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--prio-batch N<\/code><\/p>\n<div><\/div>\n<p>\u6279\u91cf\u4efb\u52a1\u7ebf\u7a0b\u4f18\u5148\u7ea7\uff1a0 \u666e\u901a\u30011 \u4e2d\u7b49\u30012 \u9ad8\u30013 \u5b9e\u65f6\uff08\u9ed8\u8ba4\uff1a0\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--poll-batch &lt;0|1&gt;<\/code><\/p>\n<div><\/div>\n<p>\u6279\u91cf\u4efb\u52a1\u542f\u7528\u5de5\u4f5c\u8f6e\u8be2\u7b49\u5f85\uff08\u9ed8\u8ba4\u540c <code>--poll<\/code>\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-c, --ctx-size N<\/code><\/p>\n<div><\/div>\n<p>\u4e0a\u4e0b\u6587\u7a97\u53e3\u5927\u5c0f\uff08\u9ed8\u8ba4\uff1a0\uff0c0 \u8868\u793a\u4ece\u6a21\u578b\u914d\u7f6e\u8bfb\u53d6\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CTX_SIZE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-n, --predict, --n-predict N<\/code><\/p>\n<div><\/div>\n<p>\u6700\u5927\u751f\u6210 token \u6570\u91cf\uff08\u9ed8\u8ba4\uff1a-1\uff0c-1 \u4ee3\u8868\u65e0\u9650\u751f\u6210\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_N_PREDICT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-b, --batch-size N<\/code><\/p>\n<div><\/div>\n<p>\u903b\u8f91\u6700\u5927\u6279\u91cf\u5c3a\u5bf8\uff08\u9ed8\u8ba4\uff1a2048\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_BATCH<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ub, --ubatch-size N<\/code><\/p>\n<div><\/div>\n<p>\u7269\u7406\u6700\u5927\u5fae\u6279\u91cf\u5c3a\u5bf8\uff08\u9ed8\u8ba4\uff1a512\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_UBATCH<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--keep N<\/code><\/p>\n<div><\/div>\n<p>\u4fdd\u7559\u539f\u59cb\u63d0\u793a\u8bcd\u524d N \u4e2a token \u4e0d\u88ab\u4e0a\u4e0b\u6587\u6ed1\u52a8\u4e22\u5f03\uff08\u9ed8\u8ba4\uff1a0\uff1b-1 = \u5168\u90e8\u4fdd\u7559\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--swa-full<\/code><\/p>\n<div><\/div>\n<p>\u542f\u7528\u5b8c\u6574\u5c3a\u5bf8 SWA \u7f13\u5b58\uff08\u9ed8\u8ba4\uff1afalse\uff09<\/p>\n<div><\/div>\n<p>\u53c2\u8003\u6587\u6863\uff1a<a title=\"autolink\" href=\"https:\/\/link.wtturl.cn\/?target=https%3A%2F%2Fgithub.com%2Fggml-org%2Fllama.cpp%2Fpull%2F13194%23issuecomment-2868343055&amp;scene=im&amp;aid=582478&amp;lang=zh\" target=\"_blank\" rel=\"noopener\">https:\/\/github.com\/ggml-org\/llama.cpp\/pull\/13194#issuecomment-2868343055<\/a><\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SWA_FULL<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-fa, --flash-attn [on|off|auto]<\/code><\/p>\n<div><\/div>\n<p>FlashAttention \u5f00\u5173\uff1aon \u5f00\u542f \/off \u5173\u95ed \/auto \u81ea\u52a8\uff08\u9ed8\u8ba4 auto\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_FLASH_ATTN<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--perf, --no-perf<\/code><\/p>\n<div><\/div>\n<p>\u5f00\u542f\u5185\u7f6e\u6027\u80fd\u8ba1\u65f6\u6253\u70b9\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_PERF<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-e, --escape, --no-escape<\/code><\/p>\n<div><\/div>\n<p>\u89e3\u6790\u8f6c\u4e49\u5b57\u7b26\uff08\\n \\r \\t &#8216; &#8221; \\\uff09\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--rope-scaling {none,linear,yarn}<\/code><\/p>\n<div><\/div>\n<p>RoPE \u9891\u7387\u7f29\u653e\u65b9\u6848\uff1b\u672a\u6307\u5b9a\u65f6\u6a21\u578b\u5185\u7f6e\u914d\u7f6e\u4f18\u5148\uff0c\u9ed8\u8ba4 linear<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_ROPE_SCALING_TYPE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--rope-scale N<\/code><\/p>\n<div><\/div>\n<p>RoPE \u4e0a\u4e0b\u6587\u7f29\u653e\u7cfb\u6570\uff0c\u4e0a\u4e0b\u6587\u7a97\u53e3\u6269\u5927 N \u500d<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_ROPE_SCALE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--rope-freq-base N<\/code><\/p>\n<div><\/div>\n<p>RoPE \u57fa\u7840\u9891\u7387\uff0c\u7528\u4e8e NTK \u7f29\u653e\uff08\u9ed8\u8ba4\u4ece\u6a21\u578b\u8bfb\u53d6\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_ROPE_FREQ_BASE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--rope-freq-scale N<\/code><\/p>\n<div><\/div>\n<p>RoPE \u9891\u7387\u7f29\u653e\u7cfb\u6570\uff0c\u4e0a\u4e0b\u6587\u7a97\u53e3\u7f29\u5c0f\u81f3 1\/N<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_ROPE_FREQ_SCALE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--yarn-orig-ctx N<\/code><\/p>\n<div><\/div>\n<p>YaRN \u7f29\u653e\uff1a\u6a21\u578b\u539f\u751f\u8bad\u7ec3\u4e0a\u4e0b\u6587\u957f\u5ea6\uff08\u9ed8\u8ba4 0\uff0c\u81ea\u52a8\u8bfb\u53d6\u6a21\u578b\u8bad\u7ec3\u7a97\u53e3\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_YARN_ORIG_CTX<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--yarn-ext-factor N<\/code><\/p>\n<div><\/div>\n<p>YaRN\uff1a\u5916\u63a8\u6df7\u5408\u7cfb\u6570\uff08\u9ed8\u8ba4 &#8211; 1.0\uff1b0.0 = \u7eaf\u63d2\u503c\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_YARN_EXT_FACTOR<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--yarn-attn-factor N<\/code><\/p>\n<div><\/div>\n<p>YaRN\uff1a\u6ce8\u610f\u529b\u5e45\u503c \/sqrt (t) \u7f29\u653e\u7cfb\u6570\uff08\u9ed8\u8ba4 &#8211; 1.0\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_YARN_ATTN_FACTOR<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--yarn-beta-slow N<\/code><\/p>\n<div><\/div>\n<p>YaRN\uff1a\u9ad8\u7ef4\u4fee\u6b63\u7cfb\u6570 alpha\uff08\u9ed8\u8ba4 &#8211; 1.0\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_YARN_BETA_SLOW<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--yarn-beta-fast N<\/code><\/p>\n<div><\/div>\n<p>YaRN\uff1a\u4f4e\u7ef4\u4fee\u6b63\u7cfb\u6570 beta\uff08\u9ed8\u8ba4 &#8211; 1.0\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_YARN_BETA_FAST<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-kvo, --kv-offload \/ -nkvo, --no-kv-offload<\/code><\/p>\n<div><\/div>\n<p>KV \u7f13\u5b58\u786c\u4ef6\u5378\u8f7d\u5f00\u5173\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_KV_OFFLOAD<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--repack \/ -nr, --no-repack<\/code><\/p>\n<div><\/div>\n<p>\u6743\u91cd\u91cd\u6253\u5305\u4f18\u5316\u5f00\u5173\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_REPACK<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--no-host<\/code><\/p>\n<div><\/div>\n<p>\u7ed5\u8fc7\u4e3b\u673a\u5185\u5b58\u7f13\u51b2\u533a\uff0c\u542f\u7528\u989d\u5916\u8bbe\u5907\u7f13\u51b2\u533a<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_NO_HOST<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ctk, --cache-type-k TYPE<\/code><\/p>\n<div><\/div>\n<p>KV \u7f13\u5b58 K \u5f20\u91cf\u6570\u636e\u7c7b\u578b<\/p>\n<div><\/div>\n<p>\u53ef\u9009\u503c\uff1af32\u3001f16\u3001bf16\u3001q8_0\u3001q4_0\u3001q4_1\u3001iq4_nl\u3001q5_0\u3001q5_1\uff08\u9ed8\u8ba4 f16\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CACHE_TYPE_K<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ctv, --cache-type-v TYPE<\/code><\/p>\n<div><\/div>\n<p>KV \u7f13\u5b58 V \u5f20\u91cf\u6570\u636e\u7c7b\u578b\uff0c\u53ef\u9009\u503c\u540c\u4e0a\uff08\u9ed8\u8ba4 f16\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CACHE_TYPE_V<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-dt, --defrag-thold N<\/code><\/p>\n<div><\/div>\n<p>KV \u7f13\u5b58\u788e\u7247\u6574\u7406\u9608\u503c\uff08\u5df2\u5f03\u7528\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_DEFRAG_THOLD<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--rpc SERVERS<\/code><\/p>\n<div><\/div>\n<p>RPC \u5206\u5e03\u5f0f\u8282\u70b9\u5730\u5740\u5217\u8868\uff0c\u591a\u8282\u70b9\u9017\u53f7\u5206\u9694 \u683c\u5f0f\uff1a\u4e3b\u673a\uff1a\u7aef\u53e3<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_RPC<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mlock<\/code><\/p>\n<div><\/div>\n<p>\u5df2\u5f03\u7528\uff0c\u63a8\u8350\u4f7f\u7528 <code>--load-mode<\/code>\uff1b\u5f3a\u5236\u7cfb\u7edf\u5c06\u6a21\u578b\u9501\u5b9a\u5728\u5185\u5b58\uff0c\u7981\u6b62\u4ea4\u6362\u538b\u7f29<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MLOCK<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mmap, --no-mmap<\/code><\/p>\n<div><\/div>\n<p>\u5df2\u5f03\u7528\uff0c\u63a8\u8350 <code>--load-mode<\/code>\uff1b\u662f\u5426\u4f7f\u7528\u5185\u5b58\u6620\u5c04\u52a0\u8f7d\u6a21\u578b\uff1b\u5173\u95ed\u65f6\u52a0\u8f7d\u66f4\u6162\uff0c\u4f46\u4e0d\u9501\u5b9a\u5185\u5b58\u53ef\u51cf\u5c11\u9875\u9762\u7f6e\u6362<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MMAP<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-dio, --direct-io \/ -ndio, --no-direct-io<\/code><\/p>\n<div><\/div>\n<p>\u5df2\u5f03\u7528\uff0c\u63a8\u8350 <code>--load-mode<\/code>\uff1b\u53ef\u7528\u65f6\u542f\u7528 DirectIO \u76f4\u8bfb<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_DIO<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-lm, --load-mode MODE<\/code><\/p>\n<div><\/div>\n<p>\u6a21\u578b\u52a0\u8f7d\u6a21\u5f0f\uff08\u9ed8\u8ba4 mmap\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<ul>\n<li>none\uff1a\u65e0\u7279\u6b8a\u52a0\u8f7d\u7b56\u7565<\/li>\n<li>mmap\uff1a\u5185\u5b58\u6620\u5c04\u52a0\u8f7d<\/li>\n<li>mlock\uff1a\u9501\u5b9a\u6a21\u578b\u81f3\u7269\u7406\u5185\u5b58\uff0c\u7981\u6b62\u4ea4\u6362<\/li>\n<li>mmap+mlock\uff1a\u6620\u5c04 + \u5185\u5b58\u9501\u5b9a\u7ec4\u5408<\/li>\n<li>dio\uff1a\u76f4\u8bfb IO \u6a21\u5f0f\uff08\u786c\u4ef6\u652f\u6301\u65f6\u751f\u6548\uff09\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_LOAD_MODE<\/code><\/li>\n<\/ul>\n<\/div>\n<div>\n<div>\n<p><code>--numa TYPE<\/code><\/p>\n<div><\/div>\n<p>\u591a NUMA \u8282\u70b9\u4f18\u5316\u7b56\u7565<\/p>\n<\/div>\n<\/div>\n<div>\n<ul>\n<li>distribute\uff1a\u4efb\u52a1\u5747\u5300\u5206\u53d1\u81f3\u6240\u6709 NUMA \u8282\u70b9<\/li>\n<li>isolate\uff1a\u4ec5\u5728\u542f\u52a8\u8fdb\u7a0b\u7684 NUMA \u8282\u70b9\u521b\u5efa\u7ebf\u7a0b<\/li>\n<li>numactl\uff1a\u8bfb\u53d6\u7cfb\u7edf numactl \u914d\u7f6e\u7ed1\u5b9a CPU\n<div><\/div>\n<p>\u9996\u6b21\u4f7f\u7528\u5efa\u8bae\u6e05\u7a7a\u7cfb\u7edf\u9875\u7f13\u5b58<\/p>\n<div><\/div>\n<p>\u53c2\u8003\uff1a<a title=\"autolink\" href=\"https:\/\/link.wtturl.cn\/?target=https%3A%2F%2Fgithub.com%2Fggml-org%2Fllama.cpp%2Fissues%2F1437&amp;scene=im&amp;aid=582478&amp;lang=zh\" target=\"_blank\" rel=\"noopener\">https:\/\/github.com\/ggml-org\/llama.cpp\/issues\/1437<\/a><\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_NUMA<\/code><\/li>\n<\/ul>\n<\/div>\n<div>\n<div>\n<p><code>-dev, --device &lt;dev1,dev2,..&gt;<\/code><\/p>\n<div><\/div>\n<p>\u7528\u4e8e\u6a21\u578b\u5378\u8f7d\u7684\u786c\u4ef6\u8bbe\u5907\u5217\u8868\uff0c\u9017\u53f7\u5206\u9694\uff1bnone = \u4e0d\u5378\u8f7d<\/p>\n<div><\/div>\n<p>\u4f7f\u7528 <code>--list-devices<\/code> \u67e5\u770b\u53ef\u7528\u786c\u4ef6<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_DEVICE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--list-devices<\/code><\/p>\n<div><\/div>\n<p>\u6253\u5370\u672c\u673a\u53ef\u7528\u52a0\u901f\u8bbe\u5907\u5217\u8868\u5e76\u9000\u51fa<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ot, --override-tensor &lt;\u5f20\u91cf\u5339\u914d\u89c4\u5219&gt;=&lt;\u7f13\u5b58\u7c7b\u578b&gt;,...<\/code><\/p>\n<div><\/div>\n<p>\u5f3a\u5236\u6307\u5b9a\u7279\u5b9a\u5f20\u91cf\u4f7f\u7528\u7684\u7f13\u51b2\u533a\u7c7b\u578b<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_OVERRIDE_TENSOR<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-cmoe, --cpu-moe<\/code><\/p>\n<div><\/div>\n<p>\u6240\u6709 MoE \u4e13\u5bb6\u5c42\u6743\u91cd\u4fdd\u7559\u5728 CPU \u5185\u5b58<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CPU_MOE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ncmoe, --n-cpu-moe N<\/code><\/p>\n<div><\/div>\n<p>\u524d N \u5c42 MoE \u4e13\u5bb6\u6743\u91cd\u4fdd\u7559\u5728 CPU \u5185\u5b58<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_N_CPU_MOE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ngl, --gpu-layers, --n-gpu-layers N<\/code><\/p>\n<div><\/div>\n<p>\u52a0\u8f7d\u81f3\u663e\u5b58\u7684\u6a21\u578b\u6700\u5927\u5c42\u6570\uff1b\u652f\u6301\u6570\u5b57 \/auto\/all\uff08\u9ed8\u8ba4 auto\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_N_GPU_LAYERS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-sm, --split-mode {none,layer,row,tensor}<\/code><\/p>\n<div><\/div>\n<p>\u591a GPU \u6a21\u578b\u62c6\u5206\u7b56\u7565<\/p>\n<\/div>\n<\/div>\n<div>\n<ul>\n<li>none\uff1a\u4ec5\u5355\u5361\u8fd0\u884c<\/li>\n<li>layer\uff08\u9ed8\u8ba4\uff09\uff1a\u5c42\u6d41\u6c34\u7ebf\u62c6\u5206\uff0cKV \u7f13\u5b58\u968f\u5c42\u5206\u914d<\/li>\n<li>row\uff1a\u6743\u91cd\u6309\u884c\u5e76\u884c\u62c6\u5206<\/li>\n<li>tensor\uff1a\u5b8c\u6574\u5f20\u91cf\u5e76\u884c\u62c6\u5206\uff08\u5b9e\u9a8c\u6027\u529f\u80fd\uff09\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPLIT_MODE<\/code><\/li>\n<\/ul>\n<\/div>\n<div>\n<div>\n<p><code>-ts, --tensor-split N0,N1,N2,...<\/code><\/p>\n<div><\/div>\n<p>\u591a\u5361\u6743\u91cd\u5206\u914d\u6bd4\u4f8b\uff0c\u9017\u53f7\u5206\u9694\u6570\u5b57\uff0c\u4f8b 3,1<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_TENSOR_SPLIT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-mg, --main-gpu INDEX<\/code><\/p>\n<div><\/div>\n<p>\u4e3b GPU \u7f16\u53f7\uff1asplit-mode=none \u65f6\u5168\u5c40\u4e3b\u5361\uff1bsplit-mode=row \u65f6\u5b58\u653e\u4e2d\u95f4\u8ba1\u7b97\u4e0e KV \u7f13\u5b58\uff08\u9ed8\u8ba4 0\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MAIN_GPU<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-fit, --fit [on|off]<\/code><\/p>\n<div><\/div>\n<p>\u81ea\u52a8\u8c03\u6574\u672a\u6307\u5b9a\u53c2\u6570\uff0c\u9002\u914d\u8bbe\u5907\u663e\u5b58\u5bb9\u91cf\uff08\u9ed8\u8ba4 on\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_FIT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-fitt, --fit-target MiB0,MiB1,MiB2,...<\/code><\/p>\n<div><\/div>\n<p>\u81ea\u52a8\u9002\u914d\u529f\u80fd\u4e3a\u6bcf\u5f20\u8bbe\u5907\u9884\u7559\u663e\u5b58\u4f59\u91cf\uff08\u5355\u4f4d MiB\uff09\uff1b\u5355\u6570\u503c\u5168\u5c40\u751f\u6548\uff0c\u9ed8\u8ba4 1024<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_FIT_TARGET<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-fitc, --fit-ctx N<\/code><\/p>\n<div><\/div>\n<p>\u81ea\u52a8\u9002\u914d\u529f\u80fd\u5141\u8bb8\u8bbe\u7f6e\u7684\u6700\u5c0f\u4e0a\u4e0b\u6587\u7a97\u53e3\u5927\u5c0f\uff0c\u9ed8\u8ba4 4096<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_FIT_CTX<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--check-tensors<\/code><\/p>\n<div><\/div>\n<p>\u6821\u9a8c\u6a21\u578b\u5f20\u91cf\u662f\u5426\u5b58\u5728\u975e\u6cd5\u5f02\u5e38\u503c\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--override-kv KEY=TYPE:VALUE,...<\/code><\/p>\n<div><\/div>\n<p>\u9ad8\u7ea7\u53c2\u6570\uff1a\u6309\u952e\u8986\u76d6\u6a21\u578b\u5143\u6570\u636e\uff0c\u591a\u7ec4\u9017\u53f7\u5206\u9694<\/p>\n<div><\/div>\n<p>\u652f\u6301\u7c7b\u578b int\/float\/bool\/str<\/p>\n<div><\/div>\n<p>\u793a\u4f8b\uff1a<code>--override-kv tokenizer.ggml.add_bos_token=bool:false,tokenizer.ggml.add_eos_token=bool:false<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--op-offload, --no-op-offload<\/code><\/p>\n<div><\/div>\n<p>\u4e3b\u673a\u5f20\u91cf\u8fd0\u7b97\u5378\u8f7d\u81f3\u52a0\u901f\u8bbe\u5907\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--lora FNAME<\/code><\/p>\n<div><\/div>\n<p>LoRA \u5fae\u8c03\u9002\u914d\u5668\u6587\u4ef6\u8def\u5f84\uff1b\u591a\u9002\u914d\u5668\u9017\u53f7\u5206\u9694\u52a0\u8f7d<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--lora-scaled FNAME:SCALE,...<\/code><\/p>\n<div><\/div>\n<p>\u5e26\u81ea\u5b9a\u4e49\u7f29\u653e\u7cfb\u6570\u7684 LoRA \u9002\u914d\u5668\uff0c\u683c\u5f0f \u6587\u4ef6\u8def\u5f84\uff1a\u7f29\u653e\u503c\uff0c\u591a\u7ec4\u9017\u53f7\u5206\u9694<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--control-vector FNAME<\/code><\/p>\n<div><\/div>\n<p>\u52a0\u8f7d\u63a7\u5236\u5411\u91cf\u6587\u4ef6\uff1b\u591a\u6587\u4ef6\u9017\u53f7\u5206\u9694\u53e0\u52a0<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--control-vector-scaled FNAME:SCALE,...