Misc. bug: The inference speed of llama-server is one-third of that of llama-cli

[zts9989]

Name and Version

llama-server --version
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
Device 0: NVIDIA GeForce GTX 1080 Ti, compute capability 6.1, VMM: yes
version: b4819
built with cc (Debian 12.2.0-14) 12.2.0 for x86_64-linux-gnu

llama-cli --version
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
Device 0: NVIDIA GeForce GTX 1080 Ti, compute capability 6.1, VMM: yes
version: b4819
built with cc (Debian 12.2.0-14) 12.2.0 for x86_64-linux-gnu

Operating systems

Linux

Which llama.cpp modules do you know to be affected?

llama-server

Command line

llama-server -m /data4/qwen2.5-14b-deep-q4_k.gguf  -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047  -ngl 160  --host 0.0.0.0

llama-server -m /data4/qwen2.5-14b-deep-q4_k.gguf  -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047  -ngl 160 -nkvo --host 0.0.0.0

llama-cli -m /data4/qwen2.5-14b-deep-q4_k.gguf -cnv -p "You are a helpful assistant." -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -if -mli -ngl 160 -nkvo

llama-cli -m /data4/qwen2.5-14b-deep-q4_k.gguf -cnv -p "You are a helpful assistant." -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -if -mli -ngl 160

Problem description & steps to reproduce

In the same environment, with the same parameters, the inference speed of llama-server is one-third of that of llama-cli.

Command lines as follows:
llama-server -m /data4/qwen2.5-14b-deep-q4_k.gguf -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -ngl 160 --host 0.0.0.0
This parameter configuration achieves an inference speed of: 25.87 t/s

llama-server -m /data4/qwen2.5-14b-deep-q4_k.gguf -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -ngl 160 -nkvo --host 0.0.0.0
This parameter configuration achieves an inference speed of: 5.83 t/s

llama-cli -m /data4/qwen2.5-14b-deep-q4_k.gguf -cnv -p "You are a helpful assistant." -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -if -mli -ngl 160 -nkvo
This parameter configuration achieves an inference speed of: 18.25 t/s

llama-cli -m /data4/qwen2.5-14b-deep-q4_k.gguf -cnv -p "You are a helpful assistant." -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -if -mli -ngl 160
This parameter configuration achieves an inference speed of: 24.27 t/s

Although the -nkvo option leaves kv calculations on the CPU, which slows down inference speed, for example, llama-cli's speed drops from 24.27 to 18.25, which is expected behavior. However, enabling -nkvo in llama-server causes a greater-than-expected drop. On my other computers, it even drops to 0.x t/s. llama-server with -nkvo is expected to have inference speeds similar to 18.xx t/s. I have reproduced this issue in multiple computer environments. I suspect there is a bug in the thread pool usage of llama-server. Thank you for this project that allows me to run LLMs locally. Looking forward to fixing this bug.

First Bad Commit

No response

Relevant log output

./build/bin/llama-cli -m /data4/qwen2.5-14b-deep-q4_k.gguf -cnv -p "You are a helpful assistant." -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -if -mli -ngl 160 

ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: NVIDIA GeForce GTX 1080 Ti, compute capability 6.1, VMM: yes
build: 0 (unknown) with cc (Debian 12.2.0-14) 12.2.0 for x86_64-linux-gnu
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce GTX 1080 Ti) - 11025 MiB free
llama_model_loader: loaded meta data with 25 key-value pairs and 579 tensors from /data4/qwen2.5-14b-deep-q4_k.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:                       general.architecture str              = qwen2
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Qw14
llama_model_loader: - kv   3:                         general.size_label str              = 15B
llama_model_loader: - kv   4:                          qwen2.block_count u32              = 48
llama_model_loader: - kv   5:                       qwen2.context_length u32              = 131072
llama_model_loader: - kv   6:                     qwen2.embedding_length u32              = 5120
llama_model_loader: - kv   7:                  qwen2.feed_forward_length u32              = 13824
llama_model_loader: - kv   8:                 qwen2.attention.head_count u32              = 40
llama_model_loader: - kv   9:              qwen2.attention.head_count_kv u32              = 8
llama_model_loader: - kv  10:                       qwen2.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  11:     qwen2.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  12:                          general.file_type u32              = 15
llama_model_loader: - kv  13:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  14:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  15:                      tokenizer.ggml.tokens arr[str,152064]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  16:                  tokenizer.ggml.token_type arr[i32,152064]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  17:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  18:                tokenizer.ggml.bos_token_id u32              = 151646
llama_model_loader: - kv  19:                tokenizer.ggml.eos_token_id u32              = 151643
llama_model_loader: - kv  20:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  21:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  22:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  23:                    tokenizer.chat_template str              = {% if not add_generation_prompt is de...
llama_model_loader: - kv  24:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  241 tensors
llama_model_loader: - type q4_K:  289 tensors
llama_model_loader: - type q6_K:   49 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q4_K - Medium
print_info: file size   = 8.37 GiB (4.87 BPW) 
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 22
load: token to piece cache size = 0.9310 MB
print_info: arch             = qwen2
print_info: vocab_only       = 0
print_info: n_ctx_train      = 131072
print_info: n_embd           = 5120
print_info: n_layer          = 48
print_info: n_head           = 40
print_info: n_head_kv        = 8
print_info: n_rot            = 128
print_info: n_swa            = 0
print_info: n_embd_head_k    = 128
print_info: n_embd_head_v    = 128
print_info: n_gqa            = 5
print_info: n_embd_k_gqa     = 1024
print_info: n_embd_v_gqa     = 1024
print_info: f_norm_eps       = 0.0e+00
print_info: f_norm_rms_eps   = 1.0e-05
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: n_ff             = 13824
print_info: n_expert         = 0
print_info: n_expert_used    = 0
print_info: causal attn      = 1
print_info: pooling type     = 0
print_info: rope type        = 2
print_info: rope scaling     = linear
print_info: freq_base_train  = 1000000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn  = 131072
print_info: rope_finetuned   = unknown
print_info: ssm_d_conv       = 0
print_info: ssm_d_inner      = 0
print_info: ssm_d_state      = 0
print_info: ssm_dt_rank      = 0
print_info: ssm_dt_b_c_rms   = 0
print_info: model type       = 14B
print_info: model params     = 14.77 B
print_info: general.name     = Qw14
print_info: vocab type       = BPE
print_info: n_vocab          = 152064
print_info: n_merges         = 151387
print_info: BOS token        = 151646 '<｜begin▁of▁sentence｜>'
print_info: EOS token        = 151643 '<｜end▁of▁sentence｜>'
print_info: EOT token        = 151643 '<｜end▁of▁sentence｜>'
print_info: PAD token        = 151643 '<｜end▁of▁sentence｜>'
print_info: LF token         = 198 'Ċ'
print_info: FIM PRE token    = 151659 '<|fim_prefix|>'
print_info: FIM SUF token    = 151661 '<|fim_suffix|>'
print_info: FIM MID token    = 151660 '<|fim_middle|>'
print_info: FIM PAD token    = 151662 '<|fim_pad|>'
print_info: FIM REP token    = 151663 '<|repo_name|>'
print_info: FIM SEP token    = 151664 '<|file_sep|>'
print_info: EOG token        = 151643 '<｜end▁of▁sentence｜>'
print_info: EOG token        = 151662 '<|fim_pad|>'
print_info: EOG token        = 151663 '<|repo_name|>'
print_info: EOG token        = 151664 '<|file_sep|>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 48 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 49/49 layers to GPU
load_tensors:        CUDA0 model buffer size =  8148.38 MiB
load_tensors:   CPU_Mapped model buffer size =   417.66 MiB
..........................................................................................
llama_init_from_model: n_seq_max     = 1
llama_init_from_model: n_ctx         = 512
llama_init_from_model: n_ctx_per_seq = 512
llama_init_from_model: n_batch       = 512
llama_init_from_model: n_ubatch      = 512
llama_init_from_model: flash_attn    = 1
llama_init_from_model: freq_base     = 1000000.0
llama_init_from_model: freq_scale    = 1
llama_init_from_model: n_ctx_per_seq (512) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 512, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 48, can_shift = 1
llama_kv_cache_init:      CUDA0 KV buffer size =    96.00 MiB
llama_init_from_model: KV self size  =   96.00 MiB, K (f16):   48.00 MiB, V (f16):   48.00 MiB
llama_init_from_model:  CUDA_Host  output buffer size =     0.58 MiB
llama_init_from_model:      CUDA0 compute buffer size =   307.00 MiB
llama_init_from_model:  CUDA_Host compute buffer size =    11.01 MiB
llama_init_from_model: graph nodes  = 1495
llama_init_from_model: graph splits = 2
common_init_from_params: setting dry_penalty_last_n to ctx_size = 512
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
main: llama threadpool init, n_threads = 24
*** User-specified prompt in conversation mode will be ignored, did you mean to set --system-prompt (-sys) instead?
main: chat template example:
You are a helpful assistant

<｜User｜>Hello<｜Assistant｜>Hi there<｜end▁of▁sentence｜><｜User｜>How are you?<｜Assistant｜>

system_info: n_threads = 24 (n_threads_batch = 24) / 48 | CUDA : ARCHS = 610 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | LLAMAFILE = 1 | AARCH64_REPACK = 1 | 

main: interactive mode on.
sampler seed: 3047
sampler params: 
	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 512
	top_k = 40, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.600
	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
sampler chain: logits -> logit-bias -> penalties -> dry -> top-k -> typical -> top-p -> min-p -> xtc -> temp-ext -> dist 
generate: n_ctx = 512, n_batch = 2048, n_predict = -1, n_keep = 1

