github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #9311] llama3.2-vision really slow when already in VRAM - high load duration #6076

New Issue
Closed
opened 2026-04-12 17:24:18 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-04-12 17:24:18 -05:00
Owner
Copy Link

Originally created by @ribbles on GitHub (Feb 24, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9311

What is the issue?

When chatting with llama3.2-vision subsequent requests are not as fast as I would expect as the model is already 100% loaded into VRAM.

I don't know if it's relevant but the "load_duration" is really high on calls after the model is already loaded, when I would expect it to be zero.
I also see the logs reporting 8 cores but the GPU has 3072 cores - no idea if relevant.

First request:

{'total_duration': 57086256700, 'load_duration': 54109228500, 'prompt_eval_count': 383, 'prompt_eval_duration': 15000000, 'eval_count': 65, 'eval_duration': 2958000000}

Second request:

{'total_duration': 91264439700, 'load_duration': 87957377600, 'prompt_eval_count': 469, 'prompt_eval_duration': 234000000, 'eval_count': 64, 'eval_duration': 3063000000}

Relevant log output

2025/02/23 22:16:30 routes.go:1186: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://0.0.0.0:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:C:\\Users\\User\\.ollama\\models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[* http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:]"
time=2025-02-23T22:16:30.279-08:00 level=INFO source=images.go:432 msg="total blobs: 19"
time=2025-02-23T22:16:30.286-08:00 level=INFO source=images.go:439 msg="total unused blobs removed: 0"
time=2025-02-23T22:16:30.293-08:00 level=INFO source=routes.go:1237 msg="Listening on [::]:11434 (version 0.5.11)"
time=2025-02-23T22:16:30.294-08:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs"
time=2025-02-23T22:16:30.294-08:00 level=INFO source=gpu_windows.go:167 msg=packages count=2
time=2025-02-23T22:16:30.294-08:00 level=INFO source=gpu_windows.go:214 msg="" package=0 cores=8 efficiency=0 threads=16
time=2025-02-23T22:16:30.294-08:00 level=INFO source=gpu_windows.go:214 msg="" package=1 cores=8 efficiency=0 threads=16
time=2025-02-23T22:16:30.723-08:00 level=INFO source=gpu.go:319 msg="detected OS VRAM overhead" id=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b library=cuda compute=5.2 driver=12.7 name="Tesla M40 24GB" overhead="716.4 MiB"
time=2025-02-23T22:16:30.729-08:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b library=cuda variant=v11 compute=5.2 driver=12.7 name="Tesla M40 24GB" total="22.5 GiB" available="21.6 GiB"
[GIN] 2025/02/23 - 22:16:32 | 200 |       538.1µs |       127.0.0.1 | HEAD     "/"
[GIN] 2025/02/23 - 22:21:20 | 200 |       157.8µs |       127.0.0.1 | HEAD     "/"
[GIN] 2025/02/23 - 22:21:20 | 200 |     75.5925ms |       127.0.0.1 | GET      "/api/tags"
[GIN] 2025/02/23 - 22:21:29 | 200 |            0s |       127.0.0.1 | HEAD     "/"
[GIN] 2025/02/23 - 22:21:29 | 200 |    125.2651ms |       127.0.0.1 | POST     "/api/show"
time=2025-02-23T22:21:29.557-08:00 level=WARN source=sched.go:137 msg="mllama doesn't support parallel requests yet"
time=2025-02-23T22:21:29.758-08:00 level=INFO source=sched.go:714 msg="new model will fit in available VRAM in single GPU, loading" model=C:\Users\User\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 gpu=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b parallel=1 available=23197925376 required="11.3 GiB"
time=2025-02-23T22:21:29.798-08:00 level=INFO source=server.go:100 msg="system memory" total="63.9 GiB" free="58.8 GiB" free_swap="68.0 GiB"
time=2025-02-23T22:21:29.810-08:00 level=INFO source=memory.go:356 msg="offload to cuda" projector.weights="1.8 GiB" projector.graph="2.8 GiB" layers.requested=-1 layers.model=41 layers.offload=41 layers.split="" memory.available="[21.6 GiB]" memory.gpu_overhead="0 B" memory.required.full="11.3 GiB" memory.required.partial="11.3 GiB" memory.required.kv="656.2 MiB" memory.required.allocations="[11.3 GiB]" memory.weights.total="5.5 GiB" memory.weights.repeating="5.1 GiB" memory.weights.nonrepeating="411.0 MiB" memory.graph.full="258.5 MiB" memory.graph.partial="669.5 MiB"
