Skip to content

Eval bug: Unusual high RAM usage on Windows when running DeepSeek V3 Q2_K_XL/IQ2_XXS, on Hybrid CPU+GPU (vs Linux). #12651

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
Panchovix opened this issue Mar 30, 2025 · 1 comment
Closed

Eval bug: Unusual high RAM usage on Windows when running DeepSeek V3 Q2_K_XL/IQ2_XXS, on Hybrid CPU+GPU (vs Linux). #12651

Panchovix opened this issue Mar 30, 2025 · 1 comment
Labels
bug-unconfirmed stale

Comments

@Panchovix
Copy link

Panchovix commented Mar 30, 2025 •
edited
Loading

Name and Version

.\llama-cli --version
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 4 CUDA devices:
  Device 0: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
  Device 1: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
  Device 2: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes
  Device 3: NVIDIA RTX A6000, compute capability 8.6, VMM: yes
version: 4992 (af6ae1ef)
built with MSVC 19.43.34809.0 for x64

Operating systems

Windows

GGML backends

CPU + CUDA

Hardware

AMD 7800X3D + 192GB RAM + RTX 5090 + RTX 4090 x2 + RTX A6000

Models

https://huggingface.co/unsloth/DeepSeek-V3-0324-GGUF/tree/main/UD-Q2_K_XL

https://huggingface.co/unsloth/DeepSeek-V3-0324-GGUF/tree/main/UD-IQ2_XXS

Problem description & steps to reproduce

Hi there, many thanks for all your work!

I have a hybrid setup that I have been testing lately, with the hardware mentioned before.

The issue consists that, loading the same model on Windows and Linux, the behavior is different. Using the command

./llama-server -m '/DeepSeek-V3-0324-UD-Q2_K_XL-merged.gguf' -c 4096-ngl 25 -ts 16,20,25,41 --no-warmup On Linux (Ubuntu 24.04, Fedora 41) and

.\llama-server.exe -m 'DeepSeek-V3-0324-UD-Q2_K_XL-merged.gguf' -c 4096 -ngl 25 -ts 16,20,25,41 --no-warmup on Windows (11).

On Linux: It starts correctly, loading the model and up to to gen. Then, the RAM usage ends near ~150GB. Then, when doing inference (via SillyTavern or http://127.0.0.1:8080), it takes a bit on pre processing but then it starts to gen soonish. RAM usage hovers around that usage an doesn't go up. ~3t/s.

On Windows: It starts correctly, loading the model and up to gen. Then, the RAM usage starts at ~120-130GB. Then, when doing inference, the RAM usage starts to go up slowly until it reaches ~190GB, and then Windows starts to use swap.. This makes the preprocessing really slow, and then on gen time, it is also very slow (~0.8 t/s). Also it seems the SSD used to load the model get constant read usage while RAM usage increases.

Note that the CUDA VRAM usage seems to be the same on both Windows and Linux.

RAM usage after loading the model

Image

RAM usage when sending a request to generate (first seconds, and you can notice I: disk usage where the model is located)

Image

RAM usage when it starts to use swap (and now C: Disk has usage)

Image

First Bad Commit

N/A

Relevant log output

PS X:\llama.cpp_windows\build\bin\Release> .\llama-server.exe -m 'DeepSeek-V3-0324-UD-IQ2_XXS-00001-of-00005.gguf' -c 4096 -ngl 25 -ts 16,20,25,41 --no-warmup
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 4 CUDA devices:
  Device 0: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
  Device 1: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
  Device 2: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes
  Device 3: NVIDIA RTX A6000, compute capability 8.6, VMM: yes
build: 4992 (af6ae1ef) with MSVC 19.43.34809.0 for x64
system info: n_threads = 8, n_threads_batch = 8, total_threads = 16

system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CUDA : ARCHS = 860,890,1200 | F16 = 1 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | AVX512 = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |

main: binding port with default address family
main: HTTP server is listening, hostname: 127.0.0.1, port: 8080, http threads: 15
main: loading model
srv    load_model: loading model 'DeepSeek-V3-0324-UD-IQ2_XXS-00001-of-00005.gguf'
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce RTX 4090) - 22754 MiB free
llama_model_load_from_file_impl: using device CUDA1 (NVIDIA GeForce RTX 4090) - 22754 MiB free
llama_model_load_from_file_impl: using device CUDA2 (NVIDIA GeForce RTX 5090) - 30620 MiB free
llama_model_load_from_file_impl: using device CUDA3 (NVIDIA RTX A6000) - 47108 MiB free
llama_model_loader: additional 4 GGUFs metadata loaded.
llama_model_loader: loaded meta data with 53 key-value pairs and 1025 tensors from G:\GGUFs\UD-IQ2_XXS\DeepSeek-V3-0324-UD-IQ2_XXS-00001-of-00005.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              = deepseek2
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = DeepSeek V3 0324 BF16
llama_model_loader: - kv   3:                       general.quantized_by str              = Unsloth
llama_model_loader: - kv   4:                         general.size_label str              = 256x20B
llama_model_loader: - kv   5:                            general.license str              = mit
llama_model_loader: - kv   6:                           general.repo_url str              = https://huggingface.co/unsloth
llama_model_loader: - kv   7:                      deepseek2.block_count u32              = 61
llama_model_loader: - kv   8:                   deepseek2.context_length u32              = 163840
llama_model_loader: - kv   9:                 deepseek2.embedding_length u32              = 7168
llama_model_loader: - kv  10:              deepseek2.feed_forward_length u32              = 18432
llama_model_loader: - kv  11:             deepseek2.attention.head_count u32              = 128
llama_model_loader: - kv  12:          deepseek2.attention.head_count_kv u32              = 128
llama_model_loader: - kv  13:                   deepseek2.rope.freq_base f32              = 10000.000000
llama_model_loader: - kv  14: deepseek2.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  15:                deepseek2.expert_used_count u32              = 8
llama_model_loader: - kv  16:        deepseek2.leading_dense_block_count u32              = 3
llama_model_loader: - kv  17:                       deepseek2.vocab_size u32              = 129280
llama_model_loader: - kv  18:            deepseek2.attention.q_lora_rank u32              = 1536
llama_model_loader: - kv  19:           deepseek2.attention.kv_lora_rank u32              = 512
llama_model_loader: - kv  20:             deepseek2.attention.key_length u32              = 192
llama_model_loader: - kv  21:           deepseek2.attention.value_length u32              = 128
llama_model_loader: - kv  22:       deepseek2.expert_feed_forward_length u32              = 2048
llama_model_loader: - kv  23:                     deepseek2.expert_count u32              = 256
llama_model_loader: - kv  24:              deepseek2.expert_shared_count u32              = 1
llama_model_loader: - kv  25:             deepseek2.expert_weights_scale f32              = 2.500000
llama_model_loader: - kv  26:              deepseek2.expert_weights_norm bool             = true
llama_model_loader: - kv  27:               deepseek2.expert_gating_func u32              = 2
llama_model_loader: - kv  28:             deepseek2.rope.dimension_count u32              = 64
llama_model_loader: - kv  29:                deepseek2.rope.scaling.type str              = yarn
llama_model_loader: - kv  30:              deepseek2.rope.scaling.factor f32              = 40.000000
