구글 ai 답변을 따라가니 이런게 있는데
[링크 : https://github.com/ggml-org/llama.cpp/issues/4666]
기본적로 동시에 사용이 가능하도록 되어있나 보다. 오호..
대신 -np 1 을 주면 두개 켜고 양쪽에서 동시에 물어보면 하나만 대답해주다가, 다른쪽이 끝나야 시작한다.
| $ ./llama-b10145/llama-server --help ----- common params ----- -h, --help, --usage print usage and exit --version show version and build info -cl, --cache-list show list of models in cache --completion-bash print source-able bash completion script for llama.cpp -t, --threads N number of CPU threads to use during generation (default: -1) (env: LLAMA_ARG_THREADS) -tb, --threads-batch N number of threads to use during batch and prompt processing (default: same as --threads) -C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range (default: "") -Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask --cpu-strict <0|1> use strict CPU placement (default: 0) --prio N set process/thread priority : low(-1), normal(0), medium(1), high(2), realtime(3) (default: 0) --poll <0...100> use polling level to wait for work (0 - no polling, default: 50) -Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch (default: same as --cpu-mask) -Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch --cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict) --prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: 0) --poll-batch <0|1> use polling to wait for work (default: same as --poll) -c, --ctx-size N size of the prompt context (default: 0, 0 = loaded from model) (env: LLAMA_ARG_CTX_SIZE) -n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity) (env: LLAMA_ARG_N_PREDICT) -b, --batch-size N logical maximum batch size (default: 2048) (env: LLAMA_ARG_BATCH) -ub, --ubatch-size N physical maximum batch size (default: 512) (env: LLAMA_ARG_UBATCH) --keep N number of tokens to keep from the initial prompt (default: 0, -1 = all) --swa-full use full-size SWA cache (default: false) [(more info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055) (env: LLAMA_ARG_SWA_FULL) -fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto') (env: LLAMA_ARG_FLASH_ATTN) --perf, --no-perf whether to enable internal libllama performance timings (default: false) (env: LLAMA_ARG_PERF) -e, --escape, --no-escape whether to process escapes sequences (\n, \r, \t, \', \", \\) (default: true) --rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by the model (env: LLAMA_ARG_ROPE_SCALING_TYPE) --rope-scale N RoPE context scaling factor, expands context by a factor of N (env: LLAMA_ARG_ROPE_SCALE) --rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from model) (env: LLAMA_ARG_ROPE_FREQ_BASE) --rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N (env: LLAMA_ARG_ROPE_FREQ_SCALE) --yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training context size) (env: LLAMA_ARG_YARN_ORIG_CTX) --yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.00, 0.0 = full interpolation) (env: LLAMA_ARG_YARN_EXT_FACTOR) --yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.00) (env: LLAMA_ARG_YARN_ATTN_FACTOR) --yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.00) (env: LLAMA_ARG_YARN_BETA_SLOW) --yarn-beta-fast N YaRN: low correction dim or beta (default: -1.00) (env: LLAMA_ARG_YARN_BETA_FAST) -kvo, --kv-offload, -nkvo, --no-kv-offload whether to enable KV cache offloading (default: enabled) (env: LLAMA_ARG_KV_OFFLOAD) --repack, -nr, --no-repack whether to enable weight repacking (default: enabled) (env: LLAMA_ARG_REPACK) --no-host bypass host buffer allowing extra buffers to be used (env: LLAMA_ARG_NO_HOST) -ctk, --cache-type-k TYPE KV cache data type for K allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1 (default: f16) (env: LLAMA_ARG_CACHE_TYPE_K) -ctv, --cache-type-v TYPE KV cache data type for V allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1 (default: f16) (env: LLAMA_ARG_CACHE_TYPE_V) -dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED) (env: LLAMA_ARG_DEFRAG_THOLD) --rpc SERVERS comma-separated list of RPC servers (host:port) (env: LLAMA_ARG_RPC) --mlock DEPRECATED in favor of `--load-mode`: force system to keep model in RAM rather than swapping or compressing (env: LLAMA_ARG_MLOCK) --mmap, --no-mmap DEPRECATED in favor of `--load-mode`: whether to memory-map model. (if mmap disabled, slower load but may reduce pageouts if not using mlock) (env: LLAMA_ARG_MMAP) -dio, --direct-io, -ndio, --no-direct-io DEPRECATED in favor of `--load-mode`: use DirectIO if available (env: LLAMA_ARG_DIO) -lm, --load-mode MODE model loading mode (default: mmap) - none: no special loading mode - mmap: memory-map model (if mmap disabled, slower load but may reduce pageouts if not using mlock) - mlock: force system to keep model in RAM rather than swapping or compressing - mmap+mlock: mmap + force system to keep model in RAM rather than swapping or compressing - dio: use DirectIO if available (env: LLAMA_ARG_LOAD_MODE) --numa TYPE attempt optimizations that help on some NUMA systems - distribute: spread execution evenly over all nodes - isolate: only spawn threads on CPUs on the node that execution started on - numactl: use the CPU map provided by numactl if run without this previously, it is recommended to drop the system page cache before using this see https://github.com/ggml-org/llama.cpp/issues/1437 (env: LLAMA_ARG_NUMA) -dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't offload) use --list-devices to see a list of available devices (env: LLAMA_ARG_DEVICE) --list-devices print list of available devices and exit -ot, --override-tensor <tensor name pattern>=<buffer type>,... override tensor buffer type (env: LLAMA_ARG_OVERRIDE_TENSOR) -cmoe, --cpu-moe keep all Mixture of Experts (MoE) weights in the CPU (env: LLAMA_ARG_CPU_MOE) -ncmoe, --n-cpu-moe N keep the Mixture of Experts (MoE) weights of the first N layers in the CPU (env: LLAMA_ARG_N_CPU_MOE) -ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM, either an exact number, 'auto', or 'all' (default: auto) (env: LLAMA_ARG_N_GPU_LAYERS) -sm, --split-mode {none,layer,row,tensor} how to split the model across multiple GPUs, one of: - none: use one GPU only - layer (default): split layers and KV across GPUs (pipelined) - row: split weight across GPUs by rows (parallelized) - tensor: split weights and KV across GPUs (parallelized, EXPERIMENTAL) (env: LLAMA_ARG_SPLIT_MODE) -ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of proportions, e.g. 3,1 (env: LLAMA_ARG_TENSOR_SPLIT) -mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for intermediate results and KV (with split-mode = row) (default: 0) (env: LLAMA_ARG_MAIN_GPU) -fit, --fit [on|off] whether to adjust unset arguments to fit in device memory ('on' or 'off', default: 'on') (env: LLAMA_ARG_FIT) -fitt, --fit-target MiB0,MiB1,MiB2,... target margin per device for --fit, comma-separated list of values, single value is broadcast across all devices, default: 1024 (env: LLAMA_ARG_FIT_TARGET) -fitc, --fit-ctx N minimum ctx size that can be set by --fit option, default: 4096 (env: LLAMA_ARG_FIT_CTX) --check-tensors check model tensor data for invalid values (default: false) --override-kv KEY=TYPE:VALUE,... advanced option to override model metadata by key. to specify multiple overrides, either use comma-separated values. types: int, float, bool, str. example: --override-kv tokenizer.ggml.add_bos_token=bool:false,tokenizer.ggml.add_eos_token=bool:false --op-offload, --no-op-offload whether to offload host tensor operations to device (default: true) --lora FNAME path to LoRA adapter (use comma-separated values to load multiple adapters) --lora-scaled FNAME:SCALE,... path to LoRA adapter with user defined scaling (format: FNAME:SCALE,...) note: use comma-separated values --control-vector FNAME add a control vector note: use comma-separated values to add multiple control vectors --control-vector-scaled FNAME:SCALE,... add a control vector with user defined scaling SCALE note: use comma-separated values (format: FNAME:SCALE,...) --control-vector-layer-range START END layer range to apply the control vector(s) to, start and end inclusive -m, --model FNAME model path to load (env: LLAMA_ARG_MODEL) -mu, --model-url MODEL_URL model download url (default: unused) (env: LLAMA_ARG_MODEL_URL) -dr, --docker-repo [<repo>/]<model>[:quant] Docker Hub model repository. repo is optional, default to ai/. quant is optional, default to :latest. example: gemma3 (default: unused) (env: LLAMA_ARG_DOCKER_REPO) -hf, -hfr, --hf-repo <user>/<model>[:quant] Hugging Face model repository; quant is optional, case-insensitive, default to Q4_K_M, or falls back to the first file in the repo if Q4_K_M doesn't exist. mmproj is also downloaded automatically if available. to disable, add --no-mmproj example: ggml-org/GLM-4.7-Flash-GGUF:Q4_K_M (default: unused) (env: LLAMA_ARG_HF_REPO) -hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in --hf-repo (default: unused) (env: LLAMA_ARG_HF_FILE) -hfv, -hfrv, --hf-repo-v <user>/<model>[:quant] Hugging Face model repository for the vocoder model (default: unused) (env: LLAMA_ARG_HF_REPO_V) -hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused) (env: LLAMA_ARG_HF_FILE_V) -hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment variable) (env: HF_TOKEN) --log-disable Log disable --log-file FNAME Log to file (env: LLAMA_ARG_LOG_FILE) --log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto') 'auto' enables colors when output is to a terminal (env: LLAMA_ARG_LOG_COLORS) -v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for debugging) --offline Offline mode: forces use of cache, prevents network access (env: LLAMA_ARG_OFFLINE) -lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be ignored. Values: - 0: generic output - 1: error - 2: warning - 3: info - 4: trace (more info) - 5: debug (default: 3) (env: LLAMA_ARG_LOG_VERBOSITY) --log-prefix, --no-log-prefix Enable prefix in log messages (env: LLAMA_ARG_LOG_PREFIX) --log-timestamps, --no-log-timestamps Enable timestamps in log messages (env: LLAMA_ARG_LOG_TIMESTAMPS) --spec-draft-type-k, -ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1 (default: f16) (env: LLAMA_ARG_SPEC_DRAFT_CACHE_TYPE_K) --spec-draft-type-v, -ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1 (default: f16) (env: LLAMA_ARG_SPEC_DRAFT_CACHE_TYPE_V) ----- sampling params ----- --samplers SAMPLERS samplers that will be used for generation in the order, separated by ';' (default: penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature) -s, --seed SEED RNG seed (default: -1, use random seed for -1) --sampler-seq, --sampling-seq SEQUENCE simplified sequence for samplers that will be used (default: edskypmxt) --ignore-eos ignore end of stream token and continue generating (implies --logit-bias EOS-inf) --temp, --temperature N temperature (default: 0.80) --top-k N top-k sampling (default: 40, 0 = disabled) (env: LLAMA_ARG_TOP_K) --top-p N top-p sampling (default: 0.95, 1.0 = disabled) --min-p N min-p sampling (default: 0.05, 0.0 = disabled) --top-nsigma, --top-n-sigma N top-n-sigma sampling (default: -1.00, -1.0 = disabled) --xtc-probability N xtc probability (default: 0.00, 0.0 = disabled) --xtc-threshold N xtc threshold (default: 0.10, 1.0 = disabled) --typical, --typical-p N locally typical sampling, parameter p (default: 1.00, 1.0 = disabled) --repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1 = ctx_size) --repeat-penalty N penalize repeat sequence of tokens (default: 1.00, 1.0 = disabled) --presence-penalty N repeat alpha presence penalty (default: 0.00, 0.0 = disabled) --frequency-penalty N repeat alpha frequency penalty (default: 0.00, 0.0 = disabled) --dry-multiplier N set DRY sampling multiplier (default: 0.00, 0.0 = disabled) --dry-base N set DRY sampling base value (default: 1.75) --dry-allowed-length N set allowed length for DRY sampling (default: 2) --dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 = context size) --dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers ('\n', ':', '"', '*') in the process; use "none" to not use any sequence breakers --adaptive-target N adaptive-p: select tokens near this probability (valid range 0.0 to 1.0; negative = disabled) (default: -1.00) [(more