<\/code><\/p>\n<div><\/div>\n<p>\u5e26\u81ea\u5b9a\u4e49\u7f29\u653e\u7cfb\u6570\u7684\u63a7\u5236\u5411\u91cf\uff0c\u683c\u5f0f \u6587\u4ef6\uff1a\u7cfb\u6570<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--control-vector-layer-range START END<\/code><\/p>\n<div><\/div>\n<p>\u63a7\u5236\u5411\u91cf\u751f\u6548\u5c42\u533a\u95f4\uff0c\u9996\u5c3e\u5c42\u5747\u5305\u542b<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-m, --model FNAME<\/code><\/p>\n<div><\/div>\n<p>\u672c\u5730\u6a21\u578b\u6587\u4ef6\u8def\u5f84<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MODEL<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-mu, --model-url MODEL_URL<\/code><\/p>\n<div><\/div>\n<p>\u6a21\u578b\u5728\u7ebf\u4e0b\u8f7d\u94fe\u63a5\uff08\u9ed8\u8ba4\u4e0d\u542f\u7528\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MODEL_URL<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-dr, --docker-repo [&lt;\u4ed3\u5e93&gt;\/]&lt;\u6a21\u578b&gt;[:\u91cf\u5316\u7248\u672c]<\/code><\/p>\n<div><\/div>\n<p>Docker Hub \u6a21\u578b\u4ed3\u5e93\u5730\u5740<\/p>\n<div><\/div>\n<p>\u4ed3\u5e93\u540d\u53ef\u9009\uff0c\u9ed8\u8ba4 ai\/\uff1b\u91cf\u5316\u6807\u7b7e\u53ef\u9009\uff0c\u9ed8\u8ba4 latest<\/p>\n<div><\/div>\n<p>\u793a\u4f8b\uff1agemma3<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_DOCKER_REPO<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-hf, -hfr, --hf-repo &lt;\u7528\u6237\u540d&gt;\/&lt;\u6a21\u578b\u4ed3\u5e93&gt;[:\u91cf\u5316]<\/code><\/p>\n<div><\/div>\n<p>HuggingFace \u6a21\u578b\u4ed3\u5e93\u5730\u5740\uff0c\u91cf\u5316\u6807\u7b7e\u5927\u5c0f\u5199\u4e0d\u654f\u611f\uff0c\u9ed8\u8ba4 Q4_K_M\uff1b\u65e0\u8be5\u91cf\u5316\u5219\u4e0b\u8f7d\u4ed3\u5e93\u9996\u4e2a GGUF \u6587\u4ef6<\/p>\n<div><\/div>\n<p>\u591a\u6a21\u6001\u6a21\u578b\u81ea\u52a8\u4e0b\u8f7d mmproj \u6295\u5f71\u6587\u4ef6\uff0c\u6dfb\u52a0<code>--no-mmproj<\/code>\u53ef\u5173\u95ed<\/p>\n<div><\/div>\n<p>\u793a\u4f8b\uff1aggml-org\/GLM-4.7-Flash-GGUF:Q4_K_M<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_HF_REPO<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-hff, --hf-file FILE<\/code><\/p>\n<div><\/div>\n<p>\u6307\u5b9a\u4e0b\u8f7d HF \u4ed3\u5e93\u5185\u7279\u5b9a\u6a21\u578b\u6587\u4ef6\uff0c\u8986\u76d6<code>--hf-repo<\/code>\u7684\u91cf\u5316\u53c2\u6570<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_HF_FILE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-hfv, -hfrv, --hf-repo-v &lt;\u7528\u6237\u540d&gt;\/&lt;\u6a21\u578b&gt;[:\u91cf\u5316]<\/code><\/p>\n<div><\/div>\n<p>\u97f3\u9891\u58f0\u7801\u5668\u4e13\u7528 HF \u4ed3\u5e93\uff08\u9ed8\u8ba4\u4e0d\u542f\u7528\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_HF_REPO_V<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-hffv, --hf-file-v FILE<\/code><\/p>\n<div><\/div>\n<p>\u58f0\u7801\u5668\u6a21\u578b\u6307\u5b9a\u6587\u4ef6\u8def\u5f84<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_HF_FILE_V<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-hft, --hf-token TOKEN<\/code><\/p>\n<div><\/div>\n<p>HuggingFace \u8bbf\u95ee\u4ee4\u724c\uff1b\u9ed8\u8ba4\u8bfb\u53d6\u73af\u5883\u53d8\u91cf HF_TOKEN<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>HF_TOKEN<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--log-disable<\/code><\/p>\n<div><\/div>\n<p>\u5173\u95ed\u5168\u90e8\u65e5\u5fd7\u8f93\u51fa<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--log-file FNAME<\/code><\/p>\n<div><\/div>\n<p>\u65e5\u5fd7\u8f93\u51fa\u81f3\u6307\u5b9a\u6587\u4ef6<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_LOG_FILE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--log-colors [on|off|auto]<\/code><\/p>\n<div><\/div>\n<p>\u5f69\u8272\u65e5\u5fd7\u5f00\u5173\uff1bauto \u4ec5\u7ec8\u7aef\u8f93\u51fa\u542f\u7528\u8272\u5f69\uff08\u9ed8\u8ba4 auto\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_LOG_COLORS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-v, --verbose, --log-verbose<\/code><\/p>\n<div><\/div>\n<p>\u65e5\u5fd7\u7b49\u7ea7\u62c9\u6ee1\uff0c\u8f93\u51fa\u5168\u90e8\u8c03\u8bd5\u4fe1\u606f\uff0c\u6392\u9519\u4e13\u7528<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--offline<\/code><\/p>\n<div><\/div>\n<p>\u79bb\u7ebf\u6a21\u5f0f\uff1a\u5f3a\u5236\u4f7f\u7528\u672c\u5730\u7f13\u5b58\uff0c\u963b\u65ad\u7f51\u7edc\u8bf7\u6c42<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_OFFLINE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-lv, --verbosity, --log-verbosity N<\/code><\/p>\n<div><\/div>\n<p>\u65e5\u5fd7\u8f93\u51fa\u9608\u503c\uff0c\u9ad8\u4e8e\u8be5\u7b49\u7ea7\u65e5\u5fd7\u4f1a\u88ab\u8fc7\u6ee4<\/p>\n<div><\/div>\n<p>0\uff1a\u666e\u901a\u8f93\u51fa<\/p>\n<div><\/div>\n<p>1\uff1a\u9519\u8bef<\/p>\n<div><\/div>\n<p>2\uff1a\u8b66\u544a<\/p>\n<div><\/div>\n<p>3\uff1a\u4fe1\u606f\uff08\u9ed8\u8ba4\uff09<\/p>\n<div><\/div>\n<p>4\uff1a\u8be6\u7ec6\u8ffd\u8e2a<\/p>\n<div><\/div>\n<p>5\uff1a\u8c03\u8bd5<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_LOG_VERBOSITY<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--log-prefix, --no-log-prefix<\/code><\/p>\n<div><\/div>\n<p>\u65e5\u5fd7\u884c\u8f93\u51fa\u6a21\u5757\u524d\u7f00\u6807\u8bc6<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--log-timestamps, --no-log-timestamps<\/code><\/p>\n<div><\/div>\n<p>\u65e5\u5fd7\u884c\u9644\u52a0\u65f6\u95f4\u6233<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-type-k, -ctkd, --cache-type-k-draft TYPE<\/code><\/p>\n<div><\/div>\n<p>\u63a8\u6d4b\u89e3\u7801\u8349\u7a3f\u6a21\u578b K \u5f20\u91cf\u7f13\u5b58\u6570\u636e\u7c7b\u578b\uff0c\u53ef\u9009\u503c\u540c\u4e3b\u6a21\u578b KV \u7f13\u5b58\uff08\u9ed8\u8ba4 f16\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_CACHE_TYPE_K<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-type-v, -ctvd, --cache-type-v-draft TYPE<\/code><\/p>\n<div><\/div>\n<p>\u63a8\u6d4b\u89e3\u7801\u8349\u7a3f\u6a21\u578b V \u5f20\u91cf\u7f13\u5b58\u6570\u636e\u7c7b\u578b\uff08\u9ed8\u8ba4 f16\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_CACHE_TYPE_V<\/code><\/p>\n<\/div>\n<\/div>\n<hr \/>\n<div>\n<h2>\u91c7\u6837\u53c2\u6570 sampling params<\/h2>\n<\/div>\n<div>\n<div>\n<p><code>--samplers SAMPLERS<\/code><\/p>\n<div><\/div>\n<p>\u6587\u672c\u751f\u6210\u91c7\u6837\u5668\u6267\u884c\u987a\u5e8f\uff0c\u5206\u53f7\u5206\u9694<\/p>\n<div><\/div>\n<p>\u9ed8\u8ba4\u987a\u5e8f\uff1apenalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-s, --seed SEED<\/code><\/p>\n<div><\/div>\n<p>\u968f\u673a\u6570\u79cd\u5b50\uff1b-1 = \u968f\u673a\u79cd\u5b50\uff08\u9ed8\u8ba4 &#8211; 1\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--sampler-seq, --sampling-seq SEQUENCE<\/code><\/p>\n<div><\/div>\n<p>\u7b80\u5316\u91c7\u6837\u5668\u5e8f\u5217\u7b80\u5199\uff0c\u9ed8\u8ba4 edskypmxt<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--ignore-eos<\/code><\/p>\n<div><\/div>\n<p>\u5ffd\u7565\u7ed3\u675f\u7b26 EOS\uff0c\u6301\u7eed\u751f\u6210\u6587\u672c\uff1b\u7b49\u6548\u81ea\u52a8\u5f00\u542f<code>--logit-bias EOS-inf<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--temp, --temperature N<\/code><\/p>\n<div><\/div>\n<p>\u751f\u6210\u6e29\u5ea6\uff0c\u63a7\u5236\u968f\u673a\u6027\uff08\u9ed8\u8ba4 0.80\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--top-k N<\/code><\/p>\n<div><\/div>\n<p>Top-K \u91c7\u6837\uff1b0 = \u5173\u95ed\uff08\u9ed8\u8ba4 40\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_TOP_K<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--top-p N<\/code><\/p>\n<div><\/div>\n<p>Top-P \u6838\u91c7\u6837\uff1b1.0 = \u5173\u95ed\uff08\u9ed8\u8ba4 0.95\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--min-p N<\/code><\/p>\n<div><\/div>\n<p>Min-P \u91c7\u6837\uff1b0.0 = \u5173\u95ed\uff08\u9ed8\u8ba4 0.05\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--top-nsigma, --top-n-sigma N<\/code><\/p>\n<div><\/div>\n<p>Top-N-Sigma \u91c7\u6837\uff1b-1 = \u5173\u95ed\uff08\u9ed8\u8ba4 &#8211; 1.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--xtc-probability N<\/code><\/p>\n<div><\/div>\n<p>XTC \u91c7\u6837\u6982\u7387\u9608\u503c\uff1b0 = \u5173\u95ed\uff08\u9ed8\u8ba4 0.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--xtc-threshold N<\/code><\/p>\n<div><\/div>\n<p>XTC \u8fc7\u6ee4\u9608\u503c\uff1b1.0 = \u5173\u95ed\uff08\u9ed8\u8ba4 0.10\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--typical, --typical-p N<\/code><\/p>\n<div><\/div>\n<p>\u5c40\u90e8\u5178\u578b\u91c7\u6837\uff1b1.0 = \u5173\u95ed\uff08\u9ed8\u8ba4 1.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--repeat-last-n N<\/code><\/p>\n<div><\/div>\n<p>\u91cd\u590d\u60e9\u7f5a\u7edf\u8ba1\u7a97\u53e3 token \u6570\u91cf\uff1b0 \u5173\u95ed\u3001-1 \u7b49\u4e8e\u4e0a\u4e0b\u6587\u603b\u957f\u5ea6\uff08\u9ed8\u8ba4 64\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--repeat-penalty N<\/code><\/p>\n<div><\/div>\n<p>\u91cd\u590d\u5e8f\u5217\u60e9\u7f5a\u7cfb\u6570\uff1b1.0 \u5173\u95ed\u60e9\u7f5a\uff08\u9ed8\u8ba4 1.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--presence-penalty N<\/code><\/p>\n<div><\/div>\n<p>\u5b58\u5728\u60e9\u7f5a\uff1b0 \u5173\u95ed\uff08\u9ed8\u8ba4 0.