llama_model_loader: - kv  31: deepseek2.rope.scaling.original_context_length u32              = 4096
llama_model_loader: - kv  32: deepseek2.rope.scaling.yarn_log_multiplier f32              = 0.100000
llama_model_loader: - kv  33:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  34:                         tokenizer.ggml.pre str              = deepseek-v3
llama_model_loader: - kv  35:                      tokenizer.ggml.tokens arr[str,129280]  = ["<∩╜£beginΓûüofΓûüsentence∩╜£>", "<∩...
llama_model_loader: - kv  36:                  tokenizer.ggml.token_type arr[i32,129280]  = [3, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  37:                      tokenizer.ggml.merges arr[str,127741]  = ["─á t", "─á a", "i n", "─á ─á", "h e...
llama_model_loader: - kv  38:                tokenizer.ggml.bos_token_id u32              = 0
llama_model_loader: - kv  39:                tokenizer.ggml.eos_token_id u32              = 1
llama_model_loader: - kv  40:            tokenizer.ggml.padding_token_id u32              = 1
llama_model_loader: - kv  41:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  42:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  43:                    tokenizer.chat_template str              = {% if not add_generation_prompt is de...
llama_model_loader: - kv  44:               general.quantization_version u32              = 2
llama_model_loader: - kv  45:                          general.file_type u32              = 19
llama_model_loader: - kv  46:                      quantize.imatrix.file str              = DeepSeek-V3-0324-GGUF/DeepSeek-V3-032...
llama_model_loader: - kv  47:                   quantize.imatrix.dataset str              = /workspace/calibration_datav3.txt
llama_model_loader: - kv  48:             quantize.imatrix.entries_count i32              = 720
llama_model_loader: - kv  49:              quantize.imatrix.chunks_count i32              = 124
llama_model_loader: - kv  50:                                   split.no u16              = 0
llama_model_loader: - kv  51:                        split.tensors.count i32              = 1025
llama_model_loader: - kv  52:                                split.count u16              = 5
llama_model_loader: - type  f32:  361 tensors
llama_model_loader: - type q3_K:   55 tensors
llama_model_loader: - type q4_K:  193 tensors
llama_model_loader: - type q5_K:  116 tensors
llama_model_loader: - type q6_K:  184 tensors
llama_model_loader: - type iq2_xxs:  116 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = IQ2_XXS - 2.0625 bpw
print_info: file size   = 203.63 GiB (2.61 BPW)
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 818
load: token to piece cache size = 0.8223 MB
print_info: arch             = deepseek2
print_info: vocab_only       = 0
print_info: n_ctx_train      = 163840
print_info: n_embd           = 7168
print_info: n_layer          = 61
print_info: n_head           = 128
print_info: n_head_kv        = 128
print_info: n_rot            = 64
print_info: n_swa            = 0
print_info: n_swa_pattern    = 1
print_info: n_embd_head_k    = 192
print_info: n_embd_head_v    = 128
print_info: n_gqa            = 1
print_info: n_embd_k_gqa     = 24576
print_info: n_embd_v_gqa     = 16384
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             = 18432
print_info: n_expert         = 256
print_info: n_expert_used    = 8
print_info: causal attn      = 1
print_info: pooling type     = 0
print_info: rope type        = 0
print_info: rope scaling     = yarn
print_info: freq_base_train  = 10000.0
print_info: freq_scale_train = 0.025
print_info: n_ctx_orig_yarn  = 4096
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       = 671B