info)](https://github.com/ggml-org/llama.cpp/pull/17927) --adaptive-decay N adaptive-p: decay rate for target adaptation over time. lower values are more reactive, higher values are more stable. (valid range 0.0 to 0.99) (default: 0.90) --dynatemp-range N dynamic temperature range (default: 0.00, 0.0 = disabled) --dynatemp-exp N dynamic temperature exponent (default: 1.00) --mirostat N use Mirostat sampling. Top K, Nucleus and Locally Typical samplers are ignored if used. (default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0) --mirostat-lr N Mirostat learning rate, parameter eta (default: 0.10) --mirostat-ent N Mirostat target entropy, parameter tau (default: 5.00) -l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion, i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello', or `--logit-bias 15043-1` to decrease likelihood of token ' Hello' --grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/ dir) --grammar-file FNAME file to read grammar from -j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g. `{}` for any JSON object For schemas w/ external $refs, use --grammar + example/json_schema_to_grammar.py instead -jf, --json-schema-file FILE File containing a JSON schema to constrain generations (https://json-schema.org/), e.g. `{}` for any JSON object For schemas w/ external $refs, use --grammar + example/json_schema_to_grammar.py instead -bs, --backend-sampling enable backend sampling (experimental) (default: disabled) (env: LLAMA_ARG_BACKEND_SAMPLING) ----- speculative params ----- --spec-draft-hf, -hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant] Same as --hf-repo, but for the draft model (default: unused) (env: LLAMA_ARG_SPEC_DRAFT_HF_REPO) --spec-draft-threads, -td, --threads-draft N number of threads to use during generation (default: same as --threads) --spec-draft-threads-batch, -tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default: same as --threads-draft) --spec-draft-cpu-mask, -Cd, --cpu-mask-draft M Draft model CPU affinity mask. Complements cpu-range-draft (default: same as --cpu-mask) --spec-draft-cpu-range, -Crd, --cpu-range-draft lo-hi Ranges of CPUs for affinity. Complements --cpu-mask-draft --spec-draft-cpu-strict, --cpu-strict-draft <0|1> Use strict CPU placement for draft model (default: same as --cpu-strict) --spec-draft-prio, --prio-draft N set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: 0) --spec-draft-poll, --poll-draft <0|1> Use polling to wait for draft model work (default: same as --poll) --spec-draft-cpu-mask-batch, -Cbd, --cpu-mask-batch-draft M Draft model CPU affinity mask. Complements cpu-range-draft (default: same as --cpu-mask) --spec-draft-cpu-strict-batch, --cpu-strict-batch-draft <0|1> Use strict CPU placement for draft model (default: --cpu-strict-draft) --spec-draft-prio-batch, --prio-batch-draft N set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: 0) --spec-draft-poll-batch, --poll-batch-draft <0|1> Use polling to wait for draft model work (default: --poll-draft) --spec-draft-override-tensor, -otd, --override-tensor-draft <tensor name pattern>=<buffer type>,... override tensor buffer type for draft model --spec-draft-cpu-moe, -cmoed, --cpu-moe-draft keep all Mixture of Experts (MoE) weights in the CPU for the draft model (env: LLAMA_ARG_SPEC_DRAFT_CPU_MOE) --spec-draft-n-cpu-moe, --spec-draft-ncmoe, -ncmoed, --n-cpu-moe-draft N keep the Mixture of Experts (MoE) weights of the first N layers in the CPU for the draft model (env: LLAMA_ARG_SPEC_DRAFT_N_CPU_MOE) --spec-draft-n-max N number of tokens to draft for speculative decoding (default: 3) (env: LLAMA_ARG_SPEC_DRAFT_N_MAX) --spec-draft-n-min N minimum number of draft tokens to use for speculative decoding (default: 0) (env: LLAMA_ARG_SPEC_DRAFT_N_MIN) --spec-draft-p-split, --draft-p-split P speculative decoding split probability (default: 0.10) (env: LLAMA_ARG_SPEC_DRAFT_P_SPLIT) --spec-draft-p-min, --draft-p-min P minimum speculative decoding probability (greedy) (default: 0.00) (env: LLAMA_ARG_SPEC_DRAFT_P_MIN) --spec-draft-backend-sampling, --no-spec-draft-backend-sampling offload draft sampling to the backend (default: enabled) (env: LLAMA_ARG_SPEC_DRAFT_BACKEND_SAMPLING) --spec-draft-device, -devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model (none = don't offload) use --list-devices to see a list of available devices --spec-draft-ngl, -ngld, --gpu-layers-draft, --n-gpu-layers-draft N max. number of draft model layers to store in VRAM, either an exact number, 'auto', or 'all' (default: auto) (env: LLAMA_ARG_N_GPU_LAYERS_DRAFT) --spec-draft-model, -md, --model-draft FNAME draft model for speculative decoding (default: unused) (env: LLAMA_ARG_SPEC_DRAFT_MODEL) --spec-type none,draft-simple,draft-eagle3,draft-mtp,draft-dflash,ngram-simple,ngram-map-k,ngram-map-k4v,ngram-mod,ngram-cache comma-separated list of types of speculative decoding to use (default: none) (env: LLAMA_ARG_SPEC_TYPE) --spec-ngram-mod-n-min N minimum number of ngram tokens to use for ngram-based speculative decoding (default: 48) --spec-ngram-mod-n-max N maximum number of ngram tokens to use for ngram-based speculative decoding (default: 64) --spec-ngram-mod-n-match N ngram-mod lookup length (default: 24) --spec-ngram-simple-size-n N ngram size N for ngram-simple speculative decoding, length of lookup n-gram (default: 12) --spec-ngram-simple-size-m N ngram size M for ngram-simple speculative decoding, length of draft m-gram (default: 48) --spec-ngram-simple-min-hits N minimum hits for ngram-simple speculative decoding (default: 1) --spec-ngram-map-k-size-n N ngram size N for ngram-map-k speculative decoding, length of lookup n-gram (default: 12) --spec-ngram-map-k-size-m N ngram size M for ngram-map-k speculative decoding, length of draft m-gram (default: 48) --spec-ngram-map-k-min-hits N minimum hits for ngram-map-k speculative decoding (default: 1) --spec-ngram-map-k4v-size-n N ngram size N for ngram-map-k4v speculative decoding, length of lookup n-gram (default: 12) --spec-ngram-map-k4v-size-m N ngram size M for ngram-map-k4v speculative decoding, length of draft m-gram (default: 48) --spec-ngram-map-k4v-min-hits N minimum hits for ngram-map-k4v speculative decoding (default: 1) --draft, --draft-n, --draft-max N the argument has been removed. use --spec-draft-n-max or --spec-ngram-mod-n-max (env: LLAMA_ARG_DRAFT_MAX) --draft-min, --draft-n-min N the argument has been removed. use --spec-draft-n-min or --spec-ngram-mod-n-min (env: LLAMA_ARG_DRAFT_MIN) --spec-ngram-size-n N the argument has been removed. use the respective --spec-ngram-*-size-n or --spec-ngram-mod-n-match --spec-ngram-size-m N the argument has been removed. use the respective --spec-ngram-*-size-m --spec-ngram-min-hits N the argument has been removed. use the respective --spec-ngram-*-min-hits ----- example-specific params ----- -lcs, --lookup-cache-static FNAME path to static lookup cache to use for lookup decoding (not updated by generation) -lcd, --lookup-cache-dynamic FNAME path to dynamic lookup cache to use for lookup decoding (updated by generation) -ctxcp, --ctx-checkpoints, --swa-checkpoints N max number of context checkpoints to create per slot (default: 32)[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293) (env: LLAMA_ARG_CTX_CHECKPOINTS) -cms, --checkpoint-min-step N minimum spacing between context checkpoints in tokens (default: 8192, 0 = no minimum) (env: LLAMA_ARG_CHECKPOINT_MIN_SPACING_NT) -cram, --cache-ram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 - disable)[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391) (env: LLAMA_ARG_CACHE_RAM) -kvu, --kv-unified, -no-kvu, --no-kv-unified use single unified KV buffer shared across all sequences (default: enabled if number of slots is auto) (env: LLAMA_ARG_KV_UNIFIED) --cache-idle-slots, --no-cache-idle-slots save idle slots to the prompt cache on new task, and clear them when using unified KV (default: enabled, requires cache-ram) (env: LLAMA_ARG_CACHE_IDLE_SLOTS) --context-shift, --no-context-shift whether to use context shift on infinite text generation (default: disabled) (env: LLAMA_ARG_CONTEXT_SHIFT) -r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode -sp, --special special tokens output enabled (default: false) --warmup, --no-warmup whether to perform warmup with an empty run (default: enabled) --spm-infill use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: disabled) --pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified (env: LLAMA_ARG_POOLING) -np, --parallel N number of server slots (default: -1, -1 = auto) (env: LLAMA_ARG_N_PARALLEL) -cb, --cont-batching, -nocb, --no-cont-batching whether to enable continuous batching (a.k.a dynamic batching) (default: enabled) (env: LLAMA_ARG_CONT_BATCHING) -mm, --mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md note: if -hf is used, this argument can be omitted (env: LLAMA_ARG_MMPROJ) -mmu, --mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md (env: LLAMA_ARG_MMPROJ_URL) --mmproj-auto, --no-mmproj, --no-mmproj-auto whether to use multimodal projector file (if available), useful when using -hf (default: enabled) (env: LLAMA_ARG_MMPROJ_AUTO) --mmproj-offload, --no-mmproj-offload whether to enable GPU offloading for multimodal projector (default: enabled) (env: LLAMA_ARG_MMPROJ_OFFLOAD) --image-min-tokens N minimum number of tokens each image can take, only used by vision models with dynamic resolution (default: read from model) (env: LLAMA_ARG_IMAGE_MIN_TOKENS) --image-max-tokens N maximum number of tokens each image can take, only used by vision models with dynamic resolution (default: read from model) (env: LLAMA_ARG_IMAGE_MAX_TOKENS) --mtmd-batch-max-tokens N maximum number of image tokens per batch when encoding images (default: 1024) (env: LLAMA_ARG_MTMD_BATCH_MAX_TOKENS) -a, --alias STRING set model name aliases, comma-separated (to be used by API) (env: LLAMA_ARG_ALIAS) --tags STRING set model tags, comma-separated (informational, not used for routing) (env: LLAMA_ARG_TAGS) --embd-normalize N normalisation for embeddings (default: 2) (-1=none, 0=max absolute int16, 1=taxicab, 2=euclidean, >2=p-norm) --host HOST ip address to listen, or bind to an UNIX socket if the address ends with .sock (default: 127.0.0.1) (env: LLAMA_ARG_HOST) --port PORT port to listen (default: 8080) (env: LLAMA_ARG_PORT) --reuse-port allow multiple sockets to bind to the same port (default: disabled) (env: LLAMA_ARG_REUSE_PORT) --path PATH path to serve static files from (default: ) (env: LLAMA_ARG_STATIC_PATH) --cors-origins ORIGINS comma-separated list of allowed origins for CORS (default: *) if set to special value 'localhost', reflect the Origin header only if it is localhost (env: LLAMA_ARG_CORS_ORIGINS) --cors-methods METHODS comma-separated list of allowed methods for CORS (default: GET, POST, DELETE, OPTIONS) (env: LLAMA_ARG_CORS_METHODS) --cors-headers HEADERS comma-separated list of allowed headers for CORS (default: *) (env: LLAMA_ARG_CORS_HEADERS) --cors-credentials, --no-cors-credentials whether to allow credentials for CORS (default: enabled) note: if this is enabled and --cors-origins is set to * (default), the Origin header will be echoed back, and credentials will always be allowed (env: LLAMA_ARG_CORS_CREDENTIALS) --api-prefix PREFIX prefix path the server serves from, without the trailing slash (default: ) (env: LLAMA_ARG_API_PREFIX) --ui-config, --webui-config JSON JSON that provides default UI settings (overrides UI defaults) (env: LLAMA_ARG_UI_CONFIG) --ui-config-file, --webui-config-file PATH JSON file that provides default UI settings (overrides UI defaults) (env: LLAMA_ARG_UI_CONFIG_FILE) --ui-mcp-proxy, --webui-mcp-proxy, --no-ui-mcp-proxy, --no-webui-mcp-proxy experimental: whether to enable MCP CORS proxy - do not enable in untrusted environments (default: disabled) (env: LLAMA_ARG_UI_MCP_PROXY) --tools TOOL1,TOOL2,... experimental: whether to enable built-in tools for AI agents - do not enable in untrusted environments (default: no tools) specify "all" to enable all tools available tools: read_file, file_glob_search, grep_search, exec_shell_command, write_file, edit_file, get_datetime