== Running in interactive mode. ==
 - Press Ctrl+C to interject at any time.
 - To return control to the AI, end your input with '\'.
 - To return control without starting a new line, end your input with '/'.
 - Not using system message. To change it, set a different value via -sys PROMPT


> hello
<think>

Hello! How can I assist you today? 😊

> 
llama_perf_sampler_print:    sampling time =       1.69 ms /    18 runs   (    0.09 ms per token, 10663.51 tokens per second)
llama_perf_context_print:        load time =    6771.74 ms
llama_perf_context_print: prompt eval time =     110.77 ms /     5 tokens (   22.15 ms per token,    45.14 tokens per second)
llama_perf_context_print:        eval time =     535.54 ms /    13 runs   (   41.20 ms per token,    **24.27 tokens per second)**
llama_perf_context_print:       total time =   19145.65 ms /    18 tokens
Interrupted by user


./build/bin/llama-cli -m /data4/qwen2.5-14b-deep-q4_k.gguf -cnv -p "You are a helpful assistant." -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -if -mli -ngl 160 -nkvo
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: NVIDIA GeForce GTX 1080 Ti, compute capability 6.1, VMM: yes
build: 0 (unknown) with cc (Debian 12.2.0-14) 12.2.0 for x86_64-linux-gnu
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce GTX 1080 Ti) - 11025 MiB free
llama_model_loader: loaded meta data with 25 key-value pairs and 579 tensors from /data4/qwen2.5-14b-deep-q4_k.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:                       general.architecture str              = qwen2
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Qw14
llama_model_loader: - kv   3:                         general.size_label str              = 15B
llama_model_loader: - kv   4:                          qwen2.block_count u32              = 48
llama_model_loader: - kv   5:                       qwen2.context_length u32              = 131072
llama_model_loader: - kv   6:                     qwen2.embedding_length u32              = 5120
llama_model_loader: - kv   7:                  qwen2.feed_forward_length u32              = 13824
llama_model_loader: - kv   8:                 qwen2.attention.head_count u32              = 40
llama_model_loader: - kv   9:              qwen2.attention.head_count_kv u32              = 8
llama_model_loader: - kv  10:                       qwen2.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  11:     qwen2.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  12:                          general.file_type u32              = 15
llama_model_loader: - kv  13:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  14:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  15:                      tokenizer.ggml.tokens arr[str,152064]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  16:                  tokenizer.ggml.token_type arr[i32,152064]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  17:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  18:                tokenizer.ggml.bos_token_id u32              = 151646
llama_model_loader: - kv  19:                tokenizer.ggml.eos_token_id u32              = 151643
llama_model_loader: - kv  20:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  21:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  22:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  23:                    tokenizer.chat_template str              = {% if not add_generation_prompt is de...
llama_model_loader: - kv  24:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  241 tensors
llama_model_loader: - type q4_K:  289 tensors
llama_model_loader: - type q6_K:   49 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q4_K - Medium
print_info: file size   = 8.37 GiB (4.87 BPW) 
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 22
load: token to piece cache size = 0.9310 MB
print_info: arch             = qwen2
print_info: vocab_only       = 0
print_info: n_ctx_train      = 131072
print_info: n_embd           = 5120
print_info: n_layer          = 48
print_info: n_head           = 40
print_info: n_head_kv        = 8
print_info: n_rot            = 128
print_info: n_swa            = 0
print_info: n_embd_head_k    = 128
print_info: n_embd_head_v    = 128
print_info: n_gqa            = 5
print_info: n_embd_k_gqa     = 1024
print_info: n_embd_v_gqa     = 1024
print_info: f_norm_eps       = 0.0e+00
print_info: f_norm_rms_eps   = 1.0e-05
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: n_ff             = 13824
print_info: n_expert         = 0
print_info: n_expert_used    = 0
print_info: causal attn      = 1
print_info: pooling type     = 0
print_info: rope type        = 2
print_info: rope scaling     = linear
print_info: freq_base_train  = 1000000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn  = 131072
print_info: rope_finetuned   = unknown
print_info: ssm_d_conv       = 0
print_info: ssm_d_inner      = 0
print_info: ssm_d_state      = 0
print_info: ssm_dt_rank      = 0
print_info: ssm_dt_b_c_rms   = 0
print_info: model type       = 14B
print_info: model params     = 14.77 B
print_info: general.name     = Qw14
print_info: vocab type       = BPE
print_info: n_vocab          = 152064
print_info: n_merges         = 151387
print_info: BOS token        = 151646 '<｜begin▁of▁sentence｜>'
print_info: EOS token        = 151643 '<｜end▁of▁sentence｜>'
print_info: EOT token        = 151643 '<｜end▁of▁sentence｜>'
print_info: PAD token        = 151643 '<｜end▁of▁sentence｜>'
print_info: LF token         = 198 'Ċ'
print_info: FIM PRE token    = 151659 '<|fim_prefix|>'
print_info: FIM SUF token    = 151661 '<|fim_suffix|>'
print_info: FIM MID token    = 151660 '<|fim_middle|>'
print_info: FIM PAD token    = 151662 '<|fim_pad|>'
print_info: FIM REP token    = 151663 '<|repo_name|>'
print_info: FIM SEP token    = 151664 '<|file_sep|>'
print_info: EOG token        = 151643 '<｜end▁of▁sentence｜>'
print_info: EOG token        = 151662 '<|fim_pad|>'
print_info: EOG token        = 151663 '<|repo_name|>'
print_info: EOG token        = 151664 '<|file_sep|>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 48 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 49/49 layers to GPU
load_tensors:        CUDA0 model buffer size =  8148.38 MiB
load_tensors:   CPU_Mapped model buffer size =   417.66 MiB
..........................................................................................
llama_init_from_model: n_seq_max     = 1
llama_init_from_model: n_ctx         = 512
llama_init_from_model: n_ctx_per_seq = 512
llama_init_from_model: n_batch       = 512
llama_init_from_model: n_ubatch      = 512
llama_init_from_model: flash_attn    = 1
llama_init_from_model: freq_base     = 1000000.0
llama_init_from_model: freq_scale    = 1
llama_init_from_model: n_ctx_per_seq (512) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 512, offload = 0, type_k = 'f16', type_v = 'f16', n_layer = 48, can_shift = 1
llama_kv_cache_init:        CPU KV buffer size =    96.00 MiB
llama_init_from_model: KV self size  =   96.00 MiB, K (f16):   48.00 MiB, V (f16):   48.00 MiB
llama_init_from_model:  CUDA_Host  output buffer size =     0.58 MiB
llama_init_from_model:      CUDA0 compute buffer size =   317.00 MiB
llama_init_from_model:  CUDA_Host compute buffer size =    11.01 MiB
llama_init_from_model: graph nodes  = 1495
llama_init_from_model: graph splits = 98
common_init_from_params: setting dry_penalty_last_n to ctx_size = 512
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
main: llama threadpool init, n_threads = 24
*** User-specified prompt in conversation mode will be ignored, did you mean to set --system-prompt (-sys) instead?
main: chat template example:
You are a helpful assistant

