llama_bin_run\llama-server.exe : ggml_cuda_init: found 2 C UDA devices (Total VRAM: 24575 MiB): 위치 줄:1 문자:1 + llama_bin_run\llama-server.exe --model "models\Q4_K_M\Qw en3.5-122B-A1 ... + ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~ + CategoryInfo : NotSpecified: (ggml_cuda_in it:...AM: 24575 MiB)::String) [], RemoteException + FullyQualifiedErrorId : NativeCommandError Device 0: NVIDIA GeForce RTX 3060, compute capability 8. 6, VMM: yes, VRAM: 12287 MiB Device 1: NVIDIA GeForce RTX 3060, compute capability 8. 6, VMM: yes, VRAM: 12287 MiB load_backend: loaded CUDA backend from C:\Users\Variet-Wor ker\Desktop\variet-llm\llama_bin_run\ggml-cuda.dll load_backend: loaded RPC backend from C:\Users\Variet-Work er\Desktop\variet-llm\llama_bin_run\ggml-rpc.dll load_backend: loaded CPU backend from C:\Users\Variet-Work er\Desktop\variet-llm\llama_bin_run\ggml-cpu-haswell.dll system info: n_threads = 6, n_threads_batch = 6, total_thr eads = 16 system_info: n_threads = 6 (n_threads_batch = 6) / 16 | CU DA : ARCHS = 500,610,700,750,800,860,890 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAM AFILE = 1 | OPENMP = 1 | REPACK = 1 | Running without SSL init: using 15 threads for HTTP server start: binding port with default address family main: loading model srv load_model: loading model 'models\Q4_K_M\Qwen3.5-12 2B-A10B-Q4_K_M-00001-of-00003.gguf' llama_model_load_from_file_impl: using device CUDA0 (NVIDI A GeForce RTX 3060) (0000:04:00.0) - 11245 MiB free llama_model_load_from_file_impl: using device CUDA1 (NVIDI A GeForce RTX 3060) (0000:06:00.0) - 11240 MiB free llama_model_loader: additional 2 GGUFs metadata loaded. llama_model_loader: loaded meta data with 55 key-value pai rs and 879 tensors from models\Q4_K_M\Qwen3.5-122B-A10B-Q4 _K_M-00001-of-00003.gguf (version GGUF V3 (latest)) llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output. llama_model_loader: - kv 0: genera l.architecture str = qwen35moe llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general. sampling.top_k i32 = 20 llama_model_loader: - kv 3: general. sampling.top_p f32 = 0.950000 llama_model_loader: - kv 4: general .sampling.temp f32 = 0.600000 llama_model_loader: - kv 5: general.name str = Qwen3.5-122B-A10B llama_model_loader: - kv 6: ge neral.basename str = Qwen3.5-122B-A10B llama_model_loader: - kv 7: genera l.quantized_by str = Unsloth llama_model_loader: - kv 8: gene ral.size_label str = 122B-A10B llama_model_loader: - kv 9: g eneral.license str = apache-2.0 llama_model_loader: - kv 10: genera l.license.link str = https://huggingface.co/Q wen/Qwen3.5-1... llama_model_loader: - kv 11: ge neral.repo_url str = https://huggingface.co/u nsloth llama_model_loader: - kv 12: general.ba se_model.count u32 = 1 llama_model_loader: - kv 13: general.bas e_model.0.name str = Qwen3.5 122B A10B llama_model_loader: - kv 14: general.base_model. 