note: for security reasons, this will limit --cors-origins to localhost by default (env: LLAMA_ARG_TOOLS) --mcp-servers-config PATH experimental: path to JSON file with MCP server definitions (Cursor-compatible format) - do not enable in untrusted environments (default: none) note: for security reasons, this will limit --cors-origins to localhost by default (env: LLAMA_ARG_MCP_SERVERS_CONFIG) --mcp-servers-json JSON experimental: inline JSON with MCP server definitions (Cursor-compatible format) - do not enable in untrusted environments (default: none) note: for security reasons, this will limit --cors-origins to localhost by default (env: LLAMA_ARG_MCP_SERVERS_JSON) -ag, --agent, -no-ag, --no-agent whether to enable CORS proxy and all built-in tools - do not enable in untrusted environments (default: disabled) note: for security reasons, this will limit --cors-origins to localhost by default (env: LLAMA_ARG_AGENT) --ui, --webui, --no-ui, --no-webui whether to enable the Web UI (default: enabled) (env: LLAMA_ARG_UI) --embedding, --embeddings restrict to only support embedding use case; use only with dedicated embedding models (default: disabled) (env: LLAMA_ARG_EMBEDDINGS) --rerank, --reranking enable reranking endpoint on server (default: disabled) (env: LLAMA_ARG_RERANKING) --api-key KEY API key to use for authentication, multiple keys can be provided as a comma-separated list (default: none) (env: LLAMA_API_KEY) --api-key-file FNAME path to file containing API keys, one per line; lines starting with a hash are treated as comments (default: none) (env: LLAMA_ARG_API_KEY_FILE) --ssl-key-file FNAME path to file a PEM-encoded SSL private key (env: LLAMA_ARG_SSL_KEY_FILE) --ssl-cert-file FNAME path to file a PEM-encoded SSL certificate (env: LLAMA_ARG_SSL_CERT_FILE) --chat-template-kwargs STRING sets additional params for the json template parser, must be a valid json object string, e.g. '{"key1":"value1","key2":"value2"}' (env: LLAMA_ARG_CHAT_TEMPLATE_KWARGS) -to, --timeout N server read/write timeout in seconds (default: 3600) (env: LLAMA_ARG_TIMEOUT) --sse-ping-interval N server SSE ping interval in seconds (-1 = disabled, default: 30) (env: LLAMA_ARG_SSE_PING_INTERVAL) --threads-http N number of threads used to process HTTP requests (default: -1) (env: LLAMA_ARG_THREADS_HTTP) --cache-prompt, --no-cache-prompt whether to enable prompt caching (default: enabled) (env: LLAMA_ARG_CACHE_PROMPT) --cache-reuse N min chunk size to attempt reusing from the cache via KV shifting, requires prompt caching to be enabled (default: 0) [(card)](https://ggml.ai/f0.png) (env: LLAMA_ARG_CACHE_REUSE) --metrics enable prometheus compatible metrics endpoint (default: disabled) (env: LLAMA_ARG_ENDPOINT_METRICS) --props enable changing global properties via POST /props (default: disabled) (env: LLAMA_ARG_ENDPOINT_PROPS) --slots, --no-slots expose slots monitoring endpoint (default: enabled) (env: LLAMA_ARG_ENDPOINT_SLOTS) --slot-save-path PATH path to save slot kv cache (default: disabled) --media-path PATH directory for loading local media files; files can be accessed via file:// URLs using relative paths (default: disabled) --models-dir PATH directory containing models for the router server (default: disabled) (env: LLAMA_ARG_MODELS_DIR) --models-preset PATH path to INI file containing model presets for the router server (default: disabled) (env: LLAMA_ARG_MODELS_PRESET) --models-max N for router server, maximum number of models to load simultaneously (default: 4, 0 = unlimited) (env: LLAMA_ARG_MODELS_MAX) --models-autoload, --no-models-autoload for router server, whether to automatically load models (default: enabled) (env: LLAMA_ARG_MODELS_AUTOLOAD) --jinja, --no-jinja whether to use jinja template engine for chat (default: enabled) (env: LLAMA_ARG_JINJA) --reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the response, and in which format they're returned; one of: - none: leaves thoughts unparsed in `message.content` - deepseek: puts thoughts in `message.reasoning_content` - deepseek-legacy: keeps `<think>` tags in `message.content` while also populating `message.reasoning_content` (default: auto) (env: LLAMA_ARG_THINK) -rea, --reasoning [on|off|auto] Use reasoning/thinking in the chat ('on', 'off', or 'auto', default: 'auto' (detect from template)) (env: LLAMA_ARG_REASONING) --reasoning-budget N token budget for thinking: -1 for unrestricted, 0 for immediate end, N>0 for token budget (default: -1) (env: LLAMA_ARG_THINK_BUDGET) --reasoning-budget-message MESSAGE message injected before the end-of-thinking tag when reasoning budget is exhausted (default: none) (env: LLAMA_ARG_THINK_BUDGET_MESSAGE) --reasoning-preserve, --no-reasoning-preserve preserve reasoning trace in the full history, not just the last assistant message (default: template default) compatible with certain templates having 'supports_preserve_reasoning' capability example: https://docs.z.ai/guides/capabilities/thinking-mode#preserved-thinking (env: LLAMA_ARG_REASONING_PRESERVE) --chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's metadata) if suffix/prefix are specified, template will be disabled only commonly used templates are accepted (unless --jinja is set before this flag): list of built-in templates: bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml, command-r, deepseek, deepseek-ocr, deepseek2, deepseek3, exaone-moe, exaone3, exaone4, falcon3, gemma, gigachat, glmedge, gpt-oss, granite, granite-4.0, granite-4.1, grok-2, hunyuan-dense, hunyuan-moe, hunyuan-vl, kimi-k2, llama2, llama2-sys, llama2-sys-bos, llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1, mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch, openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss, smolvlm, solar-open, vicuna, vicuna-orca, yandex, zephyr (env: LLAMA_ARG_CHAT_TEMPLATE) --chat-template-file JINJA_TEMPLATE_FILE set custom jinja chat template file (default: template taken from model's metadata) if suffix/prefix are specified, template will be disabled only commonly used templates are accepted (unless --jinja is set before this flag): list of built-in templates: bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml, command-r, deepseek, deepseek-ocr, deepseek2, deepseek3, exaone-moe, exaone3, exaone4, falcon3, gemma, gigachat, glmedge, gpt-oss, granite, granite-4.0, granite-4.1, grok-2, hunyuan-dense, hunyuan-moe, hunyuan-vl, kimi-k2, llama2, llama2-sys, llama2-sys-bos, llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1, mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch, openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss, smolvlm, solar-open, vicuna, vicuna-orca, yandex, zephyr (env: LLAMA_ARG_CHAT_TEMPLATE_FILE) --skip-chat-parsing, --no-skip-chat-parsing force a pure content parser, even if a Jinja template is specified; model will output everything in the content section, including any reasoning and/or tool calls (default: disabled) (env: LLAMA_ARG_SKIP_CHAT_PARSING) --prefill-assistant, --no-prefill-assistant whether to prefill the assistant's response if the last message is an assistant message (default: prefill enabled) when this flag is set, if the last message is an assistant message then it will be treated as a full message and not prefilled (env: LLAMA_ARG_PREFILL_ASSISTANT) -sps, --slot-prompt-similarity SIMILARITY how much the prompt of a request must match the prompt of a slot in order to use that slot (default: 0.10, 0.0 = disabled) --lora-init-without-apply load LoRA adapters without applying them (apply later via POST /lora-adapters) (default: disabled) --sleep-idle-seconds SECONDS number of seconds of idleness after which the server will sleep (default: -1; -1 = disabled) --log-prompts-dir PATH Log prompts to directory (auto-created if not present; only used for debugging, default: disabled) -mv, --model-vocoder FNAME vocoder model for audio generation (default: unused) --tts-use-guide-tokens Use guide tokens to improve TTS word recall --embd-gemma-default use default EmbeddingGemma model (note: can download weights from the internet) --fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the internet) --fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the internet) --fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the internet) --fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can download weights from the internet) --fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note: can download weights from the internet) --fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights from the internet) --gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet) --gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet) --vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet) --vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet) --spec-default enable default speculative decoding config |
오호.. 여기서 초기화시에 뜨는구나
| $ ./llama-b10145/llama-server --host 0.0.0.0 --port 8081 -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf 0.00.420.157 I cmn common_param: common_params_print_info: verbosity = 3 (adjust with the `-lv N` CLI arg) 0.00.420.880 W srv llama_server: ----------------- 0.00.420.884 W srv llama_server: CORS is set to allow all origins ('*') and no API key is set 0.00.420.884 W srv llama_server: this can be a security risk (cross-origin attacks) 0.00.420.884 W srv llama_server: more info: https://github.com/ggml-org/llama.cpp/pull/25655 0.00.420.885 W srv llama_server: ----------------- 0.00.422.219 I srv load_model: loading model 'model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf' 0.04.105.768 I srv load_model: initializing, n_slots = 4, n_ctx_slot = 262144, kv_unified = 'true' 0.04.123.306 I srv init: chat template supports preserving reasoning, consider enabling it via --reasoning-preserve 0.04.123.354 I srv llama_server: model loaded 0.04.123.359 I srv llama_server: listening on http://0.0.0.0:8081 |
| $ ./llama-b10145/llama-server --host 0.0.0.0 --port 8081 -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -np 1 0.00.146.608 I cmn common_param: common_params_print_info: verbosity = 3 (adjust with the `-lv N` CLI arg) 0.00.147.059 W srv llama_server: ----------------- 0.00.147.062 W srv llama_server: CORS is set to allow all origins ('*') and no API key is set 0.00.147.062 W srv llama_server: this can be a security risk (cross-origin attacks) 0.00.147.063 W srv llama_server: more info: https://github.com/ggml-org/llama.cpp/pull/25655 0.00.147.063 W srv llama_server: ----------------- 0.00.148.332 I srv load_model: loading model 'model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf' 0.03.930.586 I srv load_model: initializing, n_slots = 1, n_ctx_slot = 262144, kv_unified = 'false' 0.03.948.234 I srv init: chat template supports preserving reasoning, consider enabling it via --reasoning-preserve 0.03.948.286 I srv llama_server: model loaded 0.03.948.292 I srv llama_server: listening on http://0.0.0.0:8081 |
| $ ./llama-b10145/llama-server --host 0.0.0.0 --port 8081 -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -np 8 0.00.294.633 I cmn common_param: common_params_print_info: verbosity = 3 (adjust with the `-lv N` CLI arg) 0.00.295.199 W srv llama_server: ----------------- 0.00.295.203 W srv llama_server: CORS is set to allow all origins ('*') and no API key is set 0.00.295.203 W srv llama_server: this can be a security risk (cross-origin attacks) 0.00.295.203 W srv llama_server: more info: https://github.com/ggml-org/llama.cpp/pull/25655 0.00.295.203 W srv llama_server: ----------------- 0.00.296.551 I srv load_model: loading model 'model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf' 0.10.613.365 I srv load_model: initializing, n_slots = 8, n_ctx_slot = 32768, kv_unified = 'false' 0.10.634.191 I srv init: chat template supports preserving reasoning, consider enabling it via --reasoning-preserve 0.10.634.243 I srv llama_server: model loaded 0.10.634.251 I srv llama_server: listening on http://0.0.0.0:8081 |
+
1080ti 11GB * 2 에다가 멀티로 괴롭혀 보기
3개 동시에는 27.03+28.21+28.21 = 77.45
1 개 단독에는 60.29
4개 동시에는 23.33+23.39+23.40+23.40 = 93.52
개인당 성능은 낮아져도 전체 출력량은 늘어나는 마법(!)