<｜User｜>Hello<｜Assistant｜>Hi there<｜end▁of▁sentence｜><｜User｜>How are you?<｜Assistant｜>

system_info: n_threads = 24 (n_threads_batch = 24) / 48 | CUDA : ARCHS = 610 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | LLAMAFILE = 1 | AARCH64_REPACK = 1 | 

main: interactive mode on.
sampler seed: 3047
sampler params: 
	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 512
	top_k = 40, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.600
	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
sampler chain: logits -> logit-bias -> penalties -> dry -> top-k -> typical -> top-p -> min-p -> xtc -> temp-ext -> dist 
generate: n_ctx = 512, n_batch = 2048, n_predict = -1, n_keep = 1

== Running in interactive mode. ==
 - Press Ctrl+C to interject at any time.
 - To return control to the AI, end your input with '\'.
 - To return control without starting a new line, end your input with '/'.
 - Not using system message. To change it, set a different value via -sys PROMPT


> hello
<think>

Hello! How can I assist you today? 😊

> 
llama_perf_sampler_print:    sampling time =       1.66 ms /    18 runs   (    0.09 ms per token, 10849.91 tokens per second)
llama_perf_context_print:        load time =    1985.12 ms
llama_perf_context_print: prompt eval time =     148.26 ms /     5 tokens (   29.65 ms per token,    33.73 tokens per second)
llama_perf_context_print:        eval time =     712.26 ms /    13 runs   (   54.79 ms per token,    **18.25 tokens per second**)
llama_perf_context_print:       total time =    4656.76 ms /    18 tokens
Interrupted by user



./build/bin/llama-server -m /data4/qwen2.5-14b-deep-q4_k.gguf  -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047  -ngl 160 -nkvo --host 0.0.0.0
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: NVIDIA GeForce GTX 1080 Ti, compute capability 6.1, VMM: yes
build: 0 (unknown) with cc (Debian 12.2.0-14) 12.2.0 for x86_64-linux-gnu
system info: n_threads = 24, n_threads_batch = 24, total_threads = 48

system_info: n_threads = 24 (n_threads_batch = 24) / 48 | CUDA : ARCHS = 610 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | LLAMAFILE = 1 | AARCH64_REPACK = 1 | 

