--- /srv/rebuilderd/tmp/rebuilderdCOe5C2/inputs/llama.cpp-examples_8941+dfsg-1_arm64.deb +++ /srv/rebuilderd/tmp/rebuilderdCOe5C2/out/llama.cpp-examples_8941+dfsg-1_arm64.deb ├── file list │ @@ -1,3 +1,3 @@ │ -rw-r--r-- 0 0 0 4 2026-04-26 21:45:18.000000 debian-binary │ -rw-r--r-- 0 0 0 2396 2026-04-26 21:45:18.000000 control.tar.xz │ --rw-r--r-- 0 0 0 318532 2026-04-26 21:45:18.000000 data.tar.xz │ +-rw-r--r-- 0 0 0 319864 2026-04-26 21:45:18.000000 data.tar.xz ├── control.tar.xz │ ├── control.tar │ │ ├── ./md5sums │ │ │ ├── ./md5sums │ │ │ │┄ Files differ ├── data.tar.xz │ ├── data.tar │ │ ├── file list │ │ │ @@ -30,29 +30,29 @@ │ │ │ -rw-r--r-- 0 root (0) root (0) 3442 2026-04-26 21:45:18.000000 ./usr/share/doc/llama.cpp-examples/changelog.Debian.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 13361 2026-04-26 21:45:18.000000 ./usr/share/doc/llama.cpp-examples/copyright │ │ │ drwxr-xr-x 0 root (0) root (0) 0 2026-04-26 21:45:18.000000 ./usr/share/lintian/ │ │ │ drwxr-xr-x 0 root (0) root (0) 0 2026-04-26 21:45:18.000000 ./usr/share/lintian/overrides/ │ │ │ -rw-r--r-- 0 root (0) root (0) 177 2026-04-26 21:45:18.000000 ./usr/share/lintian/overrides/llama.cpp-examples │ │ │ drwxr-xr-x 0 root (0) root (0) 0 2026-04-26 21:45:18.000000 ./usr/share/man/ │ │ │ drwxr-xr-x 0 root (0) root (0) 0 2026-04-26 21:45:18.000000 ./usr/share/man/man1/ │ │ │ --rw-r--r-- 0 root (0) root (0) 6224 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-batched.1.gz │ │ │ +-rw-r--r-- 0 root (0) root (0) 6228 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-batched.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 6557 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-debug.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 6438 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-diffusion-cli.1.gz │ │ │ --rw-r--r-- 0 root (0) root (0) 6544 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-embedding.1.gz │ │ │ --rw-r--r-- 0 root (0) root (0) 6054 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-eval-callback.1.gz │ │ │ +-rw-r--r-- 0 root (0) root (0) 6541 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-embedding.1.gz │ │ │ +-rw-r--r-- 0 root (0) root (0) 6059 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-eval-callback.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 6436 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-finetune.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 398 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-gguf-hash.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 251 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-gguf.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 6093 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-idle.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 6051 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-lookahead.1.gz │ │ │ --rw-r--r-- 0 root (0) root (0) 6305 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-lookup-create.1.gz │ │ │ +-rw-r--r-- 0 root (0) root (0) 6307 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-lookup-create.