| $ ./llama-b9692/llama-server -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -np 8 -sm layer --host 0.0.0.0 0.00.585.225 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg) 0.00.585.228 I device_info: 0.00.591.290 I - Vulkan0 : Intel(R) UHD Graphics 630 (CFL GT2) (23816 MiB, 23816 MiB free) 0.00.597.326 I - Vulkan1 : NVIDIA GeForce GTX 1080 Ti (11510 MiB, 11248 MiB free) 0.00.603.380 I - Vulkan2 : NVIDIA GeForce GTX 1080 Ti (11510 MiB, 11390 MiB free) 0.00.603.384 I - CPU : Intel(R) Core(TM) i5-8500 CPU @ 3.00GHz (31754 MiB, 31754 MiB free) 0.00.603.455 I system_info: n_threads = 6 (n_threads_batch = 6) / 6 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 0.00.603.859 I srv init: running without SSL 0.00.605.375 I srv init: using 12 threads for HTTP server 0.00.605.651 I srv start: binding port with default address family 0.00.606.857 I srv llama_server: loading model 0.00.606.866 I srv load_model: loading model 'model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf' 0.00.606.906 I common_init_result: fitting params to device memory ... 0.00.606.906 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on) 0.29.465.233 W llama_context: n_ctx_seq (32768) < n_ctx_train (262144) -- the full capacity of the model will not be utilized 0.29.859.072 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable) 0.30.807.769 I srv load_model: initializing slots, n_slots = 8 0.31.132.316 W srv load_model: speculative decoding will use checkpoints 0.31.132.491 W common_speculative_init: no implementations specified for speculative decoding 0.31.132.503 I slot load_model: id 0 | task -1 | new slot, n_ctx = 32768 0.31.132.514 I slot load_model: id 1 | task -1 | new slot, n_ctx = 32768 0.31.132.516 I slot load_model: id 2 | task -1 | new slot, n_ctx = 32768 0.31.132.518 I slot load_model: id 3 | task -1 | new slot, n_ctx = 32768 0.31.132.519 I slot load_model: id 4 | task -1 | new slot, n_ctx = 32768 0.31.132.521 I slot load_model: id 5 | task -1 | new slot, n_ctx = 32768 0.31.132.522 I slot load_model: id 6 | task -1 | new slot, n_ctx = 32768 0.31.132.524 I slot load_model: id 7 | task -1 | new slot, n_ctx = 32768 0.31.132.789 I srv load_model: prompt cache is enabled, size limit: 8192 MiB 0.31.132.797 I srv load_model: use `--cache-ram 0` to disable the prompt cache 0.31.132.799 I srv load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391 0.31.132.800 I srv load_model: context checkpoints enabled, max = 32, min spacing = 256 0.31.132.857 I srv init: idle slots will be saved to prompt cache upon starting a new task 0.31.193.165 I init: chat template, example_format: '<|im_start|>system You are a helpful assistant<|im_end|> <|im_start|>user Hello<|im_end|> <|im_start|>assistant Hi there<|im_end|> <|im_start|>user How are you?<|im_end|> <|im_start|>assistant <think> ' 0.31.210.696 I srv init: init: chat template, thinking = 1 0.31.210.739 I srv llama_server: model loaded 0.31.210.743 I srv llama_server: server is listening on http://0.0.0.0:8080 0.31.210.748 I srv update_slots: all slots are idle 1.33.205.501 I srv params_from_: Chat format: peg-native 1.33.205.716 I slot get_availabl: id 7 | task -1 | selected slot by LRU, t_last = -1 1.33.205.718 I srv get_availabl: updating prompt cache 1.33.205.734 I srv load: - looking for better prompt, base f_keep = -1.000, sim = 0.000 1.33.205.738 I srv update: - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 262144 tokens, 8589934592 est) 1.33.205.738 I srv get_availabl: prompt cache update took 0.02 ms 1.33.205.785 I reasoning-budget: activated, budget=2147483647 tokens 1.33.205.786 I reasoning-budget: deactivated (natural end) 1.33.205.807 I slot launch_slot_: id 7 | task 0 | processing task, is_child = 0 1.33.205.810 I slot process_sing: id 0 | task -1 | saving idle slot to prompt cache 1.33.205.810 I slot process_sing: id 1 | task -1 | saving idle slot to prompt cache 1.33.205.811 I slot process_sing: id 2 | task -1 | saving idle slot to prompt cache 1.33.205.811 I slot process_sing: id 3 | task -1 | saving idle slot to prompt cache 1.33.205.811 I slot process_sing: id 4 | task -1 | saving idle slot to prompt cache 1.33.205.811 I slot process_sing: id 5 | task -1 | saving idle slot to prompt cache 1.33.205.812 I slot process_sing: id 6 | task -1 | saving idle slot to prompt cache 1.33.735.604 I srv params_from_: Chat format: peg-native 1.34.598.059 I srv params_from_: Chat format: peg-native 1.34.696.573 I slot get_availabl: id 6 | task -1 | selected slot by LRU, t_last = -1 1.34.696.575 I srv get_availabl: updating prompt cache 1.34.696.578 I srv load: - looking for better prompt, base f_keep = -1.000, sim = 0.000 1.34.696.581 I srv update: - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 262144 tokens, 8589934592 est) 1.34.696.581 I srv get_availabl: prompt cache update took 0.01 ms 1.34.696.664 I reasoning-budget: activated, budget=2147483647 tokens 1.34.696.664 I reasoning-budget: deactivated (natural end) 1.34.696.675 I slot launch_slot_: id 6 | task 2 | processing task, is_child = 0 1.34.696.676 I slot process_sing: id 0 | task -1 | saving idle slot to prompt cache 1.34.696.676 I slot process_sing: id 1 | task -1 | saving idle slot to prompt cache 1.34.696.677 I slot process_sing: id 2 | task -1 | saving idle slot to prompt cache 1.34.696.677 I slot process_sing: id 3 | task -1 | saving idle slot to prompt cache 1.34.696.677 I slot process_sing: id 4 | task -1 | saving idle slot to prompt cache 1.34.696.678 I slot process_sing: id 5 | task -1 | saving idle slot to prompt cache 1.34.696.680 I slot get_availabl: id 5 | task -1 | selected slot by LRU, t_last = -1 1.34.696.680 I srv get_availabl: updating prompt cache 1.34.696.681 I srv load: - looking for better prompt, base f_keep = -1.000, sim = 0.000 1.34.696.682 I srv update: - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 262144 tokens, 8589934592 est) 1.34.696.683 I srv get_availabl: prompt cache update took 0.00 ms 1.34.696.719 I reasoning-budget: activated, budget=2147483647 tokens 1.34.696.719 I reasoning-budget: deactivated (natural end) 1.34.696.726 I slot launch_slot_: id 5 | task 3 | processing task, is_child = 0 1.34.696.727 I slot process_sing: id 0 | task -1 | saving idle slot to prompt cache 1.34.696.727 I slot process_sing: id 1 | task -1 | saving idle slot to prompt cache 1.34.696.727 I slot process_sing: id 2 | task -1 | saving idle slot to prompt cache 1.34.696.728 I slot process_sing: id 3 | task -1 | saving idle slot to prompt cache 1.34.696.728 I slot process_sing: id 4 | task -1 | saving idle slot to prompt cache 1.34.988.803 I slot create_check: id 7 | task 0 | created context checkpoint 1 of 32 (pos_min = 28, pos_max = 28, n_tokens = 29, size = 62.813 MiB) 1.36.425.849 I slot create_check: id 5 | task 3 | created context checkpoint 1 of 32 (pos_min = 29, pos_max = 29, n_tokens = 30, size = 62.813 MiB) 1.36.474.645 I slot create_check: id 6 | task 2 | created context checkpoint 1 of 32 (pos_min = 28, pos_max = 28, n_tokens = 29, size = 62.813 MiB) 1.40.908.116 I slot print_timing: id 7 | task 0 | n_decoded = 100, tg = 22.03 t/s 1.40.941.055 I slot print_timing: id 5 | task 3 | n_decoded = 100, tg = 25.99 t/s 1.40.942.209 I slot print_timing: id 6 | task 2 | n_decoded = 100, tg = 25.99 t/s 1.43.931.725 I slot print_timing: id 7 | task 0 | n_decoded = 187, tg = 24.72 t/s 1.43.964.981 I slot print_timing: id 5 | task 3 | n_decoded = 187, tg = 27.21 t/s 1.43.967.199 I slot print_timing: id 6 | task 2 | n_decoded = 187, tg = 27.21 t/s 1.46.946.263 I slot print_timing: id 7 | task 0 | n_decoded = 273, tg = 25.81 t/s 1.46.977.938 I slot print_timing: id 5 | task 3 | n_decoded = 273, tg = 27.62 t/s 1.46.978.729 I slot print_timing: id 6 | task 2 | n_decoded = 273, tg = 27.62 t/s 1.49.960.552 I slot print_timing: id 7 | task 0 | n_decoded = 360, tg = 26.48 t/s 1.49.991.687 I slot print_timing: id 5 | task 3 | n_decoded = 360, tg = 27.91 t/s 1.49.992.249 I slot print_timing: id 6 | task 2 | n_decoded = 360, tg = 27.91 t/s 1.52.980.539 I slot print_timing: id 7 | task 0 | n_decoded = 449, tg = 27.03 t/s 1.53.012.245 I slot print_timing: id 5 | task 3 | n_decoded = 449, tg = 28.21 t/s 1.53.013.043 I slot print_timing: id 6 | task 2 | n_decoded = 449, tg = 28.21 t/s 1.55.990.978 I slot print_timing: id 7 | task 0 | n_decoded = 534, tg = 27.21 t/s 1.56.022.948 I slot print_timing: id 5 | task 3 | n_decoded = 534, tg = 28.21 t/s 1.56.023.747 I slot print_timing: id 6 | task 2 | n_decoded = 534, tg = 28.21 t/s 1.58.999.665 I slot print_timing: id 7 | task 0 | n_decoded = 620, tg = 27.40 t/s 1.59.031.691 I slot print_timing: id 5 | task 3 | n_decoded = 620, tg = 28.26 t/s 1.59.032.485 I slot print_timing: id 6 | task 2 | n_decoded = 620, tg = 28.26 t/s 2.01.148.186 I slot print_timing: id 5 | task 3 | prompt eval time = 2396.66 ms / 34 tokens ( 70.49 ms per token, 14.19 tokens per second) 2.01.148.202 I slot print_timing: id 5 | task 3 | eval time = 24054.74 ms / 680 tokens ( 35.37 ms per token, 28.27 tokens per second) 2.01.148.206 I slot print_timing: id 5 | task 3 | total time = 26451.40 ms / 714 tokens 2.01.148.217 I slot print_timing: id 5 | task 3 | graphs reused = 677 2.01.148.299 I slot release: id 5 | task 3 | stop processing: n_tokens = 713, truncated = 0 2.02.015.614 I slot print_timing: id 7 | task 0 | n_decoded = 714, tg = 27.84 t/s 2.02.041.674 I slot print_timing: id 6 | task 2 | n_decoded = 714, tg = 28.62 t/s 2.05.026.800 I slot print_timing: id 7 | task 0 | n_decoded = 832, tg = 29.03 t/s 2.05.050.825 I slot print_timing: id 6 | task 2 | n_decoded = 832, tg = 29.76 t/s 2.08.043.277 I slot print_timing: id 7 | task 0 | n_decoded = 948, tg = 29.93 t/s 2.08.066.873 I slot print_timing: id 6 | task 2 | n_decoded = 948, tg = 30.61 t/s 2.11.052.970 I slot print_timing: id 7 | task 0 | n_decoded = 1066, tg = 30.73 t/s 2.11.076.619 I slot print_timing: id 6 | task 2 | n_decoded = 1066, tg = 31.37 t/s 2.14.057.716 I slot print_timing: id 7 | task 0 | n_decoded = 1185, tg = 31.44 t/s 2.14.082.063 I slot print_timing: id 6 | task 2 | n_decoded = 1185, tg = 32.04 t/s 2.16.506.427 I slot print_timing: id 6 | task 2 | prompt eval time = 2398.32 ms / 33 tokens ( 72.68 ms per token, 13.76 tokens per second) 2.16.506.431 I slot print_timing: id 6 | task 2 | eval time = 39411.32 ms / 1280 tokens ( 30.79 ms per token, 32.48 tokens per second) 2.16.506.432 I slot print_timing: id 6 | task 2 | total time = 41809.64 ms / 1313 tokens 2.16.506.433 I slot print_timing: id 6 | task 2 | graphs reused = 1273 2.16.506.446 I slot release: id 6 | task 2 | stop processing: n_tokens = 1312, truncated = 0 2.17.062.598 I slot print_timing: id 7 | task 0 | n_decoded = 1314, tg = 32.29 t/s 2.18.497.615 I slot print_timing: id 7 | task 0 | prompt eval time = 3162.02 ms / 33 tokens ( 95.82 ms per token, 10.44 tokens per second) 2.18.497.630 I slot print_timing: id 7 | task 0 | eval time = 42129.72 ms / 1400 tokens ( 30.09 ms per token, 33.23 tokens per second) 2.18.497.635 I slot print_timing: id 7 | task 0 | total time = 45291.75 ms / 1433 tokens 2.18.497.638 I slot print_timing: id 7 | task 0 | graphs reused = 1391 2.18.497.686 I slot release: id 7 | task 0 | stop processing: n_tokens = 1432, truncated = 0 2.18.497.719 I srv update_slots: all slots are idle 2.19.344.323 I srv params_from_: Chat format: peg-native 2.19.344.536 I slot