main: HTTP server is listening, hostname: 0.0.0.0, port: 8080, http threads: 47
main: loading model
srv    load_model: loading model '/data4/qwen2.5-14b-deep-q4_k.gguf'
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce GTX 1080 Ti) - 11025 MiB free
llama_model_loader: loaded meta data with 25 key-value pairs and 579 tensors from /data4/qwen2.5-14b-deep-q4_k.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:                       general.architecture str              = qwen2
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Qw14
llama_model_loader: - kv   3:                         general.size_label str              = 15B
llama_model_loader: - kv   4:                          qwen2.block_count u32              = 48
llama_model_loader: - kv   5:                       qwen2.context_length u32              = 131072
llama_model_loader: - kv   6:                     qwen2.embedding_length u32              = 5120
llama_model_loader: - kv   7:                  qwen2.feed_forward_length u32              = 13824
llama_model_loader: - kv   8:                 qwen2.attention.head_count u32              = 40
llama_model_loader: - kv   9:              qwen2.attention.head_count_kv u32              = 8
llama_model_loader: - kv  10:                       qwen2.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  11:     qwen2.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  12:                          general.file_type u32              = 15
llama_model_loader: - kv  13:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  14:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  15:                      tokenizer.ggml.tokens arr[str,152064]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  16:                  tokenizer.ggml.token_type arr[i32,152064]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  17:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  18:                tokenizer.ggml.bos_token_id u32              = 151646
llama_model_loader: - kv  19:                tokenizer.ggml.eos_token_id u32              = 151643
llama_model_loader: - kv  20:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  21:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  22:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  23:                    tokenizer.chat_template str              = {% if not add_generation_prompt is de...
llama_model_loader: - kv  24:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  241 tensors
llama_model_loader: - type q4_K:  289 tensors
llama_model_loader: - type q6_K:   49 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q4_K - Medium
print_info: file size   = 8.37 GiB (4.87 BPW) 
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 22
load: token to piece cache size = 0.9310 MB
print_info: arch             = qwen2
print_info: vocab_only       = 0
print_info: n_ctx_train      = 131072
print_info: n_embd           = 5120
print_info: n_layer          = 48
print_info: n_head           = 40
print_info: n_head_kv        = 8
print_info: n_rot            = 128
print_info: n_swa            = 0
print_info: n_embd_head_k    = 128
print_info: n_embd_head_v    = 128
print_info: n_gqa            = 5
print_info: n_embd_k_gqa     = 1024
print_info: n_embd_v_gqa     = 1024
print_info: f_norm_eps       = 0.0e+00
print_info: f_norm_rms_eps   = 1.0e-05
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: n_ff             = 13824
print_info: n_expert         = 0
print_info: n_expert_used    = 0
print_info: causal attn      = 1
print_info: pooling type     = 0
print_info: rope type        = 2
print_info: rope scaling     = linear
print_info: freq_base_train  = 1000000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn  = 131072
print_info: rope_finetuned   = unknown
print_info: ssm_d_conv       = 0
print_info: ssm_d_inner      = 0
print_info: ssm_d_state      = 0
print_info: ssm_dt_rank      = 0
print_info: ssm_dt_b_c_rms   = 0
print_info: model type       = 14B
print_info: model params     = 14.77 B
print_info: general.name     = Qw14
print_info: vocab type       = BPE
print_info: n_vocab          = 152064
print_info: n_merges         = 151387
print_info: BOS token        = 151646 '<｜begin▁of▁sentence｜>'
print_info: EOS token        = 151643 '<｜end▁of▁sentence｜>'
print_info: EOT token        = 151643 '<｜end▁of▁sentence｜>'
print_info: PAD token        = 151643 '<｜end▁of▁sentence｜>'
print_info: LF token         = 198 'Ċ'
print_info: FIM PRE token    = 151659 '<|fim_prefix|>'
print_info: FIM SUF token    = 151661 '<|fim_suffix|>'
print_info: FIM MID token    = 151660 '<|fim_middle|>'
print_info: FIM PAD token    = 151662 '<|fim_pad|>'
print_info: FIM REP token    = 151663 '<|repo_name|>'
print_info: FIM SEP token    = 151664 '<|file_sep|>'
print_info: EOG token        = 151643 '<｜end▁of▁sentence｜>'
print_info: EOG token        = 151662 '<|fim_pad|>'
print_info: EOG token        = 151663 '<|repo_name|>'
print_info: EOG token        = 151664 '<|file_sep|>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 48 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 49/49 layers to GPU
load_tensors:        CUDA0 model buffer size =  8148.38 MiB
load_tensors:   CPU_Mapped model buffer size =   417.66 MiB
..........................................................................................
llama_init_from_model: n_seq_max     = 1
llama_init_from_model: n_ctx         = 512
llama_init_from_model: n_ctx_per_seq = 512
llama_init_from_model: n_batch       = 512
llama_init_from_model: n_ubatch      = 512
llama_init_from_model: flash_attn    = 1
llama_init_from_model: freq_base     = 1000000.0
llama_init_from_model: freq_scale    = 1
llama_init_from_model: n_ctx_per_seq (512) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 512, offload = 0, type_k = 'f16', type_v = 'f16', n_layer = 48, can_shift = 1
llama_kv_cache_init:        CPU KV buffer size =    96.00 MiB
llama_init_from_model: KV self size  =   96.00 MiB, K (f16):   48.00 MiB, V (f16):   48.00 MiB
llama_init_from_model:  CUDA_Host  output buffer size =     0.58 MiB
llama_init_from_model:      CUDA0 compute buffer size =   317.00 MiB
llama_init_from_model:  CUDA_Host compute buffer size =    11.01 MiB
llama_init_from_model: graph nodes  = 1495
llama_init_from_model: graph splits = 98
common_init_from_params: setting dry_penalty_last_n to ctx_size = 512
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
srv          init: initializing slots, n_slots = 1
slot         init: id  0 | task -1 | new slot n_ctx_slot = 512
main: model loaded
main: chat template, chat_template: {% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '' + '\n' + tool['function']['arguments'] + '\n' + '' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '' + '\n' + tool['function']['arguments'] + '\n' + '' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜>'}}{% endif %}, example_format: 'You are a helpful assistant