time=2025-02-23T22:21:29.840-08:00 level=INFO source=server.go:380 msg="starting llama server" cmd="C:\\Users\\User\\AppData\\Local\\Programs\\Ollama\\ollama.exe runner --model C:\\Users\\User\\.ollama\\models\\blobs\\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 --ctx-size 2048 --batch-size 512 --n-gpu-layers 41 --mmproj C:\\Users\\User\\.ollama\\models\\blobs\\sha256-ece5e659647a20a5c28ab9eea1c12a1ad430bc0f2a27021d00ad103b3bf5206f --threads 16 --no-mmap --parallel 1 --port 58093"
time=2025-02-23T22:21:29.851-08:00 level=INFO source=sched.go:449 msg="loaded runners" count=1
time=2025-02-23T22:21:29.851-08:00 level=INFO source=server.go:557 msg="waiting for llama runner to start responding"
time=2025-02-23T22:21:29.853-08:00 level=INFO source=server.go:591 msg="waiting for server to become available" status="llm server error"
time=2025-02-23T22:21:29.946-08:00 level=INFO source=runner.go:936 msg="starting go runner"
time=2025-02-23T22:21:29.948-08:00 level=INFO source=runner.go:937 msg=system info="CPU : LLAMAFILE = 1 | CPU : LLAMAFILE = 1 | cgo(clang)" threads=16
time=2025-02-23T22:21:29.950-08:00 level=INFO source=runner.go:995 msg="Server listening on 127.0.0.1:58093"
time=2025-02-23T22:21:30.107-08:00 level=INFO source=server.go:591 msg="waiting for server to become available" status="llm server loading model"
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: Tesla M40 24GB, compute capability 5.2, VMM: yes
load_backend: loaded CUDA backend from C:\Users\User\AppData\Local\Programs\Ollama\lib\ollama\cuda_v11\ggml-cuda.dll
load_backend: loaded CPU backend from C:\Users\User\AppData\Local\Programs\Ollama\lib\ollama\ggml-cpu-sandybridge.dll
llama_load_model_from_file: using device CUDA0 (Tesla M40 24GB) - 22081 MiB free
llama_model_loader: loaded meta data with 27 key-value pairs and 396 tensors from C:\Users\User\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 (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              = mllama
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Model
llama_model_loader: - kv   3:                         general.size_label str              = 10B
llama_model_loader: - kv   4:                         mllama.block_count u32              = 40
llama_model_loader: - kv   5:                      mllama.context_length u32              = 131072
llama_model_loader: - kv   6:                    mllama.embedding_length u32              = 4096
llama_model_loader: - kv   7:                 mllama.feed_forward_length u32              = 14336
llama_model_loader: - kv   8:                mllama.attention.head_count u32              = 32
llama_model_loader: - kv   9:             mllama.attention.head_count_kv u32              = 8
llama_model_loader: - kv  10:                      mllama.rope.freq_base f32              = 500000.000000
llama_model_loader: - kv  11:    mllama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  12:                          general.file_type u32              = 15
llama_model_loader: - kv  13:                          mllama.vocab_size u32              = 128256
llama_model_loader: - kv  14:                mllama.rope.dimension_count u32              = 128
llama_model_loader: - kv  15:    mllama.attention.cross_attention_layers arr[i32,8]       = [3, 8, 13, 18, 23, 28, 33, 38]
llama_model_loader: - kv  16:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  17:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  18:                         tokenizer.ggml.pre str              = llama-bpe
llama_model_loader: - kv  19:                      tokenizer.ggml.tokens arr[str,128257]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  20:                  tokenizer.ggml.token_type arr[i32,128257]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  21:                      tokenizer.ggml.merges arr[str,280147]  = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "...
llama_model_loader: - kv  22:                tokenizer.ggml.bos_token_id u32              = 128000
llama_model_loader: - kv  23:                tokenizer.ggml.eos_token_id u32              = 128009
llama_model_loader: - kv  24:            tokenizer.ggml.padding_token_id u32              = 128004
llama_model_loader: - kv  25:                    tokenizer.chat_template str              = {{- bos_token }}\n{%- if custom_tools ...