get_availabl: id 5 | task -1 | selected slot by LCP similarity, sim_best = 0.136 (> 0.100 thold), f_keep = 0.004 2.19.344.538 I srv get_availabl: updating prompt cache 2.19.344.818 W srv prompt_save: - saving prompt with length 713, total state size = 76.753 MiB (draft: 0.000 MiB) 2.19.439.982 I srv load: - looking for better prompt, base f_keep = 0.004, sim = 0.136 2.19.439.988 I srv update: - cache state: 1 prompts, 139.566 MiB (limits: 8192.000 MiB, 262144 tokens, 262144 est) 2.19.439.989 I srv update: - prompt 0x648aff24a1d0: 713 tokens, checkpoints: 1, 139.566 MiB 2.19.439.990 I srv get_availabl: prompt cache update took 95.45 ms 2.19.440.036 I reasoning-budget: activated, budget=2147483647 tokens 2.19.440.037 I reasoning-budget: deactivated (natural end) 2.19.440.063 I slot launch_slot_: id 5 | task 1404 | processing task, is_child = 0 2.19.440.065 I slot process_sing: id 0 | task -1 | saving idle slot to prompt cache 2.19.440.066 I slot process_sing: id 1 | task -1 | saving idle slot to prompt cache 2.19.440.066 I slot process_sing: id 2 | task -1 | saving idle slot to prompt cache 2.19.440.066 I slot process_sing: id 3 | task -1 | saving idle slot to prompt cache 2.19.440.067 I slot process_sing: id 4 | task -1 | saving idle slot to prompt cache 2.19.440.067 I slot process_sing: id 6 | task -1 | saving idle slot to prompt cache 2.19.440.296 W srv prompt_save: - saving prompt with length 1312, total state size = 88.464 MiB (draft: 0.000 MiB) 2.19.542.996 I srv update: - cache state: 2 prompts, 290.843 MiB (limits: 8192.000 MiB, 262144 tokens, 262144 est) 2.19.543.000 I srv update: - prompt 0x648aff24a1d0: 713 tokens, checkpoints: 1, 139.566 MiB 2.19.543.001 I srv update: - prompt 0x648aff2470a0: 1312 tokens, checkpoints: 1, 151.277 MiB 2.19.543.002 I slot process_sing: id 7 | task -1 | saving idle slot to prompt cache 2.19.543.233 W srv prompt_save: - saving prompt with length 1432, total state size = 90.810 MiB (draft: 0.000 MiB) 2.19.647.786 I srv update: - cache state: 3 prompts, 444.466 MiB (limits: 8192.000 MiB, 262144 tokens, 262144 est) 2.19.647.790 I srv update: - prompt 0x648aff24a1d0: 713 tokens, checkpoints: 1, 139.566 MiB 2.19.647.791 I srv update: - prompt 0x648aff2470a0: 1312 tokens, checkpoints: 1, 151.277 MiB 2.19.647.792 I srv update: - prompt 0x648aff24b750: 1432 tokens, checkpoints: 1, 153.623 MiB 2.19.647.803 I slot update_slots: id 5 | task 1404 | Checking checkpoint with [29, 29] against 3... 2.19.647.804 W slot update_slots: id 5 | task 1404 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055) 2.19.647.805 W slot update_slots: id 5 | task 1404 | erased invalidated context checkpoint (pos_min = 29, pos_max = 29, n_tokens = 30, n_swa = 0, pos_next = 0, size = 62.813 MiB) 2.19.989.405 I slot create_check: id 5 | task 1404 | created context checkpoint 1 of 32 (pos_min = 17, pos_max = 17, n_tokens = 18, size = 62.813 MiB) 2.21.697.009 I slot print_timing: id 5 | task 1404 | n_decoded = 100, tg = 60.29 t/s 2.24.702.800 I slot print_timing: id 5 | task 1404 | n_decoded = 279, tg = 59.81 t/s 2.27.704.614 I slot print_timing: id 5 | task 1404 | n_decoded = 458, tg = 59.74 t/s 2.30.718.633 I slot print_timing: id 5 | task 1404 | n_decoded = 637, tg = 59.64 t/s 2.33.731.532 I slot print_timing: id 5 | task 1404 | n_decoded = 818, tg = 59.74 t/s 2.36.742.598 I slot print_timing: id 5 | task 1404 | n_decoded = 999, tg = 59.80 t/s 2.36.970.561 I slot print_timing: id 5 | task 1404 | prompt eval time = 390.49 ms / 22 tokens ( 17.75 ms per token, 56.34 tokens per second) 2.36.970.565 I slot print_timing: id 5 | task 1404 | eval time = 16932.25 ms / 1012 tokens ( 16.73 ms per token, 59.77 tokens per second) 2.36.970.566 I slot print_timing: id 5 | task 1404 | total time = 17322.75 ms / 1034 tokens 2.36.970.566 I slot print_timing: id 5 | task 1404 | graphs reused = 2397 2.36.970.580 I slot release: id 5 | task 1404 | stop processing: n_tokens = 1033, truncated = 0 2.36.970.594 I srv update_slots: all slots are idle 3.07.804.596 I srv params_from_: Chat format: peg-native 3.07.807.070 I slot get_availabl: id 4 | task -1 | selected slot by LRU, t_last = -1 3.07.807.073 I srv get_availabl: updating prompt cache 3.07.807.087 I srv load: - looking for better prompt, base f_keep = -1.000, sim = 0.000 3.07.807.091 I srv update: - cache state: 3 prompts, 444.466 MiB (limits: 8192.000 MiB, 262144 tokens, 262144 est) 3.07.807.092 I srv update: - prompt 0x648aff24a1d0: 713 tokens, checkpoints: 1, 139.566 MiB 3.07.807.093 I srv update: - prompt 0x648aff2470a0: 1312 tokens, checkpoints: 1, 151.277 MiB 3.07.807.094 I srv update: - prompt 0x648aff24b750: 1432 tokens, checkpoints: 1, 153.623 MiB 3.07.807.094 I srv get_availabl: prompt cache update took 0.02 ms 3.07.807.137 I reasoning-budget: activated, budget=2147483647 tokens 3.07.807.138 I reasoning-budget: deactivated (natural end) 3.07.807.154 I slot launch_slot_: id 4 | task 2418 | processing task, is_child = 0 3.07.807.154 I slot process_sing: id 0 | task -1 | saving idle slot to prompt cache 3.07.807.155 I slot process_sing: id 1 | task -1 | saving idle slot to prompt cache 3.07.807.155 I slot process_sing: id 2 | task -1 | saving idle slot to prompt cache 3.07.807.155 I slot process_sing: id 3 | task -1 | saving idle slot to prompt cache 3.07.807.155 I slot process_sing: id 5 | task -1 | saving idle slot to prompt cache 3.07.807.453 W srv prompt_save: - saving prompt with length 1033, total state size = 83.009 MiB (draft: 0.000 MiB) 3.07.960.564 I srv update: - cache state: 4 prompts, 590.288 MiB (limits: 8192.000 MiB, 262144 tokens, 262144 est) 3.07.960.573 I srv update: - prompt 0x648aff24a1d0: 713 tokens, checkpoints: 1, 139.566 MiB 3.07.960.576 I srv update: - prompt 0x648aff2470a0: 1312 tokens, checkpoints: 1, 151.277 MiB 3.07.960.578 I srv update: - prompt 0x648aff24b750: 1432 tokens, checkpoints: 1, 153.623 MiB 3.07.960.579 I srv update: - prompt 0x648afefdd6f0: 1033 tokens, checkpoints: 1, 145.822 MiB 3.07.960.582 I slot process_sing: id 6 | task -1 | saving idle slot to prompt cache 3.07.961.122 W srv prompt_save: - saving prompt with length 1312, total state size = 88.464 MiB (draft: 0.000 MiB) 3.07.961.130 I srv alloc: - prompt is already in the cache, skipping 3.07.961.131 I slot process_sing: id 7 | task -1 | saving idle slot to prompt cache 3.07.961.670 W srv prompt_save: - saving prompt with length 1432, total state size = 90.810 MiB (draft: 0.000 MiB) 3.07.961.676 I srv alloc: - prompt is already in the cache, skipping 3.10.000.356 I srv params_from_: Chat format: peg-native 3.11.362.101 I srv params_from_: Chat format: peg-native 3.12.721.534 I srv params_from_: Chat format: peg-native 3.15.096.173 I slot get_availabl: id 3 | task -1 | selected slot by LRU, t_last = -1 3.15.096.176 I srv get_availabl: updating prompt cache 3.15.096.179 I srv load: - looking for better prompt, base f_keep = -1.000, sim = 0.000 3.15.096.200 I srv update: - cache state: 4 prompts, 590.288 MiB (limits: 8192.000 MiB, 262144 tokens, 262144 est) 3.15.096.201 I srv update: - prompt 0x648aff24a1d0: 713 tokens, checkpoints: 1, 139.566 MiB 3.15.096.202 I srv update: - prompt 0x648aff2470a0: 1312 tokens, checkpoints: 1, 151.277 MiB 3.15.096.203 I srv update: - prompt 0x648aff24b750: 1432 tokens, checkpoints: 1, 153.623 MiB 3.15.096.204 I srv update: - prompt 0x648afefdd6f0: 1033 tokens, checkpoints: 1, 145.822 MiB 3.15.096.204 I srv get_availabl: prompt cache update took 0.03 ms 3.15.096.272 I reasoning-budget: activated, budget=2147483647 tokens 3.15.096.273 I reasoning-budget: deactivated (natural end) 3.15.096.285 I slot launch_slot_: id 3 | task 2420 | processing task, is_child = 0 3.15.096.286 I slot process_sing: id 0 | task -1 | saving idle slot to prompt cache 3.15.096.286 I slot process_sing: id 1 | task -1 | saving idle slot to prompt cache 3.15.096.287 I slot process_sing: id 2 | task -1 | saving idle slot to prompt cache 3.15.096.287 I slot process_sing: id 5 | task -1 | saving idle slot to prompt cache 3.15.096.568 W srv prompt_save: - saving prompt with length 1033, total state size = 83.009 MiB (draft: 0.000 MiB) 3.15.096.571 I srv alloc: - prompt is already in the cache, skipping 3.15.096.572 I slot process_sing: id 6 | task -1 | saving idle slot to prompt cache 3.15.096.815 W srv prompt_save: - saving prompt with length 1312, total state size = 88.464 MiB (draft: 0.000 MiB) 3.15.096.818 I srv alloc: - prompt is already in the cache, skipping 3.15.096.818 I slot process_sing: id 7 | task -1 | saving idle slot to prompt cache 3.15.097.050 W srv prompt_save: - saving prompt with length 1432, total state size = 90.810 MiB (draft: 0.000 MiB) 3.15.097.053 I srv alloc: - prompt is already in the cache, skipping 3.15.097.054 I slot get_availabl: id 2 | task -1 | selected slot by LRU, t_last = -1 3.15.097.054 I srv get_availabl: updating prompt cache 3.15.097.070 I srv load: - looking for better prompt, base f_keep = -1.000, sim = 0.000 3.15.097.072 I srv update: - cache state: 4 prompts, 590.288 MiB (limits: 8192.000 MiB, 262144 tokens, 262144 est) 3.15.097.073 I srv update: - prompt 0x648aff24a1d0: 713 tokens, checkpoints: 1, 139.566 MiB 3.15.097.073 I srv update: - prompt 0x648aff2470a0: 1312 tokens, checkpoints: 1, 151.277 MiB 3.15.097.074 I srv update: - prompt 0x648aff24b750: 1432 tokens, checkpoints: 1, 153.623 MiB 3.15.097.075 I srv update: - prompt 0x648afefdd6f0: 1033 tokens, checkpoints: 1, 145.822 MiB 3.15.097.076 I srv get_availabl: prompt cache update took 0.02 ms 3.15.097.097 I reasoning-budget: activated, budget=2147483647 tokens 3.15.097.097 I reasoning-budget: deactivated (natural end) 3.15.097.114 I slot launch_slot_: id 2 | task 2421 | processing task, is_child = 0 3.15.097.114 I slot process_sing: id 0 | task -1 | saving idle slot to prompt cache 3.15.097.114 I slot process_sing: id 1 | task -1 | saving idle slot to prompt cache 3.15.097.115 I slot process_sing: id 5 | task -1 | saving idle slot to prompt cache 3.15.097.391 W srv prompt_save: - saving prompt with length 1033, total state size = 83.009 MiB (draft: 0.000 MiB) 3.15.097.394 I srv alloc: - prompt is already in the cache, skipping 3.15.097.394 I slot process_sing: id 6 | task -1 | saving idle slot to prompt cache 3.15.097.610 W srv prompt_save: - saving prompt with length 1312, total state size = 88.464 MiB (draft: 0.000 MiB) 3.15.097.612 I srv alloc: - prompt is already in the cache, skipping 3.15.097.612 I slot process_sing: id 7 | task -1 | saving idle slot to prompt cache 3.15.097.854 W srv prompt_save: - saving prompt with length 1432, total state size = 90.810 MiB (draft: 0.000 MiB) 3.15.097.856 I srv alloc: - prompt is already in the cache, skipping 3.15.097.858 I slot get_availabl: id 1 | task -1 | selected slot by LRU, t_last = -1 3.15.097.858 I srv get_availabl: updating prompt cache 3.15.097.873 I srv load: - looking for better prompt, base f_keep = -1.000, sim = 0.000 3.15.097.875 I srv update: - cache state: 4 prompts, 590.288 MiB (limits: 8192.000 MiB, 262144 tokens, 262144 est) 3.15.097.876 I srv update: - prompt 0x648aff24a1d0: 713 tokens, checkpoints: 1, 139.566 MiB 3.15.097.876 I srv update: - prompt 0x648aff2470a0: 1312 tokens, checkpoints: 1, 151.277 MiB 3.15.097.877 I srv update: - prompt 0x648aff24b750: 1432 tokens, checkpoints: 1, 153.623 MiB 3.15.097.878 I srv update: - prompt 0x648afefdd6f0: 1033 