<｜User｜>Hello<｜Assistant｜>Hi there<｜end▁of▁sentence｜><｜User｜>How are you?<｜Assistant｜>'
main: server is listening on http://0.0.0.0:8080 - starting the main loop
srv  update_slots: all slots are idle
slot launch_slot_: id  0 | task 0 | processing task
slot update_slots: id  0 | task 0 | new prompt, n_ctx_slot = 512, n_keep = 0, n_prompt_tokens = 9
slot update_slots: id  0 | task 0 | kv cache rm [0, end)
slot update_slots: id  0 | task 0 | prompt processing progress, n_past = 9, n_tokens = 9, progress = 1.000000
slot update_slots: id  0 | task 0 | prompt done, n_past = 9, n_tokens = 9
slot      release: id  0 | task 0 | stop processing: n_past = 23, truncated = 0
slot print_timing: id  0 | task 0 | 
prompt eval time =     264.49 ms /     9 tokens (   29.39 ms per token,    34.03 tokens per second)
       eval time =    2571.52 ms /    15 tokens (  171.43 ms per token,     **5.83 tokens per second**)
      total time =    2836.01 ms /    24 tokens
srv  update_slots: all slots are idle
srv  log_server_r: request: POST /completion 192.168.1.10 200
^Csrv    operator(): operator(): cleaning up before exit...
terminate called without an active exception
Aborted


 ./build/bin/llama-server -m /data4/qwen2.5-14b-deep-q4_k.gguf  -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047  -ngl 160  --host 0.0.0.0
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: NVIDIA GeForce GTX 1080 Ti, compute capability 6.1, VMM: yes
build: 0 (unknown) with cc (Debian 12.2.0-14) 12.2.0 for x86_64-linux-gnu
system info: n_threads = 24, n_threads_batch = 24, total_threads = 48

system_info: n_threads = 24 (n_threads_batch = 24) / 48 | CUDA : ARCHS = 610 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | LLAMAFILE = 1 | AARCH64_REPACK = 1 | 