llama_model_loader: - kv  26:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  114 tensors
llama_model_loader: - type q4_K:  245 tensors
llama_model_loader: - type q6_K:   37 tensors
llm_load_vocab: special tokens cache size = 257
llm_load_vocab: token to piece cache size = 0.7999 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = mllama
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 128256
llm_load_print_meta: n_merges         = 280147
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 131072
llm_load_print_meta: n_embd           = 4096
llm_load_print_meta: n_layer          = 40
llm_load_print_meta: n_head           = 32
llm_load_print_meta: n_head_kv        = 8
llm_load_print_meta: n_rot            = 128
llm_load_print_meta: n_swa            = 0
llm_load_print_meta: n_embd_head_k    = 128
llm_load_print_meta: n_embd_head_v    = 128
llm_load_print_meta: n_gqa            = 4
llm_load_print_meta: n_embd_k_gqa     = 1024
llm_load_print_meta: n_embd_v_gqa     = 1024
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
llm_load_print_meta: f_clamp_kqv      = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale    = 0.0e+00
llm_load_print_meta: n_ff             = 14336
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0
llm_load_print_meta: causal attn      = 1
llm_load_print_meta: pooling type     = 0
llm_load_print_meta: rope type        = 0
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 500000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 131072
llm_load_print_meta: rope_finetuned   = unknown
llm_load_print_meta: ssm_d_conv       = 0
llm_load_print_meta: ssm_d_inner      = 0
llm_load_print_meta: ssm_d_state      = 0
llm_load_print_meta: ssm_dt_rank      = 0
llm_load_print_meta: ssm_dt_b_c_rms   = 0
llm_load_print_meta: model type       = 11B
llm_load_print_meta: model ftype      = Q4_K - Medium
llm_load_print_meta: model params     = 9.78 B
llm_load_print_meta: model size       = 5.55 GiB (4.87 BPW) 
llm_load_print_meta: general.name     = Model
llm_load_print_meta: BOS token        = 128000 '<|begin_of_text|>'
llm_load_print_meta: EOS token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOT token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOM token        = 128008 '<|eom_id|>'
llm_load_print_meta: PAD token        = 128004 '<|finetune_right_pad_id|>'
llm_load_print_meta: LF token         = 128 'Ä'
llm_load_print_meta: EOG token        = 128008 '<|eom_id|>'
llm_load_print_meta: EOG token        = 128009 '<|eot_id|>'
llm_load_print_meta: max token length = 256
llama_model_load: vocab mismatch 128256 !- 128257 ...
llm_load_tensors: offloading 40 repeating layers to GPU
llm_load_tensors: offloading output layer to GPU
llm_load_tensors: offloaded 41/41 layers to GPU
llm_load_tensors:        CUDA0 model buffer size =  5397.50 MiB
llm_load_tensors:          CPU model buffer size =   281.83 MiB
llama_new_context_with_model: n_seq_max     = 1
llama_new_context_with_model: n_ctx         = 2048
llama_new_context_with_model: n_ctx_per_seq = 2048
llama_new_context_with_model: n_batch       = 512
llama_new_context_with_model: n_ubatch      = 512
llama_new_context_with_model: flash_attn    = 0
llama_new_context_with_model: freq_base     = 500000.0
llama_new_context_with_model: freq_scale    = 1
llama_new_context_with_model: n_ctx_per_seq (2048) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 2048, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 40, can_shift = 1
llama_kv_cache_init:      CUDA0 KV buffer size =   656.25 MiB
llama_new_context_with_model: KV self size  =  656.25 MiB, K (f16):  328.12 MiB, V (f16):  328.12 MiB
llama_new_context_with_model:  CUDA_Host  output buffer size =     0.50 MiB
llama_new_context_with_model:      CUDA0 compute buffer size =   258.50 MiB
llama_new_context_with_model:  CUDA_Host compute buffer size =    12.01 MiB
llama_new_context_with_model: graph nodes  = 1030
llama_new_context_with_model: graph splits = 2
mllama_model_load: model name:   Llama-3.2-11B-Vision-Instruct
mllama_model_load: description:  vision encoder for Mllama
mllama_model_load: GGUF version: 3
mllama_model_load: alignment:    32
mllama_model_load: n_tensors:    512
mllama_model_load: n_kv:         17
mllama_model_load: ftype:        f16
mllama_model_load: 
mllama_model_load: mllama_model_load: using CUDA0 backend