print_info: model params     = 671.03 B
print_info: general.name     = DeepSeek V3 0324 BF16
print_info: n_layer_dense_lead   = 3
print_info: n_lora_q             = 1536
print_info: n_lora_kv            = 512
print_info: n_ff_exp             = 2048
print_info: n_expert_shared      = 1
print_info: expert_weights_scale = 2.5
print_info: expert_weights_norm  = 1
print_info: expert_gating_func   = sigmoid
print_info: rope_yarn_log_mul    = 0.1000
print_info: vocab type       = BPE
print_info: n_vocab          = 129280
print_info: n_merges         = 127741
print_info: BOS token        = 0 '<｜begin▁of▁sentence｜>'
print_info: EOS token        = 1 '<｜end▁of▁sentence｜>'
print_info: EOT token        = 1 '<｜end▁of▁sentence｜>'
print_info: PAD token        = 1 '<｜end▁of▁sentence｜>'
print_info: LF token         = 201 '─è'
print_info: FIM PRE token    = 128801 '<｜fim▁begin｜>'
print_info: FIM SUF token    = 128800 '<｜fim▁hole｜>'
print_info: FIM MID token    = 128802 '<｜fim▁end｜>'
print_info: EOG token        = 1 '<｜end▁of▁sentence｜>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 25 repeating layers to GPU
load_tensors: offloaded 25/62 layers to GPU
load_tensors:        CUDA0 model buffer size = 14125.24 MiB
load_tensors:        CUDA1 model buffer size = 17656.55 MiB
load_tensors:        CUDA2 model buffer size = 21187.85 MiB
load_tensors:        CUDA3 model buffer size = 35313.09 MiB
load_tensors:   CPU_Mapped model buffer size = 46183.39 MiB
load_tensors:   CPU_Mapped model buffer size = 47458.50 MiB
load_tensors:   CPU_Mapped model buffer size = 26593.44 MiB
....................................................................................................
llama_context: constructing llama_context
llama_context: n_seq_max     = 1
llama_context: n_ctx         = 4096
llama_context: n_ctx_per_seq = 4096
llama_context: n_batch       = 2048
llama_context: n_ubatch      = 512
llama_context: causal_attn   = 1
llama_context: flash_attn    = 0
llama_context: freq_base     = 10000.0
llama_context: freq_scale    = 0.025
llama_context: n_ctx_per_seq (4096) < n_ctx_train (163840) -- the full capacity of the model will not be utilized
llama_context:        CPU  output buffer size =     0.49 MiB
init: kv_size = 4096, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 61, can_shift = 0
init:      CUDA0 KV buffer size =  1280.00 MiB
init:      CUDA1 KV buffer size =  1600.00 MiB
init:      CUDA2 KV buffer size =  1920.00 MiB
init:      CUDA3 KV buffer size =  3200.00 MiB
init:        CPU KV buffer size = 11520.00 MiB
llama_context: KV self size  = 19520.00 MiB, K (f16): 11712.00 MiB, V (f16): 7808.00 MiB
llama_context:      CUDA0 compute buffer size =  3014.00 MiB
llama_context:      CUDA1 compute buffer size =  1186.00 MiB
llama_context:      CUDA2 compute buffer size =  1186.00 MiB
llama_context:      CUDA3 compute buffer size =  1186.00 MiB
llama_context:  CUDA_Host compute buffer size =    88.01 MiB
llama_context: graph nodes  = 5086
llama_context: graph splits = 676 (with bs=512), 6 (with bs=1)
common_init_from_params: KV cache shifting is not supported for this context, disabling KV cache shifting
common_init_from_params: setting dry_penalty_last_n to ctx_size = 4096
srv          init: initializing slots, n_slots = 1
slot         init: id  0 | task -1 | new slot n_ctx_slot = 4096
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='', is_first_sp=true, is_last_user=false) %}{%- for message in messages %}{%- if message['role'] == 'system' %}{%- if ns.is_first_sp %}{% set ns.system_prompt = ns.system_prompt + message['content'] %}{% set ns.is_first_sp = false %}{%- else %}{% set ns.system_prompt = ns.system_prompt + '