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--frequency-penalty N<\/code><\/p>\n<div><\/div>\n<p>\u9891\u7387\u60e9\u7f5a\uff1b0 \u5173\u95ed\uff08\u9ed8\u8ba4 0.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--dry-multiplier N<\/code><\/p>\n<div><\/div>\n<p>DRY \u53bb\u91cd\u590d\u91c7\u6837\u500d\u7387\uff1b0 \u5173\u95ed\uff08\u9ed8\u8ba4 0.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--dry-base N<\/code><\/p>\n<div><\/div>\n<p>DRY \u91c7\u6837\u57fa\u7840\u7cfb\u6570\uff08\u9ed8\u8ba4 1.75\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--dry-allowed-length N<\/code><\/p>\n<div><\/div>\n<p>DRY \u68c0\u6d4b\u91cd\u590d\u5e8f\u5217\u6700\u5c0f\u957f\u5ea6\uff08\u9ed8\u8ba4 2\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--dry-penalty-last-n N<\/code><\/p>\n<div><\/div>\n<p>DRY \u60e9\u7f5a\u7edf\u8ba1\u7a97\u53e3\uff1b0 \u5173\u95ed\u3001-1 \u7b49\u4e8e\u4e0a\u4e0b\u6587\u957f\u5ea6\uff08\u9ed8\u8ba4 &#8211; 1\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--dry-sequence-breaker STRING<\/code><\/p>\n<div><\/div>\n<p>\u81ea\u5b9a\u4e49 DRY \u5e8f\u5217\u5206\u9694\u7b26\uff0c\u8986\u76d6\u9ed8\u8ba4\u6362\u884c \/ \u5192\u53f7 \/ \u5f15\u53f7 \/*\uff1b\u4f20\u5165 none \u7981\u7528\u5206\u9694\u7b26<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--adaptive-target N<\/code><\/p>\n<div><\/div>\n<p>\u81ea\u9002\u5e94 P \u91c7\u6837\u76ee\u6807\u6982\u7387\u533a\u95f4\uff1b\u8d1f\u6570\u5173\u95ed\uff08\u9ed8\u8ba4 &#8211; 1.00\uff0c\u8303\u56f4 0~1\uff09<\/p>\n<div><\/div>\n<p>\u53c2\u8003\uff1a<a title=\"autolink\" href=\"https:\/\/link.wtturl.cn\/?target=https%3A%2F%2Fgithub.com%2Fggml-org%2Fllama.cpp%2Fpull%2F17927&amp;scene=im&amp;aid=582478&amp;lang=zh\" target=\"_blank\" rel=\"noopener\">https:\/\/github.com\/ggml-org\/llama.cpp\/pull\/17927<\/a><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--adaptive-decay N<\/code><\/p>\n<div><\/div>\n<p>\u81ea\u9002\u5e94 P \u8870\u51cf\u7cfb\u6570\uff1b\u6570\u503c\u8d8a\u4f4e\u54cd\u5e94\u8d8a\u5feb\u3001\u8d8a\u9ad8\u8d8a\u7a33\u5b9a\uff080~0.99\uff0c\u9ed8\u8ba4 0.90\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--dynatemp-range N<\/code><\/p>\n<div><\/div>\n<p>\u52a8\u6001\u6e29\u5ea6\u533a\u95f4\uff1b0 \u5173\u95ed\uff08\u9ed8\u8ba4 0.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--dynatemp-exp N<\/code><\/p>\n<div><\/div>\n<p>\u52a8\u6001\u6e29\u5ea6\u6307\u6570\u7cfb\u6570\uff08\u9ed8\u8ba4 1.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mirostat N<\/code><\/p>\n<div><\/div>\n<p>Mirostat \u91c7\u6837\u5f00\u5173\uff1b0 \u5173\u95ed\u30011=Mirostat v1\u30012=Mirostat 2.0<\/p>\n<div><\/div>\n<p>\u542f\u7528\u540e\u81ea\u52a8\u5ffd\u7565 TopK\u3001\u6838\u91c7\u6837\u3001\u5178\u578b\u91c7\u6837\uff08\u9ed8\u8ba4 0\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mirostat-lr N<\/code><\/p>\n<div><\/div>\n<p>Mirostat \u5b66\u4e60\u7387 eta\uff08\u9ed8\u8ba4 0.10\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mirostat-ent N<\/code><\/p>\n<div><\/div>\n<p>Mirostat \u76ee\u6807\u71b5 tau\uff08\u9ed8\u8ba4 5.00\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-l, --logit-bias TOKEN_ID(+\/-)BIAS<\/code><\/p>\n<div><\/div>\n<p>\u4fee\u6539\u6307\u5b9a token \u751f\u6210\u6982\u7387<\/p>\n<div><\/div>\n<p>\u793a\u4f8b\uff1a<code>--logit-bias 15043+1<\/code> \u63d0\u9ad8 Hello \u751f\u6210\u6982\u7387\uff1b<code>15043-1<\/code>\u964d\u4f4e<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--grammar GRAMMAR<\/code><\/p>\n<div><\/div>\n<p>BNF \u683c\u5f0f\u8bed\u6cd5\u7ea6\u675f\uff0c\u9650\u5236\u6a21\u578b\u8f93\u51fa\u683c\u5f0f\uff1b\u793a\u4f8b\u6587\u4ef6\u89c1 grammars \u76ee\u5f55<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--grammar-file FNAME<\/code><\/p>\n<div><\/div>\n<p>\u4ece\u6587\u4ef6\u8bfb\u53d6\u8bed\u6cd5\u7ea6\u675f\u89c4\u5219<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-j, --json-schema SCHEMA<\/code><\/p>\n<div><\/div>\n<p>JSON Schema \u7ea6\u675f\u751f\u6210 JSON \u8f93\u51fa\uff0c<code>{}<\/code>\u4ee3\u8868\u4efb\u610f JSON \u5bf9\u8c61<\/p>\n<div><\/div>\n<p>\u5916\u90e8 $ref \u5f15\u7528 schema \u9700\u6539\u7528 grammar + \u914d\u5957\u8f6c\u6362\u811a\u672c<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-jf, --json-schema-file FILE<\/code><\/p>\n<div><\/div>\n<p>\u6587\u4ef6\u52a0\u8f7d JSON \u683c\u5f0f\u7ea6\u675f\u89c4\u5219\uff0c\u5916\u90e8\u5f15\u7528\u540c\u4e0a\u8ff0\u9650\u5236<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-bs, --backend-sampling<\/code><\/p>\n<div><\/div>\n<p>\u540e\u7aef\u786c\u4ef6\u91c7\u6837\u52a0\u901f\uff08\u5b9e\u9a8c\u6027\u529f\u80fd\uff0c\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_BACKEND_SAMPLING<\/code><\/p>\n<\/div>\n<\/div>\n<hr \/>\n<div>\n<h2>\u63a8\u6d4b\u89e3\u7801\u53c2\u6570 speculative params<\/h2>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-hf, -hfd, -hfrd, --hf-repo-draft &lt;\u7528\u6237&gt;\/&lt;\u6a21\u578b&gt;[:\u91cf\u5316]<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u4e13\u7528 HF \u4ed3\u5e93\uff0c\u53c2\u6570\u89c4\u5219\u540c<code>--hf-repo<\/code>\uff08\u9ed8\u8ba4\u4e0d\u542f\u7528\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_HF_REPO<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-threads, -td, --threads-draft N<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u751f\u6210\u9636\u6bb5 CPU \u7ebf\u7a0b\u6570\uff08\u9ed8\u8ba4\u540c\u4e3b\u6a21\u578b &#8211;threads\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-threads-batch, -tbd, --threads-batch-draft N<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u6279\u91cf\u9884\u5904\u7406\u7ebf\u7a0b\uff08\u540c\u8349\u7a3f\u4e3b\u7ebf\u7a0b\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-cpu-mask, -Cd, --cpu-mask-draft M<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b CPU \u4eb2\u548c\u63a9\u7801\uff0c\u914d\u5408 cpu-range-draft\uff08\u9ed8\u8ba4\u540c\u4e3b\u6a21\u578b\u63a9\u7801\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-cpu-range, -Crd, --cpu-range-draft lo-hi<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u7ed1\u5b9a CPU \u6838\u5fc3\u533a\u95f4<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-cpu-strict, --cpu-strict-draft &lt;0|1&gt;<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u542f\u7528\u4e25\u683c CPU \u7ed1\u5b9a\uff08\u9ed8\u8ba4\u540c\u4e3b\u6a21\u578b\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-prio, --prio-draft N<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u8fdb\u7a0b\u7ebf\u7a0b\u4f18\u5148\u7ea7\uff1a0 \u666e\u901a \/ 1 \u4e2d\u7b49 \/ 2 \u9ad8 \/ 3 \u5b9e\u65f6\uff08\u9ed8\u8ba4 0\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-poll, --poll-draft &lt;0|1&gt;<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u4efb\u52a1\u542f\u7528\u8f6e\u8be2\u7b49\u5f85\uff08\u9ed8\u8ba4\u540c\u4e3b\u6a21\u578b poll\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-cpu-mask-batch, -Cbd, --cpu-mask-batch-draft M<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6279\u91cf\u4efb\u52a1 CPU \u63a9\u7801\uff08\u9ed8\u8ba4\u540c\u4e3b\u6a21\u578b\u6279\u91cf\u63a9\u7801\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-cpu-strict-batch, --cpu-strict-batch-draft &lt;0|1&gt;<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6279\u91cf\u4e25\u683c CPU \u7ed1\u5b9a\uff08\u9ed8\u8ba4\u8349\u7a3f cpu-strict\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-prio-batch, --prio-batch-draft N<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6279\u91cf\u7ebf\u7a0b\u4f18\u5148\u7ea7\uff08\u9ed8\u8ba4 0\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-poll-batch, --poll-batch-draft &lt;0|1&gt;<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6279\u91cf\u8f6e\u8be2\u5f00\u5173\uff08\u9ed8\u8ba4\u540c\u8349\u7a3f poll\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-otd, --override-tensor-draft &lt;\u5f20\u91cf\u89c4\u5219&gt;=&lt;\u7f13\u5b58\u7c7b\u578b&gt;<\/code><\/p>\n<div><\/div>\n<p>\u5355\u72ec\u5f3a\u5236\u8349\u7a3f\u6a21\u578b\u5f20\u91cf\u7f13\u5b58\u7c7b\u578b<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-cmoed, --cpu-moe-draft<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u5168\u90e8 MoE \u5c42\u6743\u91cd\u5b58\u653e CPU \u5185\u5b58<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_CPU_MOE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ncmoed, --n-cpu-moe-draft