main: HTTP server is listening, hostname: 0.0.0.0, port: 8080, http threads: 47
main: loading model
srv    load_model: loading model '/data4/qwen2.5-14b-deep-q4_k.gguf'
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce GTX 1080 Ti) - 11025 MiB free
llama_model_loader: loaded meta data with 25 key-value pairs and 579 tensors from /data4/qwen2.5-14b-deep-q4_k.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:                       general.architecture str              = qwen2
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Qw14
llama_model_loader: - kv   3:                         general.size_label str              = 15B
llama_model_loader: - kv   4:                          qwen2.block_count u32              = 48
llama_model_loader: - kv   5:                       qwen2.context_length u32              = 131072
llama_model_loader: - kv   6:                     qwen2.embedding_length u32              = 5120
llama_model_loader: - kv   7:                  qwen2.feed_forward_length u32              = 13824
llama_model_loader: - kv   8:                 qwen2.attention.head_count u32              = 40
llama_model_loader: - kv   9:              qwen2.attention.head_count_kv u32              = 8
llama_model_loader: - kv  10:                       qwen2.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  11:     qwen2.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  12:                          general.file_type u32              = 15
llama_model_loader: - kv  13:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  14:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  15:                      tokenizer.ggml.tokens arr[str,152064]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  16:                  tokenizer.ggml.token_type arr[i32,152064]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  17:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  18:                tokenizer.ggml.bos_token_id u32              = 151646
llama_model_loader: - kv  19:                tokenizer.ggml.eos_token_id u32              = 151643
llama_model_loader: - kv  20:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  21:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  22:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  23:                    tokenizer.chat_template str              = {% if not add_generation_prompt is de...
llama_model_loader: - kv  24:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  241 tensors
llama_model_loader: - type q4_K:  289 tensors
llama_model_loader: - type q6_K:   49 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q4_K - Medium
print_info: file size   = 8.37 GiB (4.87 BPW) 
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 22
load: token to piece cache size = 0.9310 MB
print_info: arch             = qwen2
print_info: vocab_only       = 0
print_info: n_ctx_train      = 131072
print_info: n_embd           = 5120
print_info: n_layer          = 48
print_info: n_head           = 40
print_info: n_head_kv        = 8
print_info: n_rot            = 128
print_info: n_swa            = 0
print_info: n_embd_head_k    = 128
print_info: n_embd_head_v    = 128
print_info: n_gqa            = 5
print_info: n_embd_k_gqa     = 1024
print_info: n_embd_v_gqa     = 1024
print_info: f_norm_eps       = 0.0e+00
print_info: f_norm_rms_eps   = 1.0e-05
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: n_ff             = 13824
print_info: n_expert         = 0
print_info: n_expert_used    = 0
print_info: causal attn      = 1
print_info: pooling type     = 0
print_info: rope type        = 2
print_info: rope scaling     = linear
print_info: freq_base_train  = 1000000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn  = 131072
print_info: rope_finetuned   = unknown
print_info: ssm_d_conv       = 0
print_info: ssm_d_inner      = 0
print_info: ssm_d_state      = 0
print_info: ssm_dt_rank      = 0
print_info: ssm_dt_b_c_rms   = 0
print_info: model type       = 14B
print_info: model params     = 14.77 B
print_info: general.name     = Qw14
print_info: vocab type       = BPE
print_info: n_vocab          = 152064
print_info: n_merges         = 151387
print_info: BOS token        = 151646 '<｜begin▁of▁sentence｜>'
print_info: EOS token        = 151643 '<｜end▁of▁sentence｜>'
print_info: EOT token        = 151643 '<｜end▁of▁sentence｜>'
print_info: PAD token        = 151643 '<｜end▁of▁sentence｜>'
print_info: LF token         = 198 'Ċ'
print_info: FIM PRE token    = 151659 '<|fim_prefix|>'
print_info: FIM SUF token    = 151661 '<|fim_suffix|>'
print_info: FIM MID token    = 151660 '<|fim_middle|>'
print_info: FIM PAD token    = 151662 '<|fim_pad|>'
print_info: FIM REP token    = 151663 '<|repo_name|>'
print_info: FIM SEP token    = 151664 '<|file_sep|>'
print_info: EOG token        = 151643 '<｜end▁of▁sentence｜>'
print_info: EOG token        = 151662 '<|fim_pad|>'
print_info: EOG token        = 151663 '<|repo_name|>'
print_info: EOG token        = 151664 '<|file_sep|>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 48 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 49/49 layers to GPU
load_tensors:        CUDA0 model buffer size =  8148.38 MiB
load_tensors:   CPU_Mapped model buffer size =   417.66 MiB
..........................................................................................
llama_init_from_model: n_seq_max     = 1
llama_init_from_model: n_ctx         = 512
llama_init_from_model: n_ctx_per_seq = 512
llama_init_from_model: n_batch       = 512
llama_init_from_model: n_ubatch      = 512
llama_init_from_model: flash_attn    = 1
llama_init_from_model: freq_base     = 1000000.0
llama_init_from_model: freq_scale    = 1
llama_init_from_model: n_ctx_per_seq (512) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 512, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 48, can_shift = 1
llama_kv_cache_init:      CUDA0 KV buffer size =    96.00 MiB
llama_init_from_model: KV self size  =   96.00 MiB, K (f16):   48.00 MiB, V (f16):   48.00 MiB
llama_init_from_model:  CUDA_Host  output buffer size =     0.58 MiB
llama_init_from_model:      CUDA0 compute buffer size =   307.00 MiB
llama_init_from_model:  CUDA_Host compute buffer size =    11.01 MiB
llama_init_from_model: graph nodes  = 1495
llama_init_from_model: graph splits = 2
common_init_from_params: setting dry_penalty_last_n to ctx_size = 512
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
srv          init: initializing slots, n_slots = 1
slot         init: id  0 | task -1 | new slot n_ctx_slot = 512
main: model loaded
main: chat template, chat_template: {% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '' + '\n' + tool['function']['arguments'] + '\n' + '' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '' + '\n' + tool['function']['arguments'] + '\n' + '' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜>'}}{% endif %}, example_format: 'You are a helpful assistant