mllama_model_load: compute allocated memory: 2853.34 MB
time=2025-02-23T22:21:42.170-08:00 level=INFO source=server.go:596 msg="llama runner started in 12.32 seconds"
[GIN] 2025/02/23 - 22:21:42 | 200 |   12.6969004s |       127.0.0.1 | POST     "/api/generate"
[GIN] 2025/02/23 - 22:22:19 | 200 |       118.8µs |       127.0.0.1 | HEAD     "/"
[GIN] 2025/02/23 - 22:22:19 | 200 |     10.6673ms |       127.0.0.1 | GET      "/api/tags"
[GIN] 2025/02/23 - 22:22:21 | 200 |          86µs |       127.0.0.1 | HEAD     "/"
[GIN] 2025/02/23 - 22:22:21 | 200 |       629.9µs |       127.0.0.1 | GET      "/api/ps"
time=2025-02-23T22:23:38.035-08:00 level=WARN source=sched.go:137 msg="mllama doesn't support parallel requests yet"
[GIN] 2025/02/23 - 22:23:48 | 200 |          1m2s |      10.0.0.216 | POST     "/api/chat"

OS

Windows 10

GPU

Nvidia

CPU

Intel sandybridge

Ollama version

0.5.11

Originally created by @ribbles on GitHub (Feb 24, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9311 ### What is the issue? When chatting with llama3.2-vision subsequent requests are not as fast as I would expect as the model is already 100% loaded into VRAM. I don't know if it's relevant but the "load_duration" is really high on calls after the model is already loaded, when I would expect it to be zero. I also see the logs reporting 8 cores but the GPU has 3072 cores - no idea if relevant. First request: ``` {'total_duration': 57086256700, 'load_duration': 54109228500, 'prompt_eval_count': 383, 'prompt_eval_duration': 15000000, 'eval_count': 65, 'eval_duration': 2958000000} ``` Second request: ``` {'total_duration': 91264439700, 'load_duration': 87957377600, 'prompt_eval_count': 469, 'prompt_eval_duration': 234000000, 'eval_count': 64, 'eval_duration': 3063000000} ``` ### Relevant log output ```shell 2025/02/23 22:16:30 routes.go:1186: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://0.0.0.0:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:C:\\Users\\User\\.ollama\\models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[* http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:]" time=2025-02-23T22:16:30.279-08:00 level=INFO source=images.go:432 msg="total blobs: 19" time=2025-02-23T22:16:30.286-08:00 level=INFO source=images.go:439 msg="total unused blobs removed: 0" time=2025-02-23T22:16:30.293-08:00 level=INFO source=routes.go:1237 msg="Listening on [::]:11434 (version 0.5.11)" time=2025-02-23T22:16:30.294-08:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs" time=2025-02-23T22:16:30.294-08:00 level=INFO source=gpu_windows.go:167 msg=packages count=2 time=2025-02-23T22:16:30.294-08:00 level=INFO source=gpu_windows.go:214 msg="" package=0 cores=8 efficiency=0 threads=16 time=2025-02-23T22:16:30.294-08:00 level=INFO source=gpu_windows.go:214 msg="" package=1 cores=8 efficiency=0 threads=16 time=2025-02-23T22:16:30.723-08:00 level=INFO source=gpu.go:319 msg="detected OS VRAM overhead" id=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b library=cuda compute=5.2 driver=12.7 name="Tesla M40 24GB" overhead="716.4 MiB" time=2025-02-23T22:16:30.729-08:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b library=cuda variant=v11 compute=5.2 driver=12.7 name="Tesla M40 24GB" total="22.5 GiB" available="21.6 GiB" [GIN] 