N<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u524d N \u5c42 MoE \u6743\u91cd\u4fdd\u7559 CPU \u5185\u5b58<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_N_CPU_MOE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-n-max N<\/code><\/p>\n<div><\/div>\n<p>\u5355\u6b21\u63a8\u6d4b\u89e3\u7801\u8349\u7a3f\u6700\u5927 token \u6570\u91cf\uff08\u9ed8\u8ba4 3\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_N_MAX<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-n-min N<\/code><\/p>\n<div><\/div>\n<p>\u5355\u6b21\u63a8\u6d4b\u6700\u5c0f\u8349\u7a3f token \u6570\uff08\u9ed8\u8ba4 0\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_N_MIN<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-p-split, --draft-p-split P<\/code><\/p>\n<div><\/div>\n<p>\u63a8\u6d4b\u5206\u652f\u62c6\u5206\u6982\u7387\uff08\u9ed8\u8ba4 0.10\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_P_SPLIT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-p-min, --draft-p-min P<\/code><\/p>\n<div><\/div>\n<p>\u8d2a\u5fc3\u63a8\u6d4b\u6700\u5c0f\u6982\u7387\u9608\u503c\uff08\u9ed8\u8ba4 0.00\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_P_MIN<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-draft-backend-sampling, --no-spec-draft-backend-sampling<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u91c7\u6837\u5378\u8f7d\u81f3\u786c\u4ef6\u540e\u7aef\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_BACKEND_SAMPLING<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-devd, --device-draft &lt;dev1,dev2,..&gt;<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u786c\u4ef6\u5378\u8f7d\u8bbe\u5907\u5217\u8868\uff1bnone = \u4e0d\u5378\u8f7d\uff0c\u7528<code>--list-devices<\/code>\u67e5\u770b\u786c\u4ef6<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ngld, --gpu-layers-draft N<\/code><\/p>\n<div><\/div>\n<p>\u8349\u7a3f\u6a21\u578b\u52a0\u8f7d\u81f3\u663e\u5b58\u6700\u5927\u5c42\u6570\uff0c\u652f\u6301\u6570\u5b57 \/auto\/all\uff08\u9ed8\u8ba4 auto\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_N_GPU_LAYERS_DRAFT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-md, --model-draft FNAME<\/code><\/p>\n<div><\/div>\n<p>\u63a8\u6d4b\u89e3\u7801\u8349\u7a3f\u6a21\u578b\u672c\u5730\u6587\u4ef6\u8def\u5f84\uff08\u9ed8\u8ba4\u4e0d\u542f\u7528\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_DRAFT_MODEL<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-type \u7c7b\u578b\u5217\u8868<\/code><\/p>\n<div><\/div>\n<p>\u542f\u7528\u7684\u63a8\u6d4b\u89e3\u7801\u7b97\u6cd5\uff0c\u9017\u53f7\u5206\u9694\uff0c\u53ef\u9009\uff1a<\/p>\n<div><\/div>\n<p>none \/draft-simple\/draft-eagle3 \/draft-mtp\/draft-dflash \/draft-dspark\/ngram-simple \/ngram-map-k\/ngram-map-k4v \/ngram-mod\/ngram-cache<\/p>\n<div><\/div>\n<p>\u9ed8\u8ba4 none\uff08\u5173\u95ed\u63a8\u6d4b\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SPEC_TYPE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-mod-n-min N<\/code><\/p>\n<div><\/div>\n<p>Ngram-Mod \u63a8\u6d4b\u6700\u5c0f\u7f13\u5b58 token \u6570\u91cf\uff08\u9ed8\u8ba4 48\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-mod-n-max N<\/code><\/p>\n<div><\/div>\n<p>Ngram-Mod \u6700\u5927\u7f13\u5b58 token\uff08\u9ed8\u8ba4 64\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-mod-n-match N<\/code><\/p>\n<div><\/div>\n<p>Ngram-Mod \u5339\u914d\u7a97\u53e3\u957f\u5ea6\uff08\u9ed8\u8ba4 24\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-simple-size-n N<\/code><\/p>\n<div><\/div>\n<p>Ngram-Simple \u67e5\u8be2 n \u5143\u7ec4\u957f\u5ea6\uff08\u9ed8\u8ba4 12\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-simple-size-m N<\/code><\/p>\n<div><\/div>\n<p>Ngram-Simple \u8349\u7a3f m \u5143\u7ec4\u957f\u5ea6\uff08\u9ed8\u8ba4 48\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-simple-min-hits N<\/code><\/p>\n<div><\/div>\n<p>Ngram-Simple \u6700\u4f4e\u5339\u914d\u547d\u4e2d\u6b21\u6570\uff08\u9ed8\u8ba4 1\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-map-k-size-n N<\/code><\/p>\n<div><\/div>\n<p>Ngram-Map-K \u67e5\u8be2\u5143\u7ec4\u957f\u5ea6\uff08\u9ed8\u8ba4 12\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-map-k-size-m N<\/code><\/p>\n<div><\/div>\n<p>Ngram-Map-K \u8349\u7a3f\u5143\u7ec4\u957f\u5ea6\uff08\u9ed8\u8ba4 48\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-map-k-min-hits N<\/code><\/p>\n<div><\/div>\n<p>\u6700\u4f4e\u5339\u914d\u6b21\u6570\uff08\u9ed8\u8ba4 1\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-ngram-map-k4v-size-n \/ size-m \/ min-hits<\/code><\/p>\n<div><\/div>\n<p>\u56db\u5411\u5411\u91cf Ngram \u6620\u5c04\u53c2\u6570\uff0c\u9ed8\u8ba4\u540c\u4e0a<\/p>\n<\/div>\n<\/div>\n<div>\n<h3>\u5df2\u5e9f\u5f03\u53c2\u6570\u63d0\u793a<\/h3>\n<\/div>\n<div>\n<div>\n<p><code>--draft \/ --draft-min \/ --spec-ngram-size-n\/m\/min-hits<\/code><\/p>\n<div><\/div>\n<p>\u53c2\u6570\u5df2\u79fb\u9664\uff0c\u66ff\u6362\u4e3a\u5bf9\u5e94 spec-ngram \u7cfb\u5217\u53c2\u6570<\/p>\n<\/div>\n<\/div>\n<hr \/>\n<div>\n<h2>\u573a\u666f\u4e13\u7528\u53c2\u6570 example-specific params<\/h2>\n<\/div>\n<div>\n<div>\n<p><code>-lcs, --lookup-cache-static FNAME<\/code><\/p>\n<div><\/div>\n<p>\u9759\u6001\u67e5\u627e\u7f13\u5b58\u6587\u4ef6\uff0c\u68c0\u7d22\u89e3\u7801\u4f7f\u7528\uff0c\u751f\u6210\u8fc7\u7a0b\u4e0d\u66f4\u65b0\u7f13\u5b58<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-lcd, --lookup-cache-dynamic FNAME<\/code><\/p>\n<div><\/div>\n<p>\u52a8\u6001\u67e5\u627e\u7f13\u5b58\uff0c\u751f\u6210\u65f6\u81ea\u52a8\u66f4\u65b0<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ctxcp, --ctx-checkpoints, --swa-checkpoints N<\/code><\/p>\n<div><\/div>\n<p>\u5355\u4e2a\u4e0a\u4e0b\u6587\u69fd\u4f4d\u6700\u5927\u5feb\u7167\u6570\u91cf\uff08\u9ed8\u8ba4 32\uff09<\/p>\n<div><\/div>\n<p>\u53c2\u8003\uff1a<a title=\"autolink\" href=\"https:\/\/link.wtturl.cn\/?target=https%3A%2F%2Fgithub.com%2Fggml-org%2Fllama.cpp%2Fpull%2F15293&amp;scene=im&amp;aid=582478&amp;lang=zh\" target=\"_blank\" rel=\"noopener\">https:\/\/github.com\/ggml-org\/llama.cpp\/pull\/15293<\/a><\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CTX_CHECKPOINTS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-cms, --checkpoint-min-step N<\/code><\/p>\n<div><\/div>\n<p>\u4e0a\u4e0b\u6587\u5feb\u7167\u6700\u5c0f\u95f4\u9694 token \u6570\uff1b0 \u65e0\u9650\u5236\uff08\u9ed8\u8ba4 8192\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CHECKPOINT_MIN_SPACING_NT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-cram, --cache-ram N<\/code><\/p>\n<div><\/div>\n<p>\u7f13\u5b58\u6700\u5927\u5360\u7528\u5185\u5b58\uff08MiB\uff09\uff1b-1 \u65e0\u4e0a\u9650\u30010 \u5173\u95ed\u7f13\u5b58\uff08\u9ed8\u8ba4 8192\uff09<\/p>\n<div><\/div>\n<p>\u53c2\u8003\uff1a<a title=\"autolink\" href=\"https:\/\/link.wtturl.cn\/?target=https%3A%2F%2Fgithub.com%2Fggml-org%2Fllama.cpp%2Fpull%2F16391&amp;scene=im&amp;aid=582478&amp;lang=zh\" target=\"_blank\" rel=\"noopener\">https:\/\/github.com\/ggml-org\/llama.cpp\/pull\/16391<\/a><\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CACHE_RAM<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-kvu, --kv-unified \/ --no-kv-unified<\/code><\/p>\n<div><\/div>\n<p>\u5168\u5c40\u7edf\u4e00 KV \u7f13\u5b58\u7f13\u51b2\u533a\uff0c\u591a\u4f1a\u8bdd\u5171\u4eab\uff1b\u81ea\u52a8\u5e76\u884c\u69fd\u4f4d\u65f6\u9ed8\u8ba4\u5f00\u542f<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_KV_UNIFIED<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--cache-idle-slots, --no-cache-idle-slots<\/code><\/p>\n<div><\/div>\n<p>\u65b0\u4efb\u52a1\u5c06\u7a7a\u95f2\u4f1a\u8bdd\u5b58\u5165\u63d0\u793a\u7f13\u5b58\uff0c\u7edf\u4e00 KV \u65f6\u81ea\u52a8\u6e05\u7a7a\uff1b\u4f9d\u8d56 cache-ram\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CACHE_IDLE_SLOTS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--context-shift, --no-context-shift<\/code><\/p>\n<div><\/div>\n<p>\u65e0\u9650\u751f\u6210\u542f\u7528\u4e0a\u4e0b\u6587\u6ed1\u52a8\u7a97\u53e3\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CONTEXT_SHIFT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-r, --reverse-prompt PROMPT<\/code><\/p>\n<div><\/div>\n<p>\u4ea4\u4e92\u5f0f\u6a21\u5f0f\uff0c\u68c0\u6d4b\u5230\u6307\u5b9a\u5b57\u7b26\u4e32\u5373\u505c\u6b62\u751f\u6210\u5e76\u4ea4\u8fd8\u63a7\u5236\u53f0<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-sp, --special<\/code><\/p>\n<div><\/div>\n<p>\u8f93\u51fa\u7279\u6b8a