tokens, checkpoints: 1, 145.822 MiB 3.15.097.878 I srv get_availabl: prompt cache update took 0.02 ms 3.15.097.894 I reasoning-budget: activated, budget=2147483647 tokens 3.15.097.894 I reasoning-budget: deactivated (natural end) 3.15.097.901 I slot launch_slot_: id 1 | task 2422 | processing task, is_child = 0 3.15.097.901 I slot process_sing: id 0 | task -1 | saving idle slot to prompt cache 3.15.097.902 I slot process_sing: id 5 | task -1 | saving idle slot to prompt cache 3.15.098.156 W srv prompt_save: - saving prompt with length 1033, total state size = 83.009 MiB (draft: 0.000 MiB) 3.15.098.158 I srv alloc: - prompt is already in the cache, skipping 3.15.098.159 I slot process_sing: id 6 | task -1 | saving idle slot to prompt cache 3.15.098.362 W srv prompt_save: - saving prompt with length 1312, total state size = 88.464 MiB (draft: 0.000 MiB) 3.15.098.383 I srv alloc: - prompt is already in the cache, skipping 3.15.098.383 I slot process_sing: id 7 | task -1 | saving idle slot to prompt cache 3.15.098.613 W srv prompt_save: - saving prompt with length 1432, total state size = 90.810 MiB (draft: 0.000 MiB) 3.15.098.615 I srv alloc: - prompt is already in the cache, skipping 3.15.098.810 I slot print_timing: id 4 | task 2418 | prompt processing, n_tokens = 817, progress = 0.61, t = 7.14 s / 114.47 tokens per second 3.23.731.318 I slot print_timing: id 1 | task 2422 | prompt processing, n_tokens = 218, progress = 0.30, t = 8.63 s / 25.25 tokens per second 3.23.731.376 I slot print_timing: id 2 | task 2421 | prompt processing, n_tokens = 537, progress = 0.51, t = 8.63 s / 62.21 tokens per second 3.23.731.418 I slot print_timing: id 3 | task 2420 | prompt processing, n_tokens = 936, progress = 0.64, t = 8.63 s / 108.43 tokens per second 3.23.731.472 I slot print_timing: id 4 | task 2418 | prompt processing, n_tokens = 1174, progress = 0.88, t = 15.77 s / 74.45 tokens per second 3.30.615.240 I slot print_timing: id 1 | task 2422 | prompt processing, n_tokens = 711, progress = 0.97, t = 15.52 s / 45.82 tokens per second 3.31.453.006 I slot create_check: id 1 | task 2422 | created context checkpoint 1 of 32 (pos_min = 710, pos_max = 710, n_tokens = 711, size = 62.813 MiB) 3.31.453.012 I slot print_timing: id 2 | task 2421 | prompt processing, n_tokens = 1030, progress = 0.98, t = 16.35 s / 62.98 tokens per second 3.31.503.501 I slot create_check: id 2 | task 2421 | created context checkpoint 1 of 32 (pos_min = 1029, pos_max = 1029, n_tokens = 1030, size = 62.813 MiB) 3.31.503.507 I slot print_timing: id 3 | task 2420 | prompt processing, n_tokens = 1428, progress = 0.98, t = 16.40 s / 87.05 tokens per second 3.31.554.228 I slot create_check: id 3 | task 2420 | created context checkpoint 1 of 32 (pos_min = 1427, pos_max = 1427, n_tokens = 1428, size = 62.813 MiB) 3.31.554.234 I slot print_timing: id 4 | task 2418 | prompt processing, n_tokens = 1310, progress = 0.98, t = 23.59 s / 55.53 tokens per second 3.31.604.973 I slot create_check: id 4 | task 2418 | created context checkpoint 1 of 32 (pos_min = 1309, pos_max = 1309, n_tokens = 1310, size = 62.813 MiB) 3.32.350.167 I slot print_timing: id 1 | task 2422 | prompt processing, n_tokens = 730, progress = 0.99, t = 17.25 s / 42.32 tokens per second 3.32.350.287 I slot print_timing: id 2 | task 2421 | prompt processing, n_tokens = 1049, progress = 1.00, t = 17.25 s / 60.81 tokens per second 3.32.350.446 I slot print_timing: id 3 | task 2420 | prompt processing, n_tokens = 1448, progress = 1.00, t = 17.25 s / 83.93 tokens per second 3.32.350.632 I slot print_timing: id 4 | task 2418 | prompt processing, n_tokens = 1329, progress = 1.00, t = 24.39 s / 54.49 tokens per second 3.37.375.672 I slot print_timing: id 1 | task 2422 | n_decoded = 100, tg = 23.32 t/s 3.37.377.034 I slot print_timing: id 2 | task 2421 | n_decoded = 100, tg = 23.32 t/s 3.37.378.398 I slot print_timing: id 3 | task 2420 | n_decoded = 100, tg = 23.33 t/s 3.37.379.825 I slot print_timing: id 4 | task 2418 | n_decoded = 100, tg = 23.33 t/s 3.40.397.975 I slot print_timing: id 1 | task 2422 | n_decoded = 171, tg = 23.39 t/s 3.40.398.611 I slot print_timing: id 2 | task 2421 | n_decoded = 171, tg = 23.40 t/s 3.40.399.236 I slot print_timing: id 3 | task 2420 | n_decoded = 171, tg = 23.40 t/s 3.40.399.864 I slot print_timing: id 4 | task 2418 | n_decoded = 171, tg = 23.40 t/s 3.43.403.430 I slot print_timing: id 1 | task 2422 | n_decoded = 243, tg = 23.56 t/s 3.43.404.042 I slot print_timing: id 2 | task 2421 | n_decoded = 243, tg = 23.56 t/s 3.43.404.654 I slot print_timing: id 3 | task 2420 | n_decoded = 243, tg = 23.56 t/s 3.43.405.259 I slot print_timing: id 4 | task 2418 | n_decoded = 243, tg = 23.57 t/s 3.46.427.117 I slot print_timing: id 1 | task 2422 | n_decoded = 316, tg = 23.69 t/s 3.46.427.750 I slot print_timing: id 2 | task 2421 | n_decoded = 316, tg = 23.69 t/s 3.46.428.383 I slot print_timing: id 3 | task 2420 | n_decoded = 316, tg = 23.69 t/s 3.46.429.014 I slot print_timing: id 4 | task 2418 | n_decoded = 316, tg = 23.70 t/s 3.49.443.255 I slot print_timing: id 1 | task 2422 | n_decoded = 389, tg = 23.78 t/s 3.49.443.911 I slot print_timing: id 2 | task 2421 | n_decoded = 389, tg = 23.79 t/s 3.49.444.560 I slot print_timing: id 3 | task 2420 | n_decoded = 389, tg = 23.79 t/s 3.49.445.210 I slot print_timing: id 4 | task 2418 | n_decoded = 389, tg = 23.79 t/s 3.52.451.340 I slot print_timing: id 1 | task 2422 | n_decoded = 462, tg = 23.86 t/s 3.52.451.973 I slot print_timing: id 2 | task 2421 | n_decoded = 462, tg = 23.86 t/s 3.52.452.605 I slot print_timing: id 3 | task 2420 | n_decoded = 462, tg = 23.86 t/s 3.52.453.238 I slot print_timing: id 4 | task 2418 | n_decoded = 462, tg = 23.86 t/s 3.53.525.843 I slot print_timing: id 1 | task 2422 | prompt eval time = 17989.33 ms / 734 tokens ( 24.51 ms per token, 40.80 tokens per second) 3.53.525.847 I slot print_timing: id 1 | task 2422 | eval time = 20437.87 ms / 488 tokens ( 41.88 ms per token, 23.88 tokens per second) 3.53.525.848 I slot print_timing: id 1 | task 2422 | total time = 38427.19 ms / 1222 tokens 3.53.525.849 I slot print_timing: id 1 | task 2422 | graphs reused = 2881 3.53.525.902 I slot release: id 1 | task 2422 | stop processing: n_tokens = 1221, truncated = 0 3.53.865.447 I slot print_timing: id 4 | task 2418 | prompt eval time = 25131.75 ms / 1333 tokens ( 18.85 ms per token, 53.04 tokens per second) 3.53.865.451 I slot print_timing: id 4 | task 2418 | eval time = 20772.00 ms / 498 tokens ( 41.71 ms per token, 23.97 tokens per second) 3.53.865.452 I slot print_timing: id 4 | task 2418 | total time = 45903.75 ms / 1831 tokens 3.53.865.452 I slot print_timing: id 4 | task 2418 | graphs reused = 2890 3.53.865.567 I slot release: id 4 | task 2418 | stop processing: n_tokens = 1830, truncated = 0 3.55.473.255 I slot print_timing: id 2 | task 2421 | n_decoded = 561, tg = 25.06 t/s 3.55.474.053 I slot print_timing: id 3 | task 2420 | n_decoded = 561, tg = 25.06 t/s 3.58.477.121 I slot print_timing: id 2 | task 2421 | n_decoded = 681, tg = 26.82 t/s 3.58.477.702 I slot print_timing: id 3 | task 2420 | n_decoded = 681, tg = 26.83 t/s 4.01.479.346 I slot print_timing: id 2 | task 2421 | n_decoded = 801, tg = 28.21 t/s 4.01.480.148 I slot print_timing: id 3 | task 2420 | n_decoded = 801, tg = 28.22 t/s 4.04.490.872 I slot print_timing: id 2 | task 2421 | n_decoded = 921, tg = 29.33 t/s 4.04.491.520 I slot print_timing: id 3 | task 2420 | n_decoded = 921, tg = 29.33 t/s 4.05.363.400 I slot print_timing: id 2 | task 2421 | prompt eval time = 17991.09 ms / 1053 tokens ( 17.09 ms per token, 58.53 tokens per second) 4.05.363.404 I slot print_timing: id 2 | task 2421 | eval time = 32273.60 ms / 956 tokens ( 33.76 ms per token, 29.62 tokens per second) 4.05.363.405 I slot print_timing: id 2 | task 2421 | total time = 50264.69 ms / 2009 tokens 4.05.363.406 I slot print_timing: id 2 | task 2421 | graphs reused = 3345 4.05.363.473 I slot release: id 2 | task 2421 | stop processing: n_tokens = 2008, truncated = 0 4.07.498.689 I slot print_timing: id 3 | task 2420 | n_decoded = 1086, tg = 31.56 t/s 4.10.510.974 I slot print_timing: id 3 | task 2420 | n_decoded = 1271, tg = 33.97 t/s 4.11.128.366 I slot print_timing: id 3 | task 2420 | prompt eval time = 17992.87 ms / 1452 tokens ( 12.39 ms per token, 80.70 tokens per second) 4.11.128.369 I slot print_timing: id 3 | task 2420 | eval time = 38036.73 ms / 1309 tokens ( 29.06 ms per token, 34.41 tokens per second) 4.11.128.370 I slot print_timing: id 3 | task 2420 | total time = 56029.60 ms / 2761 tokens 4.11.128.371 I slot print_timing: id 3 | task 2420 | graphs reused = 3696 4.11.128.488 I slot release: id 3 | task 2420 | stop processing: n_tokens = 2760, truncated = 0 4.11.128.498 I srv update_slots: all slots are idle |
+
최대 컨텍스트는 유지되고, 개인당 컨텍스트는 무조건 줄어드는건가?
| $ ./llama-b9692/llama-server -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -sm layer --host 0.0.0.0 -np 1 0.08.367.856 W common_speculative_init: no implementations specified for speculative decoding 0.08.367.860 I slot load_model: id 0 | task -1 | new slot, n_ctx = 262144 0.08.368.336 I srv load_model: prompt cache is enabled, size limit: 8192 MiB |
256k ctx len을 4이서 나눠서 64k씩
| $ ./llama-b9692/llama-server -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -sm layer --host 0.0.0.0 -np 4 0.08.236.351 W common_speculative_init: no implementations specified for speculative decoding 0.08.236.356 I slot load_model: id 0 | task -1 | new slot, n_ctx = 65536 0.08.236.370 I slot load_model: id 1 | task -1 | new slot, n_ctx = 65536 0.08.236.372 I slot load_model: id 2 | task -1 | new slot, n_ctx = 65536 0.08.236.374 I slot load_model: id 3 | task -1 | new slot, n_ctx = 65536 0.08.236.795 I srv load_model: prompt cache is enabled, size limit: 8192 MiB |
그걸 다시 ctx len을 4배 늘려서 하니 인당 256k 씩 할당 완료
| $ ./llama-b9692/llama-server -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -sm layer --host 0.0.0.0 -np 4 -c 1048576 0.41.525.240 W common_speculative_init: no implementations specified for speculative decoding 0.41.525.241 I slot load_model: id 0 | task -1 | new slot, n_ctx = 262144 0.41.525.247 I slot load_model: id 1 | task -1 | new slot, n_ctx = 262144 0.41.525.247 I slot load_model: id 2 | task -1 | new slot, n_ctx = 262144 0.41.525.248 I slot load_model: id 3 | task -1 | new slot, n_ctx = 262144 0.41.525.326 I srv load_model: prompt cache is enabled, size limit: 8192 MiB |
| $ ./llama-b9692/llama-server -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -sm layer --host 0.0.0.0 -np 8 0.08.144.543 W common_speculative_init: no implementations specified for speculative decoding 0.08.144.547 I slot load_model: id 0 | task -1 | new slot, n_ctx = 32768 0.08.144.563 I slot load_model: id 1 | task -1 | new slot, n_ctx = 32768 0.08.144.565 I slot load_model: id 2 | task -1 | new slot, n_ctx = 32768 0.08.144.566 I slot load_model: id 3 | task -1 | new slot, n_ctx = 32768 0.08.144.568 I slot load_model: id 4 | task -1 | new slot, n_ctx = 32768 0.08.144.569 I slot load_model: id 5 | task -1 | new slot, n_ctx = 32768 0.08.144.571 I slot load_model: id 6 | task -1 | new slot, n_ctx = 32768 0.08.144.573 I slot load_model: id 7 | task -1 | new slot, n_ctx = 32768 0.08.145.051 I srv load_model: prompt cache is enabled, size limit: 8192 MiB |
32*7168 =229376 < 262144
음? 최대 컨텍스트 보다 적어졌다?