0.organization str = Qwen llama_model_loader: - kv 15: general.base_mo del.0.repo_url str = https://huggingface.co/Q wen/Qwen3.5-1... llama_model_loader: - kv 16: general.tags arr[str,2] = ["unsloth", "image-text- to-text"] llama_model_loader: - kv 17: qwen35m oe.block_count u32 = 48 llama_model_loader: - kv 18: qwen35moe. context_length u32 = 262144 llama_model_loader: - kv 19: qwen35moe.em bedding_length u32 = 3072 llama_model_loader: - kv 20: qwen35moe.attent ion.head_count u32 = 32 llama_model_loader: - kv 21: qwen35moe.attention .head_count_kv u32 = 2 llama_model_loader: - kv 22: qwen35moe.rope.dime nsion_sections arr[i32,4] = [11, 11, 10, 0] llama_model_loader: - kv 23: qwen35moe. rope.freq_base f32 = 10000000.000000 llama_model_loader: - kv 24: qwen35moe.attention.layer_no rm_rms_epsilon f32 = 0.000001 llama_model_loader: - kv 25: qwen35mo e.expert_count u32 = 256 llama_model_loader: - kv 26: qwen35moe.exp ert_used_count u32 = 8 llama_model_loader: - kv 27: qwen35moe.attent ion.key_length u32 = 256 llama_model_loader: - kv 28: qwen35moe.attentio n.value_length u32 = 256 llama_model_loader: - kv 29: qwen35moe.expert_feed_ forward_length u32 = 1024 llama_model_loader: - kv 30: qwen35moe.expert_shared_feed _forward_length u32 = 1024 llama_model_loader: - kv 31: qwen35moe.s sm.conv_kernel u32 = 4 llama_model_loader: - kv 32: qwen35moe. ssm.state_size u32 = 128 llama_model_loader: - kv 33: qwen35moe.s sm.group_count u32 = 16 llama_model_loader: - kv 34: qwen35moe.ssm. time_step_rank u32 = 64 llama_model_loader: - kv 35: qwen35moe. ssm.inner_size u32 = 8192 llama_model_loader: - kv 36: qwen35moe.full_atte ntion_interval u32 = 4 llama_model_loader: - kv 37: qwen35moe.rope.d imension_count u32 = 64 llama_model_loader: - kv 38: tokeni zer.ggml.model str = gpt2 llama_model_loader: - kv 39: toke nizer.ggml.pre str = qwen35 llama_model_loader: - kv 40: tokeniz er.ggml.tokens arr[str,248320] = ["!", "\"", "#", "$", "% ", "&", "'", ... llama_model_loader: - kv 41: tokenizer.g gml.token_type arr[i32,248320] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 42: tokeniz er.ggml.merges arr[str,247587] = ["휔 휔", "휔휔 휔휔", "i n", "휔 t",... llama_model_loader: - kv 43: tokenizer.ggm l.eos_token_id u32 = 248046 llama_model_loader: - kv 44: tokenizer.ggml.pa dding_token_id u32 = 248055 llama_model_loader: - kv 45: tokenizer .chat_template str = {%- set image_count = na mespace(value... llama_model_loader: - kv 46: general.quanti zation_version u32 = 2 llama_model_loader: - kv 47: gen eral.file_type u32 = 15 llama_model_loader: - kv 48: quantiz e.imatrix.file str = Qwen3.5-122B-A10B-GGUF/i matrix_unslot... llama_model_loader: - kv 49: quantize.i matrix.dataset str = unsloth_calibration_Qwen 3.5-122B-A10B... llama_model_loader: - kv 50: quantize.imatrix .entries_count u32 = 612 llama_model_loader: - kv 51: quantize.imatri x.chunks_count u32 = 76 llama_model_loader: - kv 52: split.no u16 = 0 llama_model_loader: - kv 53: split .tensors.count i32 = 879 llama_model_loader: - kv 54: split.count u16 = 3 llama_model_loader: - type f32: 361 tensors llama_model_loader: - type q8_0: 373 tensors llama_model_loader: - type q4_K: 96 tensors llama_model_loader: - type q5_K: 48 tensors llama_model_loader: - type q6_K: 1 tensors print_info: file format = GGUF V3 (latest) print_info: file type = Q4_K - Medium print_info: file size = 71.27 GiB (5.01 BPW) load: 0 unused tokens load: printing all EOG tokens: load: - 248044 ('<|endoftext|>') load: - 248046 ('<|im_end|>') load: - 248063 ('<|fim_pad|>') load: - 248064 ('<|repo_name|>') load: - 248065 ('<|file_sep|>') load: special tokens cache size = 33 