token \u6807\u8bc6\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--warmup, --no-warmup<\/code><\/p>\n<div><\/div>\n<p>\u7a0b\u5e8f\u542f\u52a8\u7a7a\u8dd1\u9884\u70ed\u6a21\u578b\u8ba1\u7b97\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spm-infill<\/code><\/p>\n<div><\/div>\n<p>\u586b\u7a7a\u4efb\u52a1\u4f7f\u7528 Suffix\/Prefix\/Middle \u987a\u5e8f\uff0c\u66ff\u4ee3\u9ed8\u8ba4 Prefix\/Suffix\/Middle\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--pooling {none,mean,cls,last,rank}<\/code><\/p>\n<div><\/div>\n<p>\u5d4c\u5165\u5411\u91cf\u6c60\u5316\u7b56\u7565\uff1b\u672a\u6307\u5b9a\u4f7f\u7528\u6a21\u578b\u9ed8\u8ba4\u914d\u7f6e<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_POOLING<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-np, --parallel N<\/code><\/p>\n<div><\/div>\n<p>\u670d\u52a1\u7aef\u5e76\u53d1\u4f1a\u8bdd\u69fd\u4f4d\u6570\u91cf\uff1b-1 \u81ea\u52a8\u9002\u914d\uff08\u9ed8\u8ba4 &#8211; 1\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_N_PARALLEL<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-cb, --cont-batching \/ -nocb, --no-cont-batching<\/code><\/p>\n<div><\/div>\n<p>\u8fde\u7eed\u52a8\u6001\u6279\u91cf\u8c03\u5ea6\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CONT_BATCHING<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-mm, --mmproj FILE<\/code><\/p>\n<div><\/div>\n<p>\u591a\u6a21\u6001\u89c6\u89c9\u6295\u5f71\u5668\u6587\u4ef6\u8def\u5f84\uff1b\u4f7f\u7528<code>-hf<\/code>\u62c9\u53d6\u6a21\u578b\u53ef\u81ea\u52a8\u4e0b\u8f7d\u65e0\u9700\u624b\u52a8\u6307\u5b9a<\/p>\n<div><\/div>\n<p>\u6587\u6863\uff1atools\/mtmd\/README.md<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MMPROJ<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-mmu, --mmproj-url URL<\/code><\/p>\n<div><\/div>\n<p>\u5728\u7ebf\u89c6\u89c9\u6295\u5f71\u5668\u4e0b\u8f7d\u5730\u5740<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MMPROJ_URL<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mmproj-auto, --no-mmproj-auto<\/code><\/p>\n<div><\/div>\n<p>\u81ea\u52a8\u52a0\u8f7d\u914d\u5957\u89c6\u89c9\u6295\u5f71\u6587\u4ef6\uff08HF \u6a21\u5f0f\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MMPROJ_AUTO<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mmproj-offload, --no-mmproj-offload<\/code><\/p>\n<div><\/div>\n<p>\u89c6\u89c9\u6295\u5f71\u5668 GPU \u786c\u4ef6\u5378\u8f7d\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MMPROJ_OFFLOAD<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--image-min-tokens N<\/code><\/p>\n<div><\/div>\n<p>\u52a8\u6001\u5206\u8fa8\u7387\u89c6\u89c9\u6a21\u578b\uff0c\u5355\u5f20\u56fe\u7247\u6700\u5c0f\u7f16\u7801 token\uff08\u9ed8\u8ba4\u8bfb\u53d6\u6a21\u578b\u914d\u7f6e\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_IMAGE_MIN_TOKENS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--image-max-tokens N<\/code><\/p>\n<div><\/div>\n<p>\u5355\u5f20\u56fe\u7247\u6700\u5927\u7f16\u7801 token \u4e0a\u9650\uff08\u9ed8\u8ba4\u8bfb\u53d6\u6a21\u578b\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_IMAGE_MAX_TOKENS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mtmd-batch-max-tokens N<\/code><\/p>\n<div><\/div>\n<p>\u56fe\u7247\u7f16\u7801\u6279\u91cf\u603b token \u4e0a\u9650\uff08\u9ed8\u8ba4 1024\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MTMD_BATCH_MAX_TOKENS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-a, --alias STRING<\/code><\/p>\n<div><\/div>\n<p>\u6a21\u578b API \u522b\u540d\uff0c\u591a\u522b\u540d\u9017\u53f7\u5206\u9694<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_ALIAS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--tags STRING<\/code><\/p>\n<div><\/div>\n<p>\u6a21\u578b\u6807\u7b7e\uff0c\u4ec5\u7528\u4e8e\u5c55\u793a\u4e0d\u53c2\u4e0e\u8def\u7531\u5339\u914d\uff0c\u9017\u53f7\u5206\u9694<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_TAGS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--embd-normalize N<\/code><\/p>\n<div><\/div>\n<p>\u5d4c\u5165\u5411\u91cf\u5f52\u4e00\u5316\u65b9\u5f0f<\/p>\n<div><\/div>\n<p>-1 = \u5173\u95ed\u5f52\u4e00\u5316<\/p>\n<div><\/div>\n<p>0 = \u6700\u5927\u7edd\u5bf9\u503c int16 \u7f29\u653e<\/p>\n<div><\/div>\n<p>1 = \u66fc\u54c8\u987f\u8ddd\u79bb\u5f52\u4e00\u5316<\/p>\n<div><\/div>\n<p>2 = \u6b27\u51e0\u91cc\u5f97\u5f52\u4e00\u5316\uff08\u9ed8\u8ba4\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<blockquote>\n<div>\n<div>2 = \u81ea\u5b9a\u4e49 P \u8303\u6570<\/div>\n<\/div>\n<\/blockquote>\n<\/div>\n<div>\n<div>\n<p><code>--host HOST<\/code><\/p>\n<div><\/div>\n<p>\u670d\u52a1\u76d1\u542c\u5730\u5740\uff1b\u4ee5.sock \u7ed3\u5c3e\u5219\u7ed1\u5b9a UNIX \u672c\u5730\u5957\u63a5\u5b57\uff08\u9ed8\u8ba4 127.0.0.1\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_HOST<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--port PORT<\/code><\/p>\n<div><\/div>\n<p>\u670d\u52a1\u76d1\u542c\u7aef\u53e3\uff08\u9ed8\u8ba4 8080\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_PORT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--reuse-port<\/code><\/p>\n<div><\/div>\n<p>\u7aef\u53e3\u590d\u7528\uff0c\u5141\u8bb8\u591a\u5957\u63a5\u5b57\u7ed1\u5b9a\u540c\u4e00\u7aef\u53e3\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_REUSE_PORT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--path PATH<\/code><\/p>\n<div><\/div>\n<p>\u9759\u6001\u8d44\u6e90\u6587\u4ef6\u6258\u7ba1\u76ee\u5f55\uff08\u9ed8\u8ba4\u7a7a\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_STATIC_PATH<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--cors-origins ORIGINS<\/code><\/p>\n<div><\/div>\n<p>\u8de8\u57df\u5141\u8bb8\u6765\u6e90\uff0c\u9017\u53f7\u5206\u9694\uff1b* \u5141\u8bb8\u5168\u90e8\uff1b\u8bbe\u4e3a<a title=\"autolink\" href=\"https:\/\/link.wtturl.cn\/?target=https%3A%2F%2Flocalhost&amp;scene=im&amp;aid=582478&amp;lang=zh\" target=\"_blank\" rel=\"noopener\">localhost<\/a>\u4ec5\u653e\u884c\u672c\u673a<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CORS_ORIGINS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--cors-methods METHODS<\/code><\/p>\n<div><\/div>\n<p>\u8de8\u57df\u5141\u8bb8 HTTP \u8bf7\u6c42\u65b9\u6cd5\uff08\u9ed8\u8ba4 GET,POST,DELETE,OPTIONS\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CORS_METHODS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--cors-headers HEADERS<\/code><\/p>\n<div><\/div>\n<p>\u8de8\u57df\u5141\u8bb8\u8bf7\u6c42\u5934\uff0c<em>\u653e\u884c\u5168\u90e8\uff08\u9ed8\u8ba4<\/em>\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CORS_HEADERS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--cors-credentials, --no-cors-credentials<\/code><\/p>\n<div><\/div>\n<p>\u5141\u8bb8\u8de8\u57df\u643a\u5e26 Cookie \u51ed\u8bc1\uff1b\u9ed8\u8ba4\u5f00\u542f\uff0corigins \u4e3a * \u65f6\u4f1a\u539f\u6837\u56de\u4f20 Origin \u5934<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CORS_CREDENTIALS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--api-prefix PREFIX<\/code><\/p>\n<div><\/div>\n<p>API \u63a5\u53e3\u5168\u5c40\u8def\u5f84\u524d\u7f00\uff0c\u672b\u5c3e\u4e0d\u5e26\u659c\u6760\uff08\u9ed8\u8ba4\u7a7a\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_API_PREFIX<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--ui-config, --webui-config JSON<\/code><\/p>\n<div><\/div>\n<p>WebUI \u754c\u9762\u9ed8\u8ba4\u914d\u7f6e JSON \u5b57\u7b26\u4e32\uff0c\u8986\u76d6\u5185\u7f6e\u9ed8\u8ba4\u503c<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_UI_CONFIG<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--ui-config-file, --webui-config-file PATH<\/code><\/p>\n<div><\/div>\n<p>\u5916\u90e8 JSON \u6587\u4ef6\u914d\u7f6e WebUI \u53c2\u6570<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_UI_CONFIG_FILE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--ui-mcp-proxy, --no-ui-mcp-proxy<\/code><\/p>\n<div><\/div>\n<p>MCP \u534f\u8bae\u8de8\u57df\u4ee3\u7406\uff08\u5b9e\u9a8c\u529f\u80fd\uff0c\u4e0d\u53ef\u5728\u4e0d\u53ef\u4fe1\u516c\u7f51\u5f00\u542f\uff0c\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_UI_MCP_PROXY<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--tools TOOL1,TOOL2,...