| $ ./llama-b9692/llama-server -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -sm layer --host 0.0.0.0 -np 32 0.20.248.497 W common_speculative_init: no implementations specified for speculative decoding 0.20.248.508 I slot load_model: id 0 | task -1 | new slot, n_ctx = 7168 0.20.248.523 I slot load_model: id 1 | task -1 | new slot, n_ctx = 7168 0.20.248.525 I slot load_model: id 2 | task -1 | new slot, n_ctx = 7168 0.20.248.527 I slot load_model: id 3 | task -1 | new slot, n_ctx = 7168 0.20.248.528 I slot load_model: id 4 | task -1 | new slot, n_ctx = 7168 0.20.248.530 I slot load_model: id 5 | task -1 | new slot, n_ctx = 7168 0.20.248.532 I slot load_model: id 6 | task -1 | new slot, n_ctx = 7168 0.20.248.533 I slot load_model: id 7 | task -1 | new slot, n_ctx = 7168 0.20.248.535 I slot load_model: id 8 | task -1 | new slot, n_ctx = 7168 0.20.248.536 I slot load_model: id 9 | task -1 | new slot, n_ctx = 7168 0.20.248.538 I slot load_model: id 10 | task -1 | new slot, n_ctx = 7168 0.20.248.539 I slot load_model: id 11 | task -1 | new slot, n_ctx = 7168 0.20.248.540 I slot load_model: id 12 | task -1 | new slot, n_ctx = 7168 0.20.248.542 I slot load_model: id 13 | task -1 | new slot, n_ctx = 7168 0.20.248.543 I slot load_model: id 14 | task -1 | new slot, n_ctx = 7168 0.20.248.545 I slot load_model: id 15 | task -1 | new slot, n_ctx = 7168 0.20.248.546 I slot load_model: id 16 | task -1 | new slot, n_ctx = 7168 0.20.248.548 I slot load_model: id 17 | task -1 | new slot, n_ctx = 7168 0.20.248.549 I slot load_model: id 18 | task -1 | new slot, n_ctx = 7168 0.20.248.551 I slot load_model: id 19 | task -1 | new slot, n_ctx = 7168 0.20.248.552 I slot load_model: id 20 | task -1 | new slot, n_ctx = 7168 0.20.248.553 I slot load_model: id 21 | task -1 | new slot, n_ctx = 7168 0.20.248.555 I slot load_model: id 22 | task -1 | new slot, n_ctx = 7168 0.20.248.556 I slot load_model: id 23 | task -1 | new slot, n_ctx = 7168 0.20.248.558 I slot load_model: id 24 | task -1 | new slot, n_ctx = 7168 0.20.248.559 I slot load_model: id 25 | task -1 | new slot, n_ctx = 7168 0.20.248.561 I slot load_model: id 26 | task -1 | new slot, n_ctx = 7168 0.20.248.562 I slot load_model: id 27 | task -1 | new slot, n_ctx = 7168 0.20.248.563 I slot load_model: id 28 | task -1 | new slot, n_ctx = 7168 0.20.248.565 I slot load_model: id 29 | task -1 | new slot, n_ctx = 7168 0.20.248.566 I slot load_model: id 30 | task -1 | new slot, n_ctx = 7168 0.20.248.568 I slot load_model: id 31 | task -1 | new slot, n_ctx = 7168 0.20.248.958 I srv load_model: prompt cache is enabled, size limit: 8192 MiB |
-c로 개인당 컨텍스트 늘리는데 예상한것 보다는 적어도 아까보단 늘어났다.
| $ ./llama-b9692/llama-server -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -sm layer --host 0.0.0.0 -np 32 -c 1048576 0.29.550.307 W common_speculative_init: no implementations specified for speculative decoding 0.29.550.308 I slot load_model: id 0 | task -1 | new slot, n_ctx = 32768 0.29.550.314 I slot load_model: id 1 | task -1 | new slot, n_ctx = 32768 0.29.550.315 I slot load_model: id 2 | task -1 | new slot, n_ctx = 32768 0.29.550.315 I slot load_model: id 3 | task -1 | new slot, n_ctx = 32768 0.29.550.315 I slot load_model: id 4 | task -1 | new slot, n_ctx = 32768 0.29.550.316 I slot load_model: id 5 | task -1 | new slot, n_ctx = 32768 0.29.550.316 I slot load_model: id 6 | task -1 | new slot, n_ctx = 32768 0.29.550.316 I slot load_model: id 7 | task -1 | new slot, n_ctx = 32768 0.29.550.316 I slot load_model: id 8 | task -1 | new slot, n_ctx = 32768 0.29.550.317 I slot load_model: id 9 | task -1 | new slot, n_ctx = 32768 0.29.550.317 I slot load_model: id 10 | task -1 | new slot, n_ctx = 32768 0.29.550.317 I slot load_model: id 11 | task -1 | new slot, n_ctx = 32768 0.29.550.318 I slot load_model: id 12 | task -1 | new slot, n_ctx = 32768 0.29.550.318 I slot load_model: id 13 | task -1 | new slot, n_ctx = 32768 0.29.550.318 I slot load_model: id 14 | task -1 | new slot, n_ctx = 32768 0.29.550.318 I slot load_model: id 15 | task -1 | new slot, n_ctx = 32768 0.29.550.319 I slot load_model: id 16 | task -1 | new slot, n_ctx = 32768 0.29.550.319 I slot load_model: id 17 | task -1 | new slot, n_ctx = 32768 0.29.550.319 I slot load_model: id 18 | task -1 | new slot, n_ctx = 32768 0.29.550.320 I slot load_model: id 19 | task -1 | new slot, n_ctx = 32768 0.29.550.320 I slot load_model: id 20 | task -1 | new slot, n_ctx = 32768 0.29.550.320 I slot load_model: id 21 | task -1 | new slot, n_ctx = 32768 0.29.550.321 I slot load_model: id 22 | task -1 | new slot, n_ctx = 32768 0.29.550.321 I slot load_model: id 23 | task -1 | new slot, n_ctx = 32768 0.29.550.321 I slot load_model: id 24 | task -1 | new slot, n_ctx = 32768 0.29.550.321 I slot load_model: id 25 | task -1 | new slot, n_ctx = 32768 0.29.550.322 I slot load_model: id 26 | task -1 | new slot, n_ctx = 32768 0.29.550.322 I slot load_model: id 27 | task -1 | new slot, n_ctx = 32768 0.29.550.322 I slot load_model: id 28 | task -1 | new slot, n_ctx = 32768 0.29.550.323 I slot load_model: id 29 | task -1 | new slot, n_ctx = 32768 0.29.550.323 I slot load_model: id 30 | task -1 | new slot, n_ctx = 32768 0.29.550.323 I slot load_model: id 31 | task -1 | new slot, n_ctx = 32768 0.29.550.438 I srv load_model: prompt cache is enabled, size limit: 8192 MiB |
여기 부터는 아마도 vram 초과해서 cpu ram으로 offloading된 듯
| $ ./llama-b9692/llama-server -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -sm layer --host 0.0.0.0 -np 64 0.21.952.543 W common_speculative_init: no implementations specified for speculative decoding 0.21.952.553 I slot load_model: id 0 | task -1 | new slot, n_ctx = 256 0.21.952.568 I slot load_model: id 1 | task -1 | new slot, n_ctx = 256 0.21.952.570 I slot load_model: id 2 | task -1 | new slot, n_ctx = 256 0.21.952.571 I slot load_model: id 3 | task -1 | new slot, n_ctx = 256 0.21.952.573 I slot load_model: id 4 | task -1 | new slot, n_ctx = 256 0.21.952.574 I slot load_model: id 5 | task -1 | new slot, n_ctx = 256 0.21.952.576 I slot load_model: id 6 | task -1 | new slot, n_ctx = 256 0.21.952.578 I slot load_model: id 7 | task -1 | new slot, n_ctx = 256 0.21.952.579 I slot load_model: id 8 | task -1 | new slot, n_ctx = 256 0.21.952.581 I slot load_model: id 9 | task -1 | new slot, n_ctx = 256 0.21.952.582 I slot load_model: id 10 | task -1 | new slot, n_ctx = 256 0.21.952.584 I slot load_model: id 11 | task -1 | new slot, n_ctx = 256 0.21.952.585 I slot load_model: id 12 | task -1 | new slot, n_ctx = 256 0.21.952.587 I slot load_model: id 13 | task -1 | new slot, n_ctx = 256 0.21.952.589 I slot load_model: id 14 | task -1 | new slot, n_ctx = 256 0.21.952.590 I slot load_model: id 15 | task -1 | new slot, n_ctx = 256 0.21.952.591 I slot load_model: id 16 | task -1 | new slot, n_ctx = 256 0.21.952.593 I slot load_model: id 17 | task -1 | new slot, n_ctx = 256 0.21.952.595 I slot load_model: id 18 | task -1 | new slot, n_ctx = 256 0.21.952.596 I slot load_model: id 19 | task -1 | new slot, n_ctx = 256 0.21.952.597 I slot load_model: id 20 | task -1 | new slot, n_ctx = 256 0.21.952.599 I slot load_model: id 21 | task -1 | new slot, n_ctx = 256 0.21.952.600 I slot load_model: id 22 | task -1 | new slot, n_ctx = 256 0.21.952.602 I slot load_model: id 23 | task -1 | new slot, n_ctx = 256 0.21.952.603 I slot load_model: id 24 | task -1 | new slot, n_ctx = 256 0.21.952.605 I slot load_model: id 25 | task -1 | new slot, n_ctx = 256 0.21.952.611 I slot load_model: id 26 | task -1 | new slot, n_ctx = 256 0.21.952.616 I slot load_model: id 27 | task -1 | new slot, n_ctx = 256 0.21.952.617 I slot load_model: id 28 | task -1 | new slot, n_ctx = 256 0.21.952.619 I slot load_model: id 29 | task -1 | new slot, n_ctx = 256 0.21.952.620 I slot load_model: id 30 | task -1 | new slot, n_ctx = 256 0.21.952.622 I slot load_model: id 31 | task -1 | new slot, n_ctx = 256 0.21.952.623 I slot load_model: id 32 | task -1 | new slot, n_ctx = 256 0.21.952.625 I slot load_model: id 33 | task -1 | new slot, n_ctx = 256 0.21.952.626 I slot load_model: id 34 | task -1 | new slot, n_ctx = 256 0.21.952.627 I slot load_model: id 35 | task -1 | new slot, n_ctx = 256 0.21.952.629 I slot load_model: id 36 | task -1 | new slot, n_ctx = 256 0.21.952.630 I slot load_model: id 37 | task -1 | new slot, n_ctx = 256 0.21.952.632 I slot load_model: id 38 | task -1 | new slot, n_ctx = 256 0.21.952.633 I slot load_model: id 39 | task -1 | new slot, n_ctx = 256 0.21.952.635 I slot load_model: id 40 | task -1 | new slot, n_ctx = 256 0.21.952.636 I slot load_model: id 41 | task -1 | new slot, n_ctx = 256 0.21.952.638 I slot load_model: id 42 | task -1 | new slot, n_ctx = 256 0.21.952.639 I slot load_model: id 43 | task -1 | new slot, n_ctx = 256 0.21.952.640 I slot load_model: id 44 | task -1 | new slot, n_ctx = 256 0.21.952.642 I slot load_model: id 45 | task -1 | new slot, n_ctx = 256 0.21.952.643 I slot load_model: id 46 | task -1 | new slot, n_ctx = 256 0.21.952.645 I slot load_model: id 47 | task -1 | new slot, n_ctx = 256 0.21.952.646 I slot load_model: id 48 | task -1 | new slot, n_ctx = 256 0.21.952.648 I slot load_model: id 49 | task -1 | new slot, n_ctx = 256 0.21.952.649 I slot load_model: id 50 | task -1 | new slot, n_ctx = 256 0.21.952.650 I slot load_model: id 51 | task -1 | new slot, n_ctx = 256 0.21.952.652 I slot load_model: id 52 | task -1 | new slot, n_ctx = 256 0.21.952.653 I slot load_model: id 53 | task -1 | new slot, n_ctx = 256 0.21.952.655 I slot load_model: id 54 | task -1 | new slot, n_ctx = 256 0.21.952.656 I slot load_model: id 55 | task -1 | new slot, n_ctx = 256 0.21.952.658 I slot load_model: id 56 | task -1 | new slot, n_ctx = 256 0.21.952.659 I slot load_model: id 57 | task -1 | new slot, n_ctx = 256 0.21.952.661 I slot load_model: id 58 | task -1 | new slot, n_ctx = 256 0.21.952.662 I slot load_model: id 59 | task -1 | new slot, n_ctx = 256 0.21.952.663 I slot load_model: id 60 | task -1 | new slot, n_ctx = 256 0.21.952.665 I slot load_model: id 61 | task -1 | new slot, n_ctx = 256 0.21.952.666 I slot load_model: id 62 | task -1 | new slot, n_ctx = 256 0.21.952.668 I slot load_model: id 63 | task -1 | new slot, n_ctx = 256 0.21.953.026 I srv load_model: prompt cache is enabled, size limit: 8192 MiB |