' + message['content'] %}{%- endif %}{%- endif %}{%- endfor %}{{ bos_token }}{{ ns.system_prompt }}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{%- set ns.is_first = false -%}{%- set ns.is_last_user = true -%}{{'<｜User｜>' + message['content'] + '<｜Assistant｜>'}}{%- endif %}{%- if message['role'] == 'assistant' and message['tool_calls'] is defined and message['tool_calls'] is not none %}{%- set ns.is_last_user = false -%}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{%- endif %}{%- set ns.is_first = false %}{%- set ns.is_tool = false -%}{%- set ns.is_output_first = true %}{%- for tool in message['tool_calls'] %}{%- if not ns.is_first %}{%- if message['content'] is none %}{{'<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '
' + '' + '
' + tool['function']['arguments'] + '
' + '' + '<｜tool▁call▁end｜>'}}{%- else %}{{message['content'] + '<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '
' + '' + '
' + tool['function']['arguments'] + '
' + '' + '<｜tool▁call▁end｜>'}}{%- endif %}{%- set ns.is_first = true -%}{%- else %}{{'
' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '
' + '' + '
' + tool['function']['arguments'] + '
' + '' + '<｜tool▁call▁end｜>'}}{%- endif %}{%- endfor %}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- if message['role'] == 'assistant' and (message['tool_calls'] is not defined or message['tool_calls'] is none)%}{%- set ns.is_last_user = false -%}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{{content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_last_user = false -%}{%- 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 %}{{'
<｜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_last_user 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://127.0.0.1:8080 - starting the main loop
srv  update_slots: all slots are idle
srv  log_server_r: request: GET /v1/models 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
slot launch_slot_: id  0 | task 0 | processing task
slot update_slots: id  0 | task 0 | new prompt, n_ctx_slot = 4096, n_keep = 0, n_prompt_tokens = 2188
slot update_slots: id  0 | task 0 | kv cache rm [0, end)
slot update_slots: id  0 | task 0 | prompt processing progress, n_past = 2048, n_tokens = 2048, progress = 0.936015
slot update_slots: id  0 | task 0 | kv cache rm [2048, end)
slot update_slots: id  0 | task 0 | prompt processing progress, n_past = 2188, n_tokens = 140, progress = 1.000000
slot update_slots: id  0 | task 0 | prompt done, n_past = 2188, n_tokens = 140
srv  log_server_r: request: POST /tokenize 127.0.0.1 200
slot      release: id  0 | task 0 | stop processing: n_past = 2572, truncated = 0
slot print_timing: id  0 | task 0 |
prompt eval time =  535821.65 ms /  2188 tokens (  244.89 ms per token,     4.08 tokens per second)
PS X:\llama.cpp_windows\build\bin\Release> .\llama-server.exe -m 'DeepSeek-V3-0324-UD-Q2_K_XL-merged.gguf' -c 4096 -ngl 25 -ts 16,20,25,41 --no-warmup
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 4 CUDA devices:
  Device 0: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
  Device 1: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
  Device 2: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes
  Device 3: NVIDIA RTX A6000, compute capability 8.6, VMM: yes
build: 4992 (af6ae1ef) with MSVC 19.43.34809.0 for x64
system info: n_threads = 8, n_threads_batch = 8, total_threads = 16