<\/code><\/p>\n<div><\/div>\n<p>\u542f\u7528\u5185\u7f6e AI \u5de5\u5177\u4ee3\u7406\uff08\u5b9e\u9a8c\uff0c\u516c\u7f51\u52ff\u5f00\uff09\uff1b\u4f20\u5165 all \u542f\u7528\u5168\u90e8<\/p>\n<div><\/div>\n<p>\u53ef\u7528\u5de5\u5177\uff1aread_file\u3001file_glob_search\u3001grep_search\u3001exec_shell_command\u3001write_file\u3001edit_file\u3001get_datetime<\/p>\n<div><\/div>\n<p>\u5b89\u5168\u9650\u5236\uff1a\u81ea\u52a8\u9650\u5236\u8de8\u57df\u4ec5<a title=\"autolink\" href=\"https:\/\/link.wtturl.cn\/?target=https%3A%2F%2Flocalhost&amp;scene=im&amp;aid=582478&amp;lang=zh\" target=\"_blank\" rel=\"noopener\">localhost<\/a>\u8bbf\u95ee<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_TOOLS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mcp-servers-config PATH<\/code><\/p>\n<div><\/div>\n<p>MCP \u670d\u52a1\u5b9a\u4e49 JSON \u6587\u4ef6\uff08Cursor \u517c\u5bb9\u683c\u5f0f\uff0c\u4e0d\u53ef\u516c\u7f51\u90e8\u7f72\uff0c\u9ed8\u8ba4\u65e0\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MCP_SERVERS_CONFIG<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--mcp-servers-json JSON<\/code><\/p>\n<div><\/div>\n<p>\u884c\u5185 JSON \u5b9a\u4e49 MCP \u670d\u52a1\u914d\u7f6e\uff08\u540c\u4e0a\u5b89\u5168\u9650\u5236\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MCP_SERVERS_JSON<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-ag, --agent, --no-agent<\/code><\/p>\n<div><\/div>\n<p>\u4e00\u952e\u5f00\u542f MCP \u4ee3\u7406 + \u5168\u90e8\u5185\u7f6e\u5de5\u5177\uff1b\u516c\u7f51\u7981\u7528\uff0c\u81ea\u52a8\u9650\u5236\u8de8\u57df<a title=\"autolink\" href=\"https:\/\/link.wtturl.cn\/?target=https%3A%2F%2Flocalhost&amp;scene=im&amp;aid=582478&amp;lang=zh\" target=\"_blank\" rel=\"noopener\">localhost<\/a>\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_AGENT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--ui, --webui, --no-ui<\/code><\/p>\n<div><\/div>\n<p>\u5f00\u542f\u7f51\u9875\u53ef\u89c6\u5316\u754c\u9762\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_UI<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--embedding, --embeddings<\/code><\/p>\n<div><\/div>\n<p>\u4ec5\u542f\u7528\u5d4c\u5165\u5411\u91cf\u63a5\u53e3\uff0c\u4ec5\u642d\u914d\u4e13\u7528 Embedding \u6a21\u578b\u4f7f\u7528\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_EMBEDDINGS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--rerank, --reranking<\/code><\/p>\n<div><\/div>\n<p>\u542f\u7528\u91cd\u6392\u5e8f Rerank \u63a5\u53e3\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_RERANKING<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--api-key KEY<\/code><\/p>\n<div><\/div>\n<p>\u63a5\u53e3\u9274\u6743\u5bc6\u94a5\uff0c\u591a\u5bc6\u94a5\u9017\u53f7\u5206\u9694\uff08\u9ed8\u8ba4\u65e0\u5bc6\u94a5\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_API_KEY<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--api-key-file FNAME<\/code><\/p>\n<div><\/div>\n<p>\u5bc6\u94a5\u6587\u4ef6\u8def\u5f84\uff0c\u6bcf\u884c\u4e00\u4e2a\u5bc6\u94a5\uff1b# \u5f00\u5934\u884c\u4e3a\u6ce8\u91ca\uff08\u9ed8\u8ba4\u65e0\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_API_KEY_FILE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--ssl-key-file FNAME<\/code><\/p>\n<div><\/div>\n<p>PEM \u683c\u5f0f SSL \u79c1\u94a5\u6587\u4ef6\u8def\u5f84<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SSL_KEY_FILE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--ssl-cert-file FNAME<\/code><\/p>\n<div><\/div>\n<p>PEM \u683c\u5f0f SSL \u8bc1\u4e66\u6587\u4ef6\u8def\u5f84<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SSL_CERT_FILE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--chat-template-kwargs STRING<\/code><\/p>\n<div><\/div>\n<p>\u804a\u5929\u6a21\u677f\u89e3\u6790\u9644\u52a0\u53c2\u6570\uff0c\u5408\u6cd5 JSON \u5bf9\u8c61\u5b57\u7b26\u4e32<\/p>\n<div><\/div>\n<p>\u793a\u4f8b\uff1a<code>{\"key1\":\"value1\"}<\/code><\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CHAT_TEMPLATE_KWARGS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-to, --timeout N<\/code><\/p>\n<div><\/div>\n<p>\u670d\u52a1\u8bfb\u5199\u8d85\u65f6\u79d2\u6570\uff08\u9ed8\u8ba4 3600\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_TIMEOUT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--sse-ping-interval N<\/code><\/p>\n<div><\/div>\n<p>SSE \u6d41\u5f0f\u5fc3\u8df3\u95f4\u9694\u79d2\uff1b-1 \u5173\u95ed\u5fc3\u8df3\uff08\u9ed8\u8ba4 30\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SSE_PING_INTERVAL<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--threads-http N<\/code><\/p>\n<div><\/div>\n<p>HTTP \u8bf7\u6c42\u5904\u7406\u7ebf\u7a0b\u6570\uff1b-1 \u81ea\u52a8\u5206\u914d\uff08\u9ed8\u8ba4 &#8211; 1\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_THREADS_HTTP<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--cache-prompt, --no-cache-prompt<\/code><\/p>\n<div><\/div>\n<p>\u542f\u7528\u63d0\u793a\u8bcd KV \u7f13\u5b58\u590d\u7528\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CACHE_PROMPT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--cache-reuse N<\/code><\/p>\n<div><\/div>\n<p>KV \u6ed1\u52a8\u590d\u7528\u6700\u5c0f\u5757\u5c3a\u5bf8\uff0c\u4f9d\u8d56 prompt \u7f13\u5b58\uff08\u9ed8\u8ba4 0\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CACHE_REUSE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--metrics<\/code><\/p>\n<div><\/div>\n<p>\u5f00\u542f Prometheus \u76d1\u63a7\u6307\u6807\u63a5\u53e3\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_ENDPOINT_METRICS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--props<\/code><\/p>\n<div><\/div>\n<p>\u5f00\u653e POST \/props \u63a5\u53e3\u52a8\u6001\u4fee\u6539\u5168\u5c40\u53c2\u6570\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_ENDPOINT_PROPS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--slots, --no-slots<\/code><\/p>\n<div><\/div>\n<p>\u5f00\u653e\u4f1a\u8bdd\u69fd\u76d1\u63a7\u63a5\u53e3\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_ENDPOINT_SLOTS<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--slot-save-path PATH<\/code><\/p>\n<div><\/div>\n<p>\u4f1a\u8bdd KV \u7f13\u5b58\u6301\u4e45\u5316\u4fdd\u5b58\u76ee\u5f55\uff08\u9ed8\u8ba4\u5173\u95ed\u6301\u4e45\u5316\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--media-path PATH<\/code><\/p>\n<div><\/div>\n<p>\u672c\u5730\u5a92\u4f53\u6587\u4ef6\u52a0\u8f7d\u76ee\u5f55\uff0c\u652f\u6301 file:\/\/ \u76f8\u5bf9\u8def\u5f84\u8bbf\u95ee\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--models-dir PATH<\/code><\/p>\n<div><\/div>\n<p>\u8def\u7531\u670d\u52a1\u6a21\u578b\u5b58\u653e\u76ee\u5f55\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MODELS_DIR<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--models-preset PATH<\/code><\/p>\n<div><\/div>\n<p>\u8def\u7531\u6a21\u578b\u9884\u8bbe INI \u914d\u7f6e\u6587\u4ef6\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MODELS_PRESET<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--models-max N<\/code><\/p>\n<div><\/div>\n<p>\u8def\u7531\u670d\u52a1\u540c\u65f6\u52a0\u8f7d\u6a21\u578b\u4e0a\u9650\uff1b0 \u65e0\u9650\u5236\uff08\u9ed8\u8ba4 4\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MODELS_MAX<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--models-autoload, --no-models-autoload<\/code><\/p>\n<div><\/div>\n<p>\u8def\u7531\u670d\u52a1\u81ea\u52a8\u52a0\u8f7d\u76ee\u5f55\u5185\u6a21\u578b\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_MODELS_AUTOLOAD<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--jinja, --no-jinja<\/code><\/p>\n<div><\/div>\n<p>\u4f7f\u7528 Jinja2 \u5f15\u64ce\u89e3\u6790\u804a\u5929\u6a21\u677f\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_JINJA<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--reasoning-format FORMAT<\/code><\/p>\n<div><\/div>\n<p>\u601d\u8003\u63a8\u7406\u6807\u7b7e\u89e3\u6790\u683c\u5f0f<\/p>\n<\/div>\n<\/div>\n<div>\n<ul>\n<li>none\uff1a\u601d\u8003\u5185\u5bb9\u4fdd\u7559\u5728 message.content \u4e0d\u62c6\u5206<\/li>\n<li>deepseek\uff1a\u601d\u8003\u5185\u5bb9\u5b58\u5165\u72ec\u7acb message.reasoning_content \u5b57\u6bb5<\/li>\n<li>deepseek-legacy\uff1a\u4fdd\u7559\u6807\u7b7e\u540c\u65f6\u62c6\u5206 reasoning_content\n<div><\/div>\n<p>\u9ed8\u8ba4 auto \u81ea\u52a8\u8bc6\u522b\u6a21\u677f<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_THINK<\/code><\/li>\n<\/ul>\n<\/div>\n<div>\n<div>\n<p><code>-rea, --reasoning [on|off|auto]<\/code><\/p>\n<div><\/div>\n<p>\u63a8\u7406\u601d\u8003\u6a21\u5f0f\u5f00\u5173\uff1bauto \u81ea\u52a8\u8bfb\u53d6\u6a21\u677f\u914d\u7f6e\uff08\u9ed8\u8ba4 