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 274 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-lookup-merge.1.gz │ │ │ --rw-r--r-- 0 root (0) root (0) 6303 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-lookup-stats.1.gz │ │ │ +-rw-r--r-- 0 root (0) root (0) 6308 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-lookup-stats.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 6293 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-lookup.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 6248 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-parallel.1.gz │ │ │ --rw-r--r-- 0 root (0) root (0) 6246 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-passkey.1.gz │ │ │ --rw-r--r-- 0 root (0) root (0) 6400 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-retrieval.1.gz │ │ │ +-rw-r--r-- 0 root (0) root (0) 6238 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-passkey.1.gz │ │ │ +-rw-r--r-- 0 root (0) root (0) 6403 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-retrieval.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 6063 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-save-load-state.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 268 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-simple-chat.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 259 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-simple.1.gz │ │ │ -rw-r--r-- 0 root (0) root (0) 6737 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-speculative-simple.1.gz │ │ │ --rw-r--r-- 0 root (0) root (0) 6727 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-speculative.1.gz │ │ │ +-rw-r--r-- 0 root (0) root (0) 6720 2026-04-26 21:45:18.000000 ./usr/share/man/man1/llama-speculative.1.gz │ │ ├── ./usr/share/man/man1/llama-batched.1.gz │ │ │ ├── llama-batched.1 │ │ │ │ @@ -1,13 +1,12 @@ │ │ │ │ .\" DO NOT MODIFY THIS FILE! It was generated by help2man 1.49.3. │ │ │ │ .TH LLAMA-BATCHED "1" "April 2026" "debian" "User Commands" │ │ │ │ .SH NAME │ │ │ │ llama-batched \- llama-batched │ │ │ │ .SH DESCRIPTION │ │ │ │ -load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ \fB\-\-\-\-\-\fR common params \fB\-\-\-\-\-\fR │ │ │ │ .PP │ │ │ │ \fB\-h\fR, \fB\-\-help\fR, \fB\-\-usage\fR print usage and exit │ │ │ │ \fB\-\-version\fR show version and build info │ │ │ │ \fB\-\-license\fR show source code license and dependencies │ │ │ │ \fB\-cl\fR, \fB\-\-cache\-list\fR show list of models in cache │ │ │ │ \fB\-\-completion\-bash\fR print source\-able bash completion script for llama.cpp │ │ │ │ @@ -412,16 +411,19 @@ │ │ │ │ \fB\-\-dry\-base\fR N set DRY sampling base value (default: 1.75) │ │ │ │ \fB\-\-dry\-allowed\-length\fR N set allowed length for DRY sampling (default: 2) │ │ │ │ \fB\-\-dry\-penalty\-last\-n\fR N set DRY penalty for the last n tokens (default: \fB\-1\fR, 0 = disable, \fB\-1\fR = │ │ │ │ .IP │ │ │ │ context size) │ │ │ │ .PP │ │ │ │ \fB\-\-dry\-sequence\-breaker\fR STRING add sequence breaker for DRY sampling, clearing out default breakers │ │ │ │ -.TP │ │ │ │ -('\en', ':', '"', '*') in the process; use "none" to not use any │ │ │ │ +.IP │ │ │ │ +('\en', ':', '"', '*') in the process; uload_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ +.PP │ │ │ │ +se "none" to not use any │ │ │ │ +.IP │ │ │ │ sequence breakers │ │ │ │ .PP │ │ │ │ \fB\-\-adaptive\-target\fR N adaptive\-p: select tokens near this probability (valid range 0.0 to │ │ │ │ .TP │ │ │ │ 1.0; negative = disabled) (default: \fB\-1\fR.00) │ │ │ │ [(more info)](https://github.com/ggml\-org/llama.cpp/pull/17927) │ │ │ │ .PP │ │ ├── ./usr/share/man/man1/llama-embedding.1.gz │ │ │ ├── llama-embedding.1 │ │ │ │ @@ -1,13 +1,12 @@ │ │ │ │ .\" DO NOT MODIFY THIS FILE! It was generated by help2man 1.49.3. │ │ │ │ .TH LLAMA-EMBEDDING "1" "April 2026" "debian" "User Commands" │ │ │ │ .SH NAME │ │ │ │ llama-embedding \- llama-embedding │ │ │ │ .SH DESCRIPTION │ │ │ │ -load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ \fB\-\-\-\-\-\fR common params \fB\-\-\-\-\-\fR │ │ │ │ .PP │ │ │ │ \fB\-h\fR, \fB\-\-help\fR, \fB\-\-usage\fR print usage and exit │ │ │ │ \fB\-\-version\fR show version and build info │ │ │ │ \fB\-\-license\fR show source code license and dependencies │ │ │ │ \fB\-cl\fR, \fB\-\-cache\-list\fR show list of models in cache │ │ │ │ \fB\-\-completion\-bash\fR print source\-able bash completion script for llama.cpp │ │ │ │ @@ -485,9 +484,10 @@ │ │ │ │ .PP │ │ │ │ \fB\-\-embd\-separator\fR STRING separator of embeddings (default \en) for example "<#sep#>" │ │ │ │ \fB\-\-cls\-separator\fR STRING separator of classification sequences (default \et) for example │ │ │ │ .IP │ │ │ │ "<#seq#>" │ │ │ │ .PP │ │ │ │ \fB\-\-embd\-gemma\-default\fR use default EmbeddingGemma model (note: can download weights from the │ │ │ │ -.IP │ │ │ │ +.TP │ │ │ │ +load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ internet) │ │ ├── ./usr/share/man/man1/llama-eval-callback.1.gz │ │ │ ├── llama-eval-callback.1 │ │ │ │ @@ -1,13 +1,12 @@ │ │ │ │ .\" DO NOT MODIFY THIS FILE! It was generated by help2man 1.49.3. │ │ │ │ .TH LLAMA-EVAL-CALLBACK "1" "April 2026" "debian" "User Commands" │ │ │ │ .SH NAME │ │ │ │ llama-eval-callback \- llama-eval-callback │ │ │ │ .SH DESCRIPTION │ │ │ │ -load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ \fB\-\-\-\-\-\fR common params \fB\-\-\-\-\-\fR │ │ │ │ .PP │ │ │ │ \fB\-\-\-\-\-\fR sampling params \fB\-\-\-\-\-\fR │ │ │ │ .PP │ │ │ │ \fB\-\-samplers\fR SAMPLERS samplers that will be used for generation in the order, separated by │ │ │ │ .TP │ │ │ │ \&';' │ │ │ │ @@ -399,16 +398,18 @@ │ │ │ │ .PP │ │ │ │ \fB\-hff\fR, \fB\-\-hf\-file\fR FILE Hugging Face model file. If specified, it will override the quant in │ │ │ │ .TP │ │ │ │ \fB\-\-hf\-repo\fR (default: unused) │ │ │ │ (env: LLAMA_ARG_HF_FILE) │ │ │ │ .PP │ │ │ │ \fB\-hfv\fR, \fB\-hfrv\fR, \fB\-\-hf\-repo\-v\fR /[:quant] │ │ │ │ +.IP │ │ │ │ +Hugging Face model repository forload_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ .TP │ │ │ │ -Hugging Face model repository for the vocoder model (default: unused) │ │ │ │ +the vocoder model (default: unused) │ │ │ │ (env: LLAMA_ARG_HF_REPO_V) │ │ │ │ .PP │ │ │ │ \fB\-hffv\fR, \fB\-\-hf\-file\-v\fR FILE Hugging Face model file for the vocoder model (default: unused) │ │ │ │ .IP │ │ │ │ (env: LLAMA_ARG_HF_FILE_V) │ │ │ │ .PP │ │ │ │ \fB\-hft\fR, \fB\-\-hf\-token\fR TOKEN Hugging Face access token (default: value from HF_TOKEN environment │ │ ├── ./usr/share/man/man1/llama-lookup-create.1.gz │ │ │ ├── llama-lookup-create.1 │ │ │ │ @@ -1,13 +1,12 @@ │ │ │ │ .