system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CUDA : ARCHS = 860,890,1200 | F16 = 1 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | AVX512 = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |

main: binding port with default address family
main: HTTP server is listening, hostname: 127.0.0.1, port: 8080, http threads: 15
main: loading model
srv    load_model: loading model 'DeepSeek-V3-0324-UD-Q2_K_XL-merged.gguf'
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce RTX 4090) - 22754 MiB free
llama_model_load_from_file_impl: using device CUDA1 (NVIDIA GeForce RTX 4090) - 22754 MiB free
llama_model_load_from_file_impl: using device CUDA2 (NVIDIA GeForce RTX 5090) - 30620 MiB free
llama_model_load_from_file_impl: using device CUDA3 (NVIDIA RTX A6000) - 47108 MiB free
llama_model_loader: loaded meta data with 49 key-value pairs and 1025 tensors from I:\HuggingFaceModelDownloader\Storage\GGUFs\DeepSeek-V3-0324-UD-Q2_K_XL-merged.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              = deepseek2
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = DeepSeek V3 0324 BF16
llama_model_loader: - kv   3:                       general.quantized_by str              = Unsloth
llama_model_loader: - kv   4:                         general.size_label str              = 256x20B
llama_model_loader: - kv   5:                            general.license str              = mit
llama_model_loader: - kv   6:                           general.repo_url str              = https://huggingface.co/unsloth
llama_model_loader: - kv   7:                      deepseek2.block_count u32              = 61
llama_model_loader: - kv   8:                   deepseek2.context_length u32              = 163840
llama_model_loader: - kv   9:                 deepseek2.embedding_length u32              = 7168
llama_model_loader: - kv  10:              deepseek2.feed_forward_length u32              = 18432
llama_model_loader: - kv  11:             deepseek2.attention.head_count u32              = 128
llama_model_loader: - kv  12:          deepseek2.attention.head_count_kv u32              = 128
llama_model_loader: - kv  13:                   deepseek2.rope.freq_base f32              = 10000.000000
llama_model_loader: - kv  14: deepseek2.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  15:                deepseek2.expert_used_count u32              = 8
llama_model_loader: - kv  16:        deepseek2.leading_dense_block_count u32              = 3
llama_model_loader: - kv  17:                       deepseek2.vocab_size u32              = 129280
llama_model_loader: - kv  18:            deepseek2.attention.q_lora_rank u32              = 1536
llama_model_loader: - kv  19:           deepseek2.attention.kv_lora_rank u32              = 512
llama_model_loader: - kv  20:             deepseek2.attention.key_length u32              = 192
llama_model_loader: - kv  21:           deepseek2.attention.value_length u32              = 128
llama_model_loader: - kv  22:       deepseek2.expert_feed_forward_length u32              = 2048
llama_model_loader: - kv  23:                     deepseek2.expert_count u32              = 256
llama_model_loader: - kv  24:              deepseek2.expert_shared_count u32              = 1
llama_model_loader: - kv  25:             deepseek2.expert_weights_scale f32              = 2.500000
llama_model_loader: - kv  26:              deepseek2.expert_weights_norm bool             = true
llama_model_loader: - kv  27:               deepseek2.expert_gating_func u32              = 2
llama_model_loader: - kv  28:             deepseek2.rope.dimension_count u32              = 64
llama_model_loader: - kv  29:                deepseek2.rope.scaling.type str              = yarn
llama_model_loader: - kv  30:              deepseek2.rope.scaling.factor f32              = 40.000000
llama_model_loader: - kv  31: deepseek2.rope.scaling.original_context_length u32              = 4096
llama_model_loader: - kv  32: deepseek2.rope.scaling.yarn_log_multiplier f32              = 0.100000
llama_model_loader: - kv  33:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  34:                         tokenizer.ggml.pre str              = deepseek-v3
llama_model_loader: - kv  35:                      tokenizer.ggml.tokens arr[str,129280]  = ["<∩╜£beginΓûüofΓûüsentence∩╜£>", "<∩...
llama_model_loader: - kv  36:                  tokenizer.ggml.token_type arr[i32,129280]  = [3, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  37:                      tokenizer.ggml.merges arr[str,127741]  = ["─á t", "─á a", "i n", "─á ─á", "h e...
llama_model_loader: - kv  38:                tokenizer.ggml.bos_token_id u32              = 0
llama_model_loader: - kv  39:                tokenizer.ggml.eos_token_id u32              = 1
llama_model_loader: - kv  40:            tokenizer.ggml.padding_token_id u32              = 1
llama_model_loader: - kv  41:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  42:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  43:                    tokenizer.chat_template str              = {% if not add_generation_prompt is de...
llama_model_loader: - kv  44:               general.quantization_version u32              = 2
llama_model_loader: - kv  45:                          general.file_type u32              = 10