auto\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_REASONING<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--reasoning-budget N<\/code><\/p>\n<div><\/div>\n<p>\u601d\u8003\u9636\u6bb5 token \u9884\u7b97\uff1b-1 \u65e0\u4e0a\u9650\u30010 \u76f4\u63a5\u7ed3\u675f\u601d\u8003\u3001\u6b63\u6570\u9650\u5236\u601d\u8003\u957f\u5ea6\uff08\u9ed8\u8ba4 &#8211; 1\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_THINK_BUDGET<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--reasoning-budget-message MESSAGE<\/code><\/p>\n<div><\/div>\n<p>\u601d\u8003 token \u8017\u5c3d\u65f6\u63d2\u5165\u7684\u63d0\u793a\u6587\u672c\uff08\u9ed8\u8ba4\u7a7a\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_THINK_BUDGET_MESSAGE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--reasoning-preserve, --no-reasoning-preserve<\/code><\/p>\n<div><\/div>\n<p>\u5b8c\u6574\u5bf9\u8bdd\u5386\u53f2\u6c38\u4e45\u4fdd\u5b58\u63a8\u7406\u5185\u5bb9\uff0c\u4e0d\u53ea\u4fdd\u5b58\u5728\u6700\u540e\u4e00\u8f6e\u52a9\u624b\u56de\u590d\uff1b\u9075\u5faa\u6a21\u578b\u6a21\u677f\u80fd\u529b\u6807\u8bc6<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_REASONING_PRESERVE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--chat-template JINJA_TEMPLATE<\/code><\/p>\n<div><\/div>\n<p>\u5185\u7f6e Jinja \u804a\u5929\u6a21\u677f\u540d\u79f0\uff0c\u8986\u76d6\u6a21\u578b\u81ea\u5e26\u6a21\u677f\uff1b\u6307\u5b9a\u524d\u7f00 \/ \u540e\u7f00\u4f1a\u7981\u7528\u6a21\u677f<\/p>\n<div><\/div>\n<p>\u5185\u7f6e\u6a21\u677f\u5217\u8868\uff1abailing\u3001bailing-think\u3001bailing2\u3001chatglm3\u3001chatglm4\u3001chatml\u3001command-r\u3001deepseek\u3001deepseek-ocr\u3001deepseek2\u3001deepseek3\u3001exaone-moe\u3001exaone3\u3001exaone4\u3001falcon3\u3001gemma\u3001gigachat\u3001glmedge\u3001gpt-oss\u3001granite\u3001granite-4.0\u3001granite-4.1\u3001grok-2\u3001hunyuan-dense\u3001hunyuan-moe\u3001hunyuan-vl\u3001kimi-k2\u3001llama2\u3001llama2-sys\u3001llama2-sys-bos\u3001llama2-sys-strip\u3001llama3\u3001llama4\u3001megrez\u3001minicpm\u3001mistral-v1\u3001mistral-v3\u3001mistral-v3-tekken\u3001mistral-v7\u3001mistral-v7-tekken\u3001monarch\u3001openchat\u3001orion\u3001pangu-embedded\u3001phi3\u3001phi4\u3001rwkv-world\u3001seed_oss\u3001smolvlm\u3001solar-open\u3001vicuna\u3001vicuna-orca\u3001yandex\u3001zephyr<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CHAT_TEMPLATE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--chat-template-file JINJA_TEMPLATE_FILE<\/code><\/p>\n<div><\/div>\n<p>\u5916\u90e8\u6587\u4ef6\u81ea\u5b9a\u4e49 Jinja \u804a\u5929\u6a21\u677f\uff0c\u89c4\u5219\u540c\u4e0a\u5185\u7f6e\u6a21\u677f<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_CHAT_TEMPLATE_FILE<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--skip-chat-parsing, --no-skip-chat-parsing<\/code><\/p>\n<div><\/div>\n<p>\u5f3a\u5236\u7eaf\u6587\u672c\u89e3\u6790\uff0c\u5ffd\u7565 Jinja \u6a21\u677f\uff1b\u6240\u6709\u8f93\u51fa\uff08\u63a8\u7406\u3001\u5de5\u5177\u8c03\u7528\uff09\u5168\u90e8\u653e\u5165 content \u5b57\u6bb5\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_SKIP_CHAT_PARSING<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--prefill-assistant, --no-prefill-assistant<\/code><\/p>\n<div><\/div>\n<p>\u6700\u540e\u4e00\u6761\u6d88\u606f\u4e3a\u52a9\u624b\u56de\u590d\u65f6\u81ea\u52a8\u7eed\u5199\uff1b\u5173\u95ed\u5219\u89c6\u4e3a\u5b8c\u6574\u6d88\u606f\u4e0d\u518d\u9884\u586b\u5145\uff08\u9ed8\u8ba4\u5f00\u542f\uff09<\/p>\n<div><\/div>\n<p>\u73af\u5883\u53d8\u91cf\uff1a<code>LLAMA_ARG_PREFILL_ASSISTANT<\/code><\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-sps, --slot-prompt-similarity SIMILARITY<\/code><\/p>\n<div><\/div>\n<p>\u4f1a\u8bdd\u590d\u7528\u63d0\u793a\u8bcd\u76f8\u4f3c\u5ea6\u9608\u503c\uff1b0 \u5173\u95ed\u590d\u7528\uff08\u9ed8\u8ba4 0.10\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--lora-init-without-apply<\/code><\/p>\n<div><\/div>\n<p>\u52a0\u8f7d LoRA \u9002\u914d\u5668\u4f46\u4e0d\u7acb\u5373\u751f\u6548\uff0c\u53ef\u901a\u8fc7 POST \/lora-adapters \u63a5\u53e3\u52a8\u6001\u5207\u6362\uff08\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--sleep-idle-seconds SECONDS<\/code><\/p>\n<div><\/div>\n<p>\u7a7a\u95f2\u8d85\u65f6\u4f11\u7720\u79d2\u6570\uff1b-1 \u7981\u7528\u4f11\u7720\uff08\u9ed8\u8ba4 &#8211; 1\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--log-prompts-dir PATH<\/code><\/p>\n<div><\/div>\n<p>\u63d0\u793a\u8bcd\u65e5\u5fd7\u4fdd\u5b58\u76ee\u5f55\uff08\u81ea\u52a8\u521b\u5efa\uff0c\u4ec5\u8c03\u8bd5\u7528\uff0c\u9ed8\u8ba4\u5173\u95ed\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>-mv, --model-vocoder FNAME<\/code><\/p>\n<div><\/div>\n<p>\u97f3\u9891 TTS \u58f0\u7801\u5668\u6a21\u578b\u6587\u4ef6\uff08\u9ed8\u8ba4\u4e0d\u542f\u7528\uff09<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--tts-use-guide-tokens<\/code><\/p>\n<div><\/div>\n<p>TTS \u8bed\u97f3\u751f\u6210\u542f\u7528\u5f15\u5bfc token \u63d0\u5347\u8bcd\u8bed\u8fd8\u539f\u5ea6<\/p>\n<\/div>\n<\/div>\n<div>\n<h3>\u5185\u7f6e\u4e13\u7528\u6a21\u578b\u5feb\u6377\u53c2\u6570\uff08\u4e00\u952e\u52a0\u8f7d\uff09<\/h3>\n<\/div>\n<div>\n<div>\n<p><code>--embd-gemma-default<\/code><\/p>\n<div><\/div>\n<p>\u81ea\u52a8\u4e0b\u8f7d Gemma \u5d4c\u5165\u6a21\u578b\u6743\u91cd<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--fim-qwen-1.5b-default<\/code><\/p>\n<div><\/div>\n<p>Qwen2.5-Coder 1.5B \u4ee3\u7801\u8865\u5168\u6a21\u578b<\/p>\n<div><\/div>\n<p><code>--fim-qwen-3b-default<\/code> Qwen2.5-Coder 3B<\/p>\n<div><\/div>\n<p><code>--fim-qwen-7b-default<\/code> Qwen2.5-Coder 7B<\/p>\n<div><\/div>\n<p><code>--fim-qwen-7b-spec<\/code> Qwen7B \u4e3b\u6a21\u578b + 0.5B \u8349\u7a3f\u63a8\u6d4b\u89e3\u7801<\/p>\n<div><\/div>\n<p><code>--fim-qwen-14b-spec<\/code> Qwen14B+0.5B \u8349\u7a3f<\/p>\n<div><\/div>\n<p><code>--fim-qwen-30b-default<\/code> Qwen3 Coder 30B A3B \u6307\u4ee4\u6a21\u578b<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--gpt-oss-20b-default<\/code> \/ <code>--gpt-oss-120b-default<\/code><\/p>\n<div><\/div>\n<p>\u81ea\u52a8\u4e0b\u8f7d\u5bf9\u5e94 GPT-OSS \u5f00\u6e90\u6a21\u578b\u6743\u91cd<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--vision-gemma-4b-default<\/code> \/ <code>--vision-gemma-12b-default<\/code><\/p>\n<div><\/div>\n<p>Gemma3 4B\/12B \u89c6\u89c9 QAT \u91cf\u5316\u6a21\u578b<\/p>\n<\/div>\n<\/div>\n<div>\n<div>\n<p><code>--spec-default<\/code><\/p>\n<div><\/div>\n<p>\u542f\u7528\u9ed8\u8ba4\u63a8\u6d4b\u89e3\u7801\u914d\u7f6e<br \/>\n\u9644\uff1a<\/p>\n<p><a href=\"https:\/\/www.o6s.net\/wp-content\/uploads\/2026\/07\/llamaenveditor.zip\">llamaenveditor<\/a>\u00a0\u7a0b\u5e8f\u53ca\u6e90\u4ee3\u7801<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-2478\" src=\"https:\/\/www.o6s.net\/wp-content\/uploads\/2026\/07\/llama.cppWindows\u73af\u5883\u53d8\u91cf\u7f16\u8f91\u5668.png\" alt=\"\" width=\"876\" height=\"533\" srcset=\"https:\/\/www.o6s.net\/wp-content\/uploads\/2026\/07\/llama.cppWindows\u73af\u5883\u53d8\u91cf\u7f16\u8f91\u5668.png 876w, https:\/\/www.o6s.net\/wp-content\/uploads\/2026\/07\/llama.cppWindows\u73af\u5883\u53d8\u91cf\u7f16\u8f91\u5668-300x183.png 300w, https:\/\/www.o6s.net\/wp-content\/uploads\/2026\/07\/llama.cppWindows\u73af\u5883\u53d8\u91cf\u7f16\u8f91\u5668-768x467.png 768w\" sizes=\"auto, (max-width: 876px) 100vw, 876px\" \/><\/p>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u6ce8\uff1a\u6587\u540e\u9644llama.cpp\u73af\u5883\u53d8\u91cf\u7f16\u8f91\u5668\u53ca\u6e90\u7801 \u901a\u7528\u53c2\u6570 common params -h, &#8211;help, [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[21],"tags":[],"class_list":["post-2475","post","type-post","status-publish","format-standard","hentry","category-21"],"_links":{"self":[{"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/posts\/2475","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/comments?post=2475"}],"version-history":[{"count":3,"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/posts\/2475\/revisions"}],"predecessor-version":[{"id":2480,"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/posts\/2475\/revisions\/2480"}],"wp:attachment":[{"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/media?parent=2475"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/categories?post=2475"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.o6s.net\/index.php\/wp-json\/wp\/v2\/tags?post=2475"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}