| $ ./llama-b9692/llama-server -m model/qwen3.6_35B/Qwen3.6-35B-A3B-UD-Q2_K_XL.gguf -sm layer --host 0.0.0.0 -np 128 0.43.993.646 W common_speculative_init: no implementations specified for speculative decoding 0.43.993.651 I slot load_model: id 0 | task -1 | new slot, n_ctx = 256 0.43.993.666 I slot load_model: id 1 | task -1 | new slot, n_ctx = 256 0.43.993.669 I slot load_model: id 2 | task -1 | new slot, n_ctx = 256 0.43.993.670 I slot load_model: id 3 | task -1 | new slot, n_ctx = 256 0.43.993.672 I slot load_model: id 4 | task -1 | new slot, n_ctx = 256 0.43.993.673 I slot load_model: id 5 | task -1 | new slot, n_ctx = 256 0.43.993.675 I slot load_model: id 6 | task -1 | new slot, n_ctx = 256 0.43.993.676 I slot load_model: id 7 | task -1 | new slot, n_ctx = 256 0.43.993.678 I slot load_model: id 8 | task -1 | new slot, n_ctx = 256 0.43.993.679 I slot load_model: id 9 | task -1 | new slot, n_ctx = 256 0.43.993.681 I slot load_model: id 10 | task -1 | new slot, n_ctx = 256 0.43.993.682 I slot load_model: id 11 | task -1 | new slot, n_ctx = 256 0.43.993.684 I slot load_model: id 12 | task -1 | new slot, n_ctx = 256 0.43.993.685 I slot load_model: id 13 | task -1 | new slot, n_ctx = 256 0.43.993.687 I slot load_model: id 14 | task -1 | new slot, n_ctx = 256 0.43.993.688 I slot load_model: id 15 | task -1 | new slot, n_ctx = 256 0.43.993.689 I slot load_model: id 16 | task -1 | new slot, n_ctx = 256 0.43.993.691 I slot load_model: id 17 | task -1 | new slot, n_ctx = 256 0.43.993.692 I slot load_model: id 18 | task -1 | new slot, n_ctx = 256 0.43.993.694 I slot load_model: id 19 | task -1 | new slot, n_ctx = 256 0.43.993.695 I slot load_model: id 20 | task -1 | new slot, n_ctx = 256 0.43.993.697 I slot load_model: id 21 | task -1 | new slot, n_ctx = 256 0.43.993.698 I slot load_model: id 22 | task -1 | new slot, n_ctx = 256 0.43.993.700 I slot load_model: id 23 | task -1 | new slot, n_ctx = 256 0.43.993.701 I slot load_model: id 24 | task -1 | new slot, n_ctx = 256 0.43.993.703 I slot load_model: id 25 | task -1 | new slot, n_ctx = 256 0.43.993.704 I slot load_model: id 26 | task -1 | new slot, n_ctx = 256 0.43.993.705 I slot load_model: id 27 | task -1 | new slot, n_ctx = 256 0.43.993.707 I slot load_model: id 28 | task -1 | new slot, n_ctx = 256 0.43.993.708 I slot load_model: id 29 | task -1 | new slot, n_ctx = 256 0.43.993.710 I slot load_model: id 30 | task -1 | new slot, n_ctx = 256 0.43.993.711 I slot load_model: id 31 | task -1 | new slot, n_ctx = 256 0.43.993.714 I slot load_model: id 32 | task -1 | new slot, n_ctx = 256 0.43.993.715 I slot load_model: id 33 | task -1 | new slot, n_ctx = 256 0.43.993.717 I slot load_model: id 34 | task -1 | new slot, n_ctx = 256 0.43.993.719 I slot load_model: id 35 | task -1 | new slot, n_ctx = 256 0.43.993.720 I slot load_model: id 36 | task -1 | new slot, n_ctx = 256 0.43.993.722 I slot load_model: id 37 | task -1 | new slot, n_ctx = 256 0.43.993.723 I slot load_model: id 38 | task -1 | new slot, n_ctx = 256 0.43.993.724 I slot load_model: id 39 | task -1 | new slot, n_ctx = 256 0.43.993.726 I slot load_model: id 40 | task -1 | new slot, n_ctx = 256 0.43.993.727 I slot load_model: id 41 | task -1 | new slot, n_ctx = 256 0.43.993.729 I slot load_model: id 42 | task -1 | new slot, n_ctx = 256 0.43.993.730 I slot load_model: id 43 | task -1 | new slot, n_ctx = 256 0.43.993.731 I slot load_model: id 44 | task -1 | new slot, n_ctx = 256 0.43.993.733 I slot load_model: id 45 | task -1 | new slot, n_ctx = 256 0.43.993.734 I slot load_model: id 46 | task -1 | new slot, n_ctx = 256 0.43.993.736 I slot load_model: id 47 | task -1 | new slot, n_ctx = 256 0.43.993.737 I slot load_model: id 48 | task -1 | new slot, n_ctx = 256 0.43.993.739 I slot load_model: id 49 | task -1 | new slot, n_ctx = 256 0.43.993.740 I slot load_model: id 50 | task -1 | new slot, n_ctx = 256 0.43.993.742 I slot load_model: id 51 | task -1 | new slot, n_ctx = 256 0.43.993.743 I slot load_model: id 52 | task -1 | new slot, n_ctx = 256 0.43.993.745 I slot load_model: id 53 | task -1 | new slot, n_ctx = 256 0.43.993.746 I slot load_model: id 54 | task -1 | new slot, n_ctx = 256 0.43.993.748 I slot load_model: id 55 | task -1 | new slot, n_ctx = 256 0.43.993.749 I slot load_model: id 56 | task -1 | new slot, n_ctx = 256 0.43.993.750 I slot load_model: id 57 | task -1 | new slot, n_ctx = 256 0.43.993.752 I slot load_model: id 58 | task -1 | new slot, n_ctx = 256 0.43.993.753 I slot load_model: id 59 | task -1 | new slot, n_ctx = 256 0.43.993.755 I slot load_model: id 60 | task -1 | new slot, n_ctx = 256 0.43.993.756 I slot load_model: id 61 | task -1 | new slot, n_ctx = 256 0.43.993.758 I slot load_model: id 62 | task -1 | new slot, n_ctx = 256 0.43.993.759 I slot load_model: id 63 | task -1 | new slot, n_ctx = 256 0.43.993.761 I slot load_model: id 64 | task -1 | new slot, n_ctx = 256 0.43.993.762 I slot load_model: id 65 | task -1 | new slot, n_ctx = 256 0.43.993.763 I slot load_model: id 66 | task -1 | new slot, n_ctx = 256 0.43.993.765 I slot load_model: id 67 | task -1 | new slot, n_ctx = 256 0.43.993.766 I slot load_model: id 68 | task -1 | new slot, n_ctx = 256 0.43.993.768 I slot load_model: id 69 | task -1 | new slot, n_ctx = 256 0.43.993.769 I slot load_model: id 70 | task -1 | new slot, n_ctx = 256 0.43.993.771 I slot load_model: id 71 | task -1 | new slot, n_ctx = 256 0.43.993.772 I slot load_model: id 72 | task -1 | new slot, n_ctx = 256 0.43.993.774 I slot load_model: id 73 | task -1 | new slot, n_ctx = 256 0.43.993.775 I slot load_model: id 74 | task -1 | new slot, n_ctx = 256 0.43.993.776 I slot load_model: id 75 | task -1 | new slot, n_ctx = 256 0.43.993.778 I slot load_model: id 76 | task -1 | new slot, n_ctx = 256 0.43.993.779 I slot load_model: id 77 | task -1 | new slot, n_ctx = 256 0.43.993.781 I slot load_model: id 78 | task -1 | new slot, n_ctx = 256 0.43.993.782 I slot load_model: id 79 | task -1 | new slot, n_ctx = 256 0.43.993.784 I slot load_model: id 80 | task -1 | new slot, n_ctx = 256 0.43.993.785 I slot load_model: id 81 | task -1 | new slot, n_ctx = 256 0.43.993.787 I slot load_model: id 82 | task -1 | new slot, n_ctx = 256 0.43.993.788 I slot load_model: id 83 | task -1 | new slot, n_ctx = 256 0.43.993.789 I slot load_model: id 84 | task -1 | new slot, n_ctx = 256 0.43.993.791 I slot load_model: id 85 | task -1 | new slot, n_ctx = 256 0.43.993.792 I slot load_model: id 86 | task -1 | new slot, n_ctx = 256 0.43.993.794 I slot load_model: id 87 | task -1 | new slot, n_ctx = 256 0.43.993.795 I slot load_model: id 88 | task -1 | new slot, n_ctx = 256 0.43.993.797 I slot load_model: id 89 | task -1 | new slot, n_ctx = 256 0.43.993.798 I slot load_model: id 90 | task -1 | new slot, n_ctx = 256 0.43.993.800 I slot load_model: id 91 | task -1 | new slot, n_ctx = 256 0.43.993.801 I slot load_model: id 92 | task -1 | new slot, n_ctx = 256 0.43.993.802 I slot load_model: id 93 | task -1 | new slot, n_ctx = 256 0.43.993.804 I slot load_model: id 94 | task -1 | new slot, n_ctx = 256 0.43.993.805 I slot load_model: id 95 | task -1 | new slot, n_ctx = 256 0.43.993.807 I slot load_model: id 96 | task -1 | new slot, n_ctx = 256 0.43.993.808 I slot load_model: id 97 | task -1 | new slot, n_ctx = 256 0.43.993.810 I slot load_model: id 98 | task -1 | new slot, n_ctx = 256 0.43.993.811 I slot load_model: id 99 | task -1 | new slot, n_ctx = 256 0.43.993.813 I slot load_model: id 100 | task -1 | new slot, n_ctx = 256 0.43.993.814 I slot load_model: id 101 | task -1 | new slot, n_ctx = 256 0.43.993.816 I slot load_model: id 102 | task -1 | new slot, n_ctx = 256 0.43.993.817 I slot load_model: id 103 | task -1 | new slot, n_ctx = 256 0.43.993.819 I slot load_model: id 104 | task -1 | new slot, n_ctx = 256 0.43.993.820 I slot load_model: id 105 | task -1 | new slot, n_ctx = 256 0.43.993.822 I slot load_model: id 106 | task -1 | new slot, n_ctx = 256 0.43.993.823 I slot load_model: id 107 | task -1 | new slot, n_ctx = 256 0.43.993.824 I slot load_model: id 108 | task -1 | new slot, n_ctx = 256 0.43.993.826 I slot load_model: id 109 | task -1 | new slot, n_ctx = 256 0.43.993.827 I slot load_model: id 110 | task -1 | new slot, n_ctx = 256 0.43.993.829 I slot load_model: id 111 | task -1 | new slot, n_ctx = 256 0.43.993.830 I slot load_model: id 112 | task -1 | new slot, n_ctx = 256 0.43.993.832 I slot load_model: id 113 | task -1 | new slot, n_ctx = 256 0.43.993.833 I slot load_model: id 114 | task -1 | new slot, n_ctx = 256 0.43.993.835 I slot load_model: id 115 | task -1 | new slot, n_ctx = 256 0.43.993.836 I slot load_model: id 116 | task -1 | new slot, n_ctx = 256 0.43.993.838 I slot load_model: id 117 | task -1 | new slot, n_ctx = 256 0.43.993.839 I slot load_model: id 118 | task -1 | new slot, n_ctx = 256 0.43.993.840 I slot load_model: id 119 | task -1 | new slot, n_ctx = 256 0.43.993.842 I slot load_model: id 120 | task -1 | new slot, n_ctx = 256 0.43.993.843 I slot load_model: id 121 | task -1 | new slot, n_ctx = 256 0.43.993.845 I slot load_model: id 122 | task -1 | new slot, n_ctx = 256 0.43.993.846 I slot load_model: id 123 | task -1 | new slot, n_ctx = 256 0.43.993.848 I slot load_model: id 124 | task -1 | new slot, n_ctx = 256 0.43.993.849 I slot load_model: id 125 | task -1 | new slot, n_ctx = 256 0.43.993.851 I slot load_model: id 126 | task -1 | new slot, n_ctx = 256 0.43.993.852 I slot load_model: id 127 | task -1 | new slot, n_ctx = 256 0.43.994.142 I srv load_model: prompt cache is enabled, size limit: 8192 MiB |
'프로그램 사용 > ai 프로그램' 카테고리의 다른 글
| vllm concurrent test/benchmark (0) | 2026.07.29 |
|---|---|
| vllm parallelism - concurrency (0) | 2026.07.28 |
| nvidia thor - llama.cpp LLM (0) | 2026.07.24 |
| llama.cpp for arm64 (0) | 2026.07.20 |
| qwen3.5 0.8B (0) | 2026.07.19 |