llama_model_loader: - kv  46:                                   split.no u16              = 0
llama_model_loader: - kv  47:                        split.tensors.count i32              = 1025
llama_model_loader: - kv  48:                                split.count u16              = 0
llama_model_loader: - type  f32:  361 tensors
llama_model_loader: - type q2_K:  116 tensors
llama_model_loader: - type q3_K:   58 tensors
llama_model_loader: - type q4_K:  306 tensors
llama_model_loader: - type q6_K:  184 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q2_K - Medium
print_info: file size   = 230.59 GiB (2.95 BPW)
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 818
load: token to piece cache size = 0.8223 MB
print_info: arch             = deepseek2
print_info: vocab_only       = 0
print_info: n_ctx_train      = 163840
print_info: n_embd           = 7168
print_info: n_layer          = 61
print_info: n_head           = 128
print_info: n_head_kv        = 128
print_info: n_rot            = 64
print_info: n_swa            = 0
print_info: n_swa_pattern    = 1
print_info: n_embd_head_k    = 192
print_info: n_embd_head_v    = 128
print_info: n_gqa            = 1
print_info: n_embd_k_gqa     = 24576
print_info: n_embd_v_gqa     = 16384
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             = 18432
print_info: n_expert         = 256
print_info: n_expert_used    = 8
print_info: causal attn      = 1
print_info: pooling type     = 0
print_info: rope type        = 0
print_info: rope scaling     = yarn
print_info: freq_base_train  = 10000.0
print_info: freq_scale_train = 0.025
print_info: n_ctx_orig_yarn  = 4096
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       = 671B
print_info: model params     = 671.03 B
print_info: general.name     = DeepSeek V3 0324 BF16
print_info: n_layer_dense_lead   = 3
print_info: n_lora_q             = 1536
print_info: n_lora_kv            = 512
print_info: n_ff_exp             = 2048
print_info: n_expert_shared      = 1
print_info: expert_weights_scale = 2.5
print_info: expert_weights_norm  = 1
print_info: expert_gating_func   = sigmoid
print_info: rope_yarn_log_mul    = 0.1000
print_info: vocab type       = BPE
print_info: n_vocab          = 129280
print_info: n_merges         = 127741
print_info: BOS token        = 0 '<｜begin▁of▁sentence｜>'
print_info: EOS token        = 1 '<｜end▁of▁sentence｜>'
print_info: EOT token        = 1 '<｜end▁of▁sentence｜>'
print_info: PAD token        = 1 '<｜end▁of▁sentence｜>'
print_info: LF token         = 201 '─è'
print_info: FIM PRE token    = 128801 '<｜fim▁begin｜>'
print_info: FIM SUF token    = 128800 '<｜fim▁hole｜>'
print_info: FIM MID token    = 128802 '<｜fim▁end｜>'
print_info: EOG token        = 1 '<｜end▁of▁sentence｜>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 25 repeating layers to GPU
load_tensors: offloaded 25/62 layers to GPU
load_tensors:        CUDA0 model buffer size = 16127.24 MiB
load_tensors:        CUDA1 model buffer size = 20159.05 MiB
load_tensors:        CUDA2 model buffer size = 24190.85 MiB
load_tensors:        CUDA3 model buffer size = 40318.09 MiB
load_tensors:   CPU_Mapped model buffer size = 135323.84 MiB
....................................................................................................
llama_context: constructing llama_context
llama_context: n_seq_max     = 1
llama_context: n_ctx         = 4096
llama_context: n_ctx_per_seq = 4096
llama_context: n_batch       = 2048
llama_context: n_ubatch      = 512
llama_context: causal_attn   = 1
llama_context: flash_attn    = 0
llama_context: freq_base     = 10000.0
llama_context: freq_scale    = 0.025
llama_context: n_ctx_per_seq (4096) < n_ctx_train (163840) -- the full capacity of the model will not be utilized
llama_context:        CPU  output buffer size =     0.49 MiB
init: kv_size = 4096, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 61, can_shift = 0
init:      CUDA0 KV buffer size =  1280.00 MiB
init:      CUDA1 KV buffer size =  1600.00 MiB
init:      CUDA2 KV buffer size =  1920.00 MiB
init:      CUDA3 KV buffer size =  3200.00 MiB
init:        CPU KV buffer size = 11520.00 MiB
llama_context: KV self size  = 19520.00 MiB, K (f16): 11712.00 MiB, V (f16): 7808.00 MiB
llama_context:      CUDA0 compute buffer size =  2790.00 MiB
llama_context:      CUDA1 compute buffer size =  1186.00 MiB
llama_context:      CUDA2 compute buffer size =  1186.00 MiB
llama_context:      CUDA3 compute buffer size =  1186.00 MiB
llama_context:  CUDA_Host compute buffer size =    88.01 MiB
llama_context: graph nodes  = 5086
llama_context: graph splits = 676 (with bs=512), 6 (with bs=1)
common_init_from_params: KV cache shifting is not supported for this context, disabling KV cache shifting
common_init_from_params: setting dry_penalty_last_n to ctx_size = 4096
srv          init: initializing slots, n_slots = 1
slot         init: id  0 | task -1 | new slot n_ctx_slot = 4096
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='', is_first_sp=true, is_last_user=false) %}{%- for message in messages %}{%- if message['role'] == 'system' %}{%- if ns.is_first_sp %}{% set ns.system_prompt = ns.system_prompt + message['content'] %}{% set ns.is_first_sp = false %}{%- else %}{% set ns.system_prompt = ns.system_prompt + '