\" DO NOT MODIFY THIS FILE! It was generated by help2man 1.49.3. │ │ │ │ .TH LLAMA-LOOKUP-CREATE "1" "April 2026" "debian" "User Commands" │ │ │ │ .SH NAME │ │ │ │ llama-lookup-create \- llama-lookup-create │ │ │ │ .SH DESCRIPTION │ │ │ │ -load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ \fB\-\-\-\-\-\fR common params \fB\-\-\-\-\-\fR │ │ │ │ .PP │ │ │ │ \fB\-h\fR, \fB\-\-help\fR, \fB\-\-usage\fR print usage and exit │ │ │ │ \fB\-\-version\fR show version and build info │ │ │ │ \fB\-\-license\fR show source code license and dependencies │ │ │ │ \fB\-cl\fR, \fB\-\-cache\-list\fR show list of models in cache │ │ │ │ \fB\-\-completion\-bash\fR print source\-able bash completion script for llama.cpp │ │ │ │ @@ -412,16 +411,19 @@ │ │ │ │ \fB\-\-dry\-base\fR N set DRY sampling base value (default: 1.75) │ │ │ │ \fB\-\-dry\-allowed\-length\fR N set allowed length for DRY sampling (default: 2) │ │ │ │ \fB\-\-dry\-penalty\-last\-n\fR N set DRY penalty for the last n tokens (default: \fB\-1\fR, 0 = disable, \fB\-1\fR = │ │ │ │ .IP │ │ │ │ context size) │ │ │ │ .PP │ │ │ │ \fB\-\-dry\-sequence\-breaker\fR STRING add sequence breaker for DRY sampling, clearing out default breakers │ │ │ │ -.TP │ │ │ │ -('\en', ':', '"', '*') in the process; use "none" to not use any │ │ │ │ +.IP │ │ │ │ +('\en', ':', '"', '*') in the process; uload_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ +.PP │ │ │ │ +se "none" to not use any │ │ │ │ +.IP │ │ │ │ sequence breakers │ │ │ │ .PP │ │ │ │ \fB\-\-adaptive\-target\fR N adaptive\-p: select tokens near this probability (valid range 0.0 to │ │ │ │ .TP │ │ │ │ 1.0; negative = disabled) (default: \fB\-1\fR.00) │ │ │ │ [(more info)](https://github.com/ggml\-org/llama.cpp/pull/17927) │ │ │ │ .PP │ │ ├── ./usr/share/man/man1/llama-lookup-stats.1.gz │ │ │ ├── llama-lookup-stats.1 │ │ │ │ @@ -1,13 +1,12 @@ │ │ │ │ .\" DO NOT MODIFY THIS FILE! It was generated by help2man 1.49.3. │ │ │ │ .TH LLAMA-LOOKUP-STATS "1" "April 2026" "debian" "User Commands" │ │ │ │ .SH NAME │ │ │ │ llama-lookup-stats \- llama-lookup-stats │ │ │ │ .SH DESCRIPTION │ │ │ │ -load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ \fB\-\-\-\-\-\fR common params \fB\-\-\-\-\-\fR │ │ │ │ .PP │ │ │ │ \fB\-h\fR, \fB\-\-help\fR, \fB\-\-usage\fR print usage and exit │ │ │ │ \fB\-\-version\fR show version and build info │ │ │ │ \fB\-\-license\fR show source code license and dependencies │ │ │ │ \fB\-cl\fR, \fB\-\-cache\-list\fR show list of models in cache │ │ │ │ \fB\-\-completion\-bash\fR print source\-able bash completion script for llama.cpp │ │ │ │ @@ -412,16 +411,19 @@ │ │ │ │ \fB\-\-dry\-base\fR