load: token to piece cache size = 1.7581 MB print_info: arch = qwen35moe print_info: vocab_only = 0 print_info: no_alloc = 0 print_info: n_ctx_train = 262144 print_info: n_embd = 3072 print_info: n_embd_inp = 3072 print_info: n_layer = 48 print_info: n_head = 32 print_info: n_head_kv = 2 print_info: n_rot = 64 print_info: n_swa = 0 print_info: is_swa_any = 0 print_info: n_embd_head_k = 256 print_info: n_embd_head_v = 256 print_info: n_gqa = 16 print_info: n_embd_k_gqa = 512 print_info: n_embd_v_gqa = 512 print_info: f_norm_eps = 0.0e+00 print_info: f_norm_rms_eps = 1.0e-06 print_info: f_clamp_kqv = 0.0e+00 print_info: f_max_alibi_bias = 0.0e+00 print_info: f_logit_scale = 0.0e+00 print_info: f_attn_scale = 0.0e+00 print_info: n_ff = 0 print_info: n_expert = 256 print_info: n_expert_used = 8 print_info: n_expert_groups = 0 print_info: n_group_used = 0 print_info: causal attn = 1 print_info: pooling type = -1 print_info: rope type = 40 print_info: rope scaling = linear print_info: freq_base_train = 10000000.0 print_info: freq_scale_train = 1 print_info: n_ctx_orig_yarn = 262144 print_info: rope_yarn_log_mul = 0.0000 print_info: rope_finetuned = unknown print_info: mrope sections = [11, 11, 10, 0] print_info: ssm_d_conv = 4 print_info: ssm_d_inner = 8192 print_info: ssm_d_state = 128 print_info: ssm_dt_rank = 64 print_info: ssm_n_group = 16 print_info: ssm_dt_b_c_rms = 0 print_info: model type = 122B.A10B print_info: model params = 122.11 B print_info: general.name = Qwen3.5-122B-A10B print_info: vocab type = BPE print_info: n_vocab = 248320 print_info: n_merges = 247587 print_info: BOS token = 11 ',' print_info: EOS token = 248046 '<|im_end|>' print_info: EOT token = 248046 '<|im_end|>' print_info: PAD token = 248055 '<|vision_pad|> ' print_info: LF token = 198 '훹' print_info: FIM PRE token = 248060 '<|fim_prefix|> ' print_info: FIM SUF token = 248062 '<|fim_suffix|> ' print_info: FIM MID token = 248061 '<|fim_middle|> ' print_info: FIM PAD token = 248063 '<|fim_pad|>' print_info: FIM REP token = 248064 '<|repo_name|>' print_info: FIM SEP token = 248065 '<|file_sep|>' print_info: EOG token = 248044 '<|endoftext|>' print_info: EOG token = 248046 '<|im_end|>' print_info: EOG token = 248063 '<|fim_pad|>' print_info: EOG token = 248064 '<|repo_name|>' print_info: EOG token = 248065 '<|file_sep|>' print_info: max token length = 256 load_tensors: loading model tensors, this can take a while ... (mmap = true, direct_io = false) llama_model_loader: tensor overrides to CPU are used with mmap enabled - consider using --no-mmap for better perform ance load_tensors: offloading output layer to GPU load_tensors: offloading 47 repeating layers to GPU load_tensors: offloaded 49/49 layers to GPU load_tensors: CPU_Mapped model buffer size = 47000.12 Mi B load_tensors: CPU_Mapped model buffer size = 25271.75 Mi B load_tensors: CUDA0 model buffer size = 2499.98 Mi B load_tensors: CUDA1 model buffer size = 2891.90 Mi B .......................................................... .......................................... common_init_result: added <|endoftext|> logit bias = -inf common_init_result: added <|im_end|> logit bias = -inf common_init_result: added <|fim_pad|> logit bias = -inf common_init_result: added <|repo_name|> logit bias = -inf common_init_result: added <|file_sep|> logit bias = -inf llama_context: constructing llama_context llama_context: n_seq_max = 1 llama_context: n_ctx = 32768 llama_context: n_ctx_seq = 32768 llama_context: n_batch = 2048 llama_context: n_ubatch = 512 llama_context: causal_attn = 1 llama_context: flash_attn = enabled llama_context: kv_unified = false llama_context: freq_base = 10000000.0 llama_context: freq_scale = 1 llama_context: n_ctx_seq (32768) < n_ctx_train (262144) -- the full capacity of the model will not be utilized llama_context: CUDA_Host output buffer size = 0.95 M iB llama_kv_cache: CUDA0 KV buffer size = 108.00 MiB llama_kv_cache: CUDA1 KV buffer size = 108.00 MiB llama_kv_cache: size = 216.00 MiB ( 32768 cells, 12 laye rs, 1/1 seqs), K (q4_0): 108.00 MiB, V (q4_0): 108.00 M iB llama_kv_cache: attn_rot_k = 1 llama_kv_cache: attn_rot_v = 1 llama_memory_recurrent: CUDA0 RS buffer size = 78. 67 MiB llama_memory_recurrent: CUDA1 RS buffer size = 70. 39 MiB llama_memory_recurrent: size = 149.06 MiB ( 1 cells, 48 layers, 1 seqs), R (f32): 5.06 MiB, S (f32): 144. 00 MiB sched_reserve: reserving ... sched_reserve: resolving fused Gated Delta Net support: sched_reserve: fused Gated Delta Net (autoregressive) enab led sched_reserve: fused Gated Delta Net (chunked) enabled sched_reserve: CUDA0 compute buffer size = 966.50 M iB sched_reserve: CUDA1 compute buffer size = 503.00 M iB sched_reserve: CUDA_Host compute buffer size = 76.29 M iB sched_reserve: graph nodes = 4617 sched_reserve: graph splits = 170 (with bs=512), 99 (with bs=1) sched_reserve: reserve took 18.29 ms, sched copies = 1 common_init_from_params: warming up the model with an empt y run - please wait ... (--no-warmup to disable) srv load_model: initializing slots, n_slots = 1 common_speculative_is_compat: the target context does not support partial sequence removal srv load_model: speculative decoding not supported by t his context slot load_model: id 0 | task -1 | new slot, n_ctx = 327 68 srv load_model: prompt cache is enabled, size limit: 81 92 MiB srv load_model: use `--cache-ram 0` to disable the prom pt cache srv load_model: for more info see https://github.com/gg ml-org/llama.cpp/pull/16391 srv init: init: --clear-idle requires --kv-unifie d, disabling 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 ' srv init: init: chat template, thinking = 1 main: model loaded main: server is listening on http://0.0.0.0:8000 main: starting the main loop... srv update_slots: all slots are idle srv params_from_: Chat format: peg-native slot get_availabl: id 0 | task -1 | selected slot by LRU, t_last = -1 srv get_availabl: updating prompt cache srv load: - looking for better prompt, base f_ke ep = -1.000, sim = 0.000 srv update: - cache state: 0 prompts, 0.000 MiB (l imits: 8192.000 MiB, 32768 tokens, 8589934592 est) srv get_availabl: prompt cache update took 0.01 ms slot launch_slot_: id 0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typica l -> top-p -> min-p -> ?xtc -> temp-ext -> dist slot launch_slot_: id 0 | task 0 | processing task, is_ch ild = 0 slot update_slots: id 0 | task 0 | new prompt, n_ctx_slot = 32768, n_keep = 0, task.n_tokens = 18 slot update_slots: id 0 | task 0 | n_tokens = 0, memory_s eq_rm [0, end) slot update_slots: id 0 | task 0 | prompt processing prog ress, n_tokens = 14, batch.n_tokens = 14, progress = 0.777 778 slot update_slots: id 0 | task 0 | n_tokens = 14, memory_ seq_rm [14, end) reasoning-budget: activated, budget=2147483647 tokens slot init_sampler: id 0 | task 0 | init sampler, took 0.0 1 ms, tokens: text = 18, total = 18 slot update_slots: id 0 | task 0 | prompt processing done , n_tokens = 18, batch.n_tokens = 4 slot print_timing: id 0 | task 0 | prompt eval time = 1668.15 ms / 18 tokens ( 92.68 ms per token, 10.79 tokens per second) eval time = 2269.44 ms / 20 tokens ( 113.47 ms per token, 8.81 tokens per second) total time = 3937.59 ms / 38 tokens slot release: id 0 | task 0 | stop processing: n_tok ens = 37, truncated = 0 srv update_slots: all slots are idle srv log_server_r: done request: POST /v1/chat/completions 127.0.0.1 200 srv params_from_: Chat format: peg-native slot get_availabl: id 0 | task -1 | selected slot by LCP similarity, sim_best = 1.000 (> 0.100 thold), f_keep = 0.4 86 srv get_availabl: updating prompt cache srv prompt_save: - saving prompt with length 37, total state size = 149.308 MiB srv load: - looking for better prompt, base f_ke ep = 0.486, sim = 1.000 srv update: - cache state: 1 prompts, 149.308 MiB (limits: 8192.000 MiB, 32768 tokens, 32768 est) srv update: - prompt 00000282FC468F50: 37 t okens, checkpoints: 0, 149.308 MiB srv get_availabl: prompt cache update took 52.31 ms slot launch_slot_: id 0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typica l -> top-p -> min-p -> ?xtc -> temp-ext -> dist slot launch_slot_: id 0 | task 22 | processing task, is_c hild = 0 slot update_slots: id 0 | task 22 | new prompt, n_ctx_slo t = 32768, n_keep = 0, task.n_tokens = 18 slot update_slots: id 0 | task 22 | n_past = 18, slot.pro mpt.tokens.size() = 37, seq_id = 0, pos_min = 36, n_swa = 0 slot update_slots: id 0 | task 22 | forcing full prompt r e-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-or g/llama.cpp/pull/13194#issuecomment-2868343055) slot update_slots: id 0 | task 22 | n_tokens = 0, memory_ seq_rm [0, end) slot update_slots: id 0 | task 22 | prompt processing pro gress, n_tokens = 14, batch.n_tokens = 14, progress = 0.77 7778 slot update_slots: id 0 | task 22 | n_tokens = 14, memory _seq_rm [14, end) reasoning-budget: activated, budget=2147483647 tokens slot init_sampler: id 0 | task 22 | init sampler, took 0. 01 ms, tokens: text = 18, total = 18 slot update_slots: id 0 | task 22 | prompt processing don e, n_tokens = 18, batch.n_tokens = 4 slot print_timing: id 0 | task 22 | prompt eval time = 869.82 ms / 18 tokens ( 48.32 ms per token, 20.69 tokens per second) eval time = 40320.89 ms / 200 tokens ( 201.60 ms per token, 4.96 tokens per second) total time = 41190.70 ms / 218 tokens slot release: id 0 | task 22 | stop processing: n_to kens = 217, truncated = 0 srv update_slots: all slots are idle srv log_server_r: done request: POST /v1/chat/completions 127.0.0.1 200 srv params_from_: Chat format: peg-native slot get_availabl: id 0 | task -1 | selected slot by LRU, t_last = 105302505 srv get_availabl: updating prompt cache srv prompt_save: - saving prompt with length 217, total state size = 150.498 MiB