' + message['content'] %}{%- endif %}{%- endif %}{%- endfor %}{{ bos_token }}{{ ns.system_prompt }}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{%- set ns.is_first = false -%}{%- set ns.is_last_user = true -%}{{'<｜User｜>' + message['content'] + '<｜Assistant｜>'}}{%- endif %}{%- if message['role'] == 'assistant' and message['tool_calls'] is defined and message['tool_calls'] is not none %}{%- set ns.is_last_user = false -%}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{%- endif %}{%- set ns.is_first = false %}{%- set ns.is_tool = false -%}{%- set ns.is_output_first = true %}{%- for tool in message['tool_calls'] %}{%- if not ns.is_first %}{%- if message['content'] is none %}{{'<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '
' + '```json' + '
' + tool['function']['arguments'] + '
' + '```' + '<｜tool▁call▁end｜>'}}{%- else %}{{message['content'] + '<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '
' + '```json' + '
' + tool['function']['arguments'] + '
' + '```' + '<｜tool▁call▁end｜>'}}{%- endif %}{%- set ns.is_first = true -%}{%- else %}{{'
' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '
' + '```json' + '
' + tool['function']['arguments'] + '
' + '```' + '<｜tool▁call▁end｜>'}}{%- endif %}{%- endfor %}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- if message['role'] == 'assistant' and (message['tool_calls'] is not defined or message['tool_calls'] is none)%}{%- set ns.is_last_user = false -%}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{{content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_last_user = false -%}{%- 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 %}{{'
<｜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_last_user 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://127.0.0.1:8080 - starting the main loop
@Panchovix Panchovix changed the title Eval bug: llama.cpp on Windows, for DeepSeek V3 Q2_K_XL uses too much RAM vs Linux (Hybrid GPU + CPU) and starts using swap. Eval bug: llama.cpp on Windows, for DeepSeek V3 Q2_K_XL/IQ2_XXS uses too much RAM vs Linux (Hybrid GPU + CPU) and starts using swap. Mar 30, 2025
@Panchovix Panchovix changed the title Eval bug: llama.cpp on Windows, for DeepSeek V3 Q2_K_XL/IQ2_XXS uses too much RAM vs Linux (Hybrid GPU + CPU) and starts using swap. Eval bug: Unusual high RAM usage on Windows when running DeepSeek V3 Q2_K_XL/IQ2_XXS, on Hybrid CPU+GPU (vs Linux). Mar 30, 2025
@github-actions github-actions bot added the stale label Apr 30, 2025
@github-actions GitHub Actions
Copy link
Contributor

This issue was closed because it has been inactive for 14 days since being marked as stale.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
bug-unconfirmed stale
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@Panchovix