N set DRY sampling base value (default: 1.75) │ │ │ │ \fB\-\-dry\-allowed\-length\fR N set allowed length for DRY sampling (default: 2) │ │ │ │ \fB\-\-dry\-penalty\-last\-n\fR N set DRY penalty for the last n tokens (default: \fB\-1\fR, 0 = disable, \fB\-1\fR = │ │ │ │ .IP │ │ │ │ context size) │ │ │ │ .PP │ │ │ │ \fB\-\-dry\-sequence\-breaker\fR STRING add sequence breaker for DRY sampling, clearing out default breakers │ │ │ │ -.TP │ │ │ │ -('\en', ':', '"', '*') in the process; use "none" to not use any │ │ │ │ +.IP │ │ │ │ +('\en', ':', '"', '*') in the process; uload_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ +.PP │ │ │ │ +se "none" to not use any │ │ │ │ +.IP │ │ │ │ sequence breakers │ │ │ │ .PP │ │ │ │ \fB\-\-adaptive\-target\fR N adaptive\-p: select tokens near this probability (valid range 0.0 to │ │ │ │ .TP │ │ │ │ 1.0; negative = disabled) (default: \fB\-1\fR.00) │ │ │ │ [(more info)](https://github.com/ggml\-org/llama.cpp/pull/17927) │ │ │ │ .PP │ │ ├── ./usr/share/man/man1/llama-passkey.1.gz │ │ │ ├── llama-passkey.1 │ │ │ │ @@ -1,13 +1,12 @@ │ │ │ │ .\" DO NOT MODIFY THIS FILE! It was generated by help2man 1.49.3. │ │ │ │ .TH LLAMA-PASSKEY "1" "April 2026" "debian" "User Commands" │ │ │ │ .SH NAME │ │ │ │ llama-passkey \- llama-passkey │ │ │ │ .SH DESCRIPTION │ │ │ │ -load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ \fB\-\-\-\-\-\fR common params \fB\-\-\-\-\-\fR │ │ │ │ .PP │ │ │ │ \fB\-h\fR, \fB\-\-help\fR, \fB\-\-usage\fR print usage and exit │ │ │ │ \fB\-\-version\fR show version and build info │ │ │ │ \fB\-\-license\fR show source code license and dependencies │ │ │ │ \fB\-cl\fR, \fB\-\-cache\-list\fR show list of models in cache │ │ │ │ \fB\-\-completion\-bash\fR print source\-able bash completion script for llama.cpp │ │ │ │ @@ -81,14 +80,16 @@ │ │ │ │ \fB\-e\fR, \fB\-\-escape\fR, \fB\-\-no\-escape\fR whether to process escapes sequences (\en, \er, \et, \e', \e", \e\e) │ │ │ │ .IP │ │ │ │ (default: true) │ │ │ │ .PP │ │ │ │ \fB\-\-rope\-scaling\fR {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by │ │ │ │ .TP │ │ │ │ the model │ │ │ │ +load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ +.IP │ │ │ │ (env: LLAMA_ARG_ROPE_SCALING_TYPE) │ │ │ │ .PP │ │ │ │ \fB\-\-rope\-scale\fR N RoPE context scaling factor, expands context by a factor of N │ │ │ │ .IP │ │ │ │ (env: LLAMA_ARG_ROPE_SCALE) │ │ │ │ .PP │ │ │ │ \fB\-\-rope\-freq\-base\fR N RoPE base frequency, used by NTK\-aware scaling (default: loaded from │ │ ├── ./usr/share/man/man1/llama-retrieval.1.gz │ │ │ ├── llama-retrieval.1 │ │ │ │ @@ -1,13 +1,12 @@ │ │ │ │ .\" DO NOT MODIFY THIS FILE! It was generated by help2man 1.49.3. │ │ │ │ .TH LLAMA-RETRIEVAL "1" "April 2026" "debian" "User Commands" │ │ │ │ .SH NAME │ │ │ │ llama-retrieval \- llama-retrieval │ │ │ │ .SH DESCRIPTION │ │ │ │ -load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ \fB\-\-\-\-\-\fR common params \fB\-\-\-\-\-\fR │ │ │ │ .PP │ │ │ │ \fB\-h\fR, \fB\-\-help\fR, \fB\-\-usage\fR print usage and exit │ │ │ │ \fB\-\-version\fR show version and build info │ │ │ │ \fB\-\-license\fR show source code license and dependencies │ │ │ │ \fB\-cl\fR, \fB\-\-cache\-list\fR show list of models in cache │ │ │ │ \fB\-\-completion\-bash\fR print source\-able bash completion script for llama.cpp │ │ │ │ @@ -412,16 +411,19 @@ │ │ │ │ \fB\-\-dry\-base\fR N set DRY sampling base value (default: 1.75) │ │ │ │ \fB\-\-dry\-allowed\-length\fR N set allowed length for DRY sampling (default: 2) │ │ │ │ \fB\-\-dry\-penalty\-last\-n\fR N set DRY penalty for the last n tokens (default: \fB\-1\fR, 0 = disable, \fB\-1\fR = │ │ │ │ .IP │ │ │ │ context size) │ │ │ │ .PP │ │ │ │ \fB\-\-dry\-sequence\-breaker\fR STRING add sequence breaker for DRY sampling, clearing out default breakers │ │ │ │ -.TP │ │ │ │ -('\en', ':', '"', '*') in the process; use "none" to not use any │ │ │ │ +.IP │ │ │ │ +('\en', ':', '"', '*') in the process; uload_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ +.PP │ │ │ │ +se "none" to not use any │ │ │ │ +.IP │ │ │ │ sequence breakers │ │ │ │ .PP │ │ │ │ \fB\-\-adaptive\-target\fR N adaptive\-p: select tokens near this probability (valid range 0.0 to │ │ │ │ .TP │ │ │ │ 1.0; negative = disabled) (default: \fB\-1\fR.00) │ │ │ │ [(more info)](https://github.com/ggml\-org/llama.cpp/pull/17927) │ │ │ │ .PP │ │ ├── ./usr/share/man/man1/llama-speculative.1.gz │ │ │ ├── llama-speculative.1 │ │ │ │ @@ -1,13 +1,12 @@ │ │ │ │ .\" DO NOT MODIFY THIS FILE! It was generated by help2man 1.49.3. │ │ │ │ .TH LLAMA-SPECULATIVE "1" "April 2026" "debian" "User Commands" │ │ │ │ .SH NAME │ │ │ │ llama-speculative \- llama-speculative │ │ │ │ .SH DESCRIPTION │ │ │ │ -load_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ \fB\-\-\-\-\-\fR common params \fB\-\-\-\-\-\fR │ │ │ │ .PP │ │ │ │ \fB\-h\fR, \fB\-\-help\fR, \fB\-\-usage\fR print usage and exit │ │ │ │ \fB\-\-version\fR show version and build info │ │ │ │ \fB\-\-license\fR show source code license and dependencies │ │ │ │ \fB\-cl\fR, \fB\-\-cache\-list\fR show list of models in cache │ │ │ │ \fB\-\-completion\-bash\fR print source\-able bash completion script for llama.cpp │ │ │ │ @@ -488,15 +487,17 @@ │ │ │ │ .PP │ │ │ │ \fB\-td\fR, \fB\-\-threads\-draft\fR N number of threads to use during generation (default: same as │ │ │ │ .IP │ │ │ │ \fB\-\-threads\fR) │ │ │ │ .SS "-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:" │ │ │ │ .IP │ │ │ │ same as \fB\-\-threads\-draft\fR) │ │ │ │ -.SS "-Cd, --cpu-mask-draft M Draft model CPU affinity mask. Complements cpu-range-draft (default:" │ │ │ │ +.PP │ │ │ │ +\fB\-Cd\fR, \fB\-\-cpu\-mask\-draft\fR M Draft model CPU affinity mask. Complements cpu\-range\-draft (defauload_backend: loaded CPU backend from \fI\,/usr/lib/aarch64\-linux\-gnu/ggml/backends0/libggml\-cpu\-armv8.2_2.so\/\fP │ │ │ │ +lt: │ │ │ │ .IP │ │ │ │ same as \fB\-\-cpu\-mask\fR) │ │ │ │ .PP │ │ │ │ \fB\-Crd\fR, \fB\-\-cpu\-range\-draft\fR lo\-hi Ranges of CPUs for affinity. Complements \fB\-\-cpu\-mask\-draft\fR │ │ │ │ \fB\-\-cpu\-strict\-draft\fR <0|1> Use strict CPU placement for draft model (default: same as │ │ │ │ .IP │ │ │ │ \fB\-\-cpu\-strict\fR)