<｜User｜>Hello<｜Assistant｜>Hi there<｜end▁of▁sentence｜><｜User｜>How are you?<｜Assistant｜>'
main: server is listening on http://0.0.0.0:8080 - starting the main loop
srv  update_slots: all slots are idle
slot launch_slot_: id  0 | task 0 | processing task
slot update_slots: id  0 | task 0 | new prompt, n_ctx_slot = 512, n_keep = 0, n_prompt_tokens = 9
slot update_slots: id  0 | task 0 | kv cache rm [0, end)
slot update_slots: id  0 | task 0 | prompt processing progress, n_past = 9, n_tokens = 9, progress = 1.000000
slot update_slots: id  0 | task 0 | prompt done, n_past = 9, n_tokens = 9
slot      release: id  0 | task 0 | stop processing: n_past = 23, truncated = 0
slot print_timing: id  0 | task 0 | 
prompt eval time =      78.55 ms /     9 tokens (    8.73 ms per token,   114.58 tokens per second)
       eval time =     579.88 ms /    15 tokens (   38.66 ms per token,    **25.87 tokens per second**)
      total time =     658.43 ms /    24 tokens
srv  update_slots: all slots are idle
srv  log_server_r: request: POST /completion 192.168.1.10 200
^Cterminate called without an active exception
srv    operator(): operator(): cleaning up before exit...
Aborted

[ggerganov]

Can you post the speed numbers by adding -t 1 to all 4 commands?

[zts9989]

Can you post the speed numbers by adding -t 1 to all 4 commands?

Of course! Your suggestion is an excellent debugging method indeed!

llama-server -m /data4/qwen2.5-14b-deep-q4_k.gguf -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -ngl 160 --host 0.0.0.0 -t 1
This parameter configuration achieves an inference speed of: 27.12 t/s

llama-server -m /data4/qwen2.5-14b-deep-q4_k.gguf -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -ngl 160 --host 0.0.0.0 -t 1 -nkvo
This parameter configuration achieves an inference speed of: 19.73 t/s

llama-cli -m /data4/qwen2.5-14b-deep-q4_k.gguf -cnv -p "You are a helpful assistant." -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -if -mli -ngl 160 -t 1
This parameter configuration achieves an inference speed of: 26.12 t/s

llama-cli -m /data4/qwen2.5-14b-deep-q4_k.gguf -cnv -p "You are a helpful assistant." -fa -c 512 --temp 0.6 --top-p 0.95 -s 3047 -if -mli -ngl 160 -nkvo -t 1
This parameter configuration achieves an inference speed of: 18.99 t/s

[zts9989]

I discovered another key point for reproducing the bug: When I compiled the program, I used:
cmake -B build -DGGML_CUDA=ON -DGGML_BUILD_NUMBER=3 -DGGML_OPENMP=OFF -DGGML_SCHED_MAX_COPIES=1

The GGML_OPENMP=OFF flag does not affect the performance of llama-cli, but it severely impacts the performance of llama-server.
Therefore, another critical factor for reproducing the bug is disabling OPENMP.

[ggerganov]

The GGML_OPENMP=OFF flag does not affect the performance of llama-cli, but it severely impacts the performance of llama-server.

I didn't get - does it impact it in a negative or in a positive way when you use GGML_OPENMP=OFF?

[zts9989]

Let me summarize all the test data (note the two lines marked with asterisks*):
When compiling with GGML_OPENMP=OFF:
llama-cli -t 24 24.27 t/s
llama-cli -t 24 -nkvo 18.25 t/s
llama-svr -t 24 25.87 t/s
llama-svr -t 24 -nkvo 5.83 t/s *******************
llama-cli -t 1 26.12 t/s
llama-cli -t 1 -nkvo 18.99 t/s
llama-svr -t 1 27.12 t/s
llama-svr -t 1 -nkvo 19.73 t/s

Additional tests with GGML_OPENMP=ON (24-thread only):
llama-svr -t 24 -nkvo 18.79 t/s ********************

Observations:

llama-cli and llama-svr appear to use thread pools differently.
llama-cli shows minimal sensitivity to the GGML_OPENMP flag.
llama-svr exhibits abnormal performance degradation with GGML_OPENMP=OFF.

[zts9989]

While conducting inference with Deepseek R1 671B Q4 #11397 (comment) , I discovered anomalous data in llama-server. I initially achieved up to 15t/s inference speed using llama-cli, but encountered a significant drop to ** #0.1t/s** when working with the API through llama-server. I subsequently replicated this performance degradation with the "-nkvo" parameter in multiple computing environments (compiled with GGML_OPENMP=OFF across all test systems). To demonstrate the issue more efficiently, I'm using the qwen2.5 14B model as an example here.

[jukofyork]

Could it be related to the -DGGML_SCHED_MAX_COPIES=1 for the --override-tensor option? Do you still get this without this?

[zts9989]

@jukofyork Thanks for your attention and suggestions. I recompiled and ran the program again. It seems to be unrelated to the options -DGGML_SCHED_MAX_COPIES and --override-tensor. It's only related to GGML_OPENMP.

cmake -B build3 -DGGML_CUDA=ON -DGGML_BUILD_NUMBER=3 -DGGML_OPENMP=OFF

   prompt eval time =     238.19 ms /     6 tokens (   39.70 ms per token,    25.19 tokens per second)
   eval time =    2662.75 ms /    16 tokens (  166.42 ms per token,     6.01 tokens per second)
   total time =    2900.94 ms /    22 tokens