2025/02/23 - 22:16:32 | 200 | 538.1µs | 127.0.0.1 | HEAD "/" [GIN] 2025/02/23 - 22:21:20 | 200 | 157.8µs | 127.0.0.1 | HEAD "/" [GIN] 2025/02/23 - 22:21:20 | 200 | 75.5925ms | 127.0.0.1 | GET "/api/tags" [GIN] 2025/02/23 - 22:21:29 | 200 | 0s | 127.0.0.1 | HEAD "/" [GIN] 2025/02/23 - 22:21:29 | 200 | 125.2651ms | 127.0.0.1 | POST "/api/show" time=2025-02-23T22:21:29.557-08:00 level=WARN source=sched.go:137 msg="mllama doesn't support parallel requests yet" time=2025-02-23T22:21:29.758-08:00 level=INFO source=sched.go:714 msg="new model will fit in available VRAM in single GPU, loading" model=C:\Users\User\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 gpu=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b parallel=1 available=23197925376 required="11.3 GiB" time=2025-02-23T22:21:29.798-08:00 level=INFO source=server.go:100 msg="system memory" total="63.9 GiB" free="58.8 GiB" free_swap="68.0 GiB" time=2025-02-23T22:21:29.810-08:00 level=INFO source=memory.go:356 msg="offload to cuda" projector.weights="1.8 GiB" projector.graph="2.8 GiB" layers.requested=-1 layers.model=41 layers.offload=41 layers.split="" memory.available="[21.6 GiB]" memory.gpu_overhead="0 B" memory.required.full="11.3 GiB" memory.required.partial="11.3 GiB" memory.required.kv="656.2 MiB" memory.required.allocations="[11.3 GiB]" memory.weights.total="5.5 GiB" memory.weights.repeating="5.1 GiB" memory.weights.nonrepeating="411.0 MiB" memory.graph.full="258.5 MiB" memory.graph.partial="669.5 MiB" time=2025-02-23T22:21:29.840-08:00 level=INFO source=server.go:380 msg="starting llama server" cmd="C:\\Users\\User\\AppData\\Local\\Programs\\Ollama\\ollama.exe runner --model C:\\Users\\User\\.ollama\\models\\blobs\\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 --ctx-size 2048 --batch-size 512 --n-gpu-layers 41 --mmproj C:\\Users\\User\\.ollama\\models\\blobs\\sha256-ece5e659647a20a5c28ab9eea1c12a1ad430bc0f2a27021d00ad103b3bf5206f --threads 16 --no-mmap --parallel 1 --port 58093" time=2025-02-23T22:21:29.851-08:00 level=INFO source=sched.go:449 msg="loaded runners" count=1 time=2025-02-23T22:21:29.851-08:00 level=INFO source=server.go:557 msg="waiting for llama runner to start responding" time=2025-02-23T22:21:29.853-08:00 level=INFO source=server.go:591 msg="waiting for server to become available" status="llm server error" time=2025-02-23T22:21:29.946-08:00 level=INFO source=runner.go:936 msg="starting go runner" time=2025-02-23T22:21:29.948-08:00 level=INFO source=runner.go:937 msg=system info="CPU : LLAMAFILE = 1 | CPU : LLAMAFILE = 1 | cgo(clang)" threads=16 time=2025-02-23T22:21:29.950-08:00 level=INFO source=runner.go:995 msg="Server listening on 127.0.0.1:58093" time=2025-02-23T22:21:30.107-08:00 level=INFO source=server.go:591 msg="waiting for server to become available" status="llm server loading model" 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: Tesla M40 24GB, compute capability 5.2, VMM: yes load_backend: loaded CUDA backend from C:\Users\User\AppData\Local\Programs\Ollama\lib\ollama\cuda_v11\ggml-cuda.dll load_backend: loaded CPU backend from C:\Users\User\AppData\Local\Programs\Ollama\lib\ollama\ggml-cpu-sandybridge.dll llama_load_model_from_file: using device CUDA0 (Tesla M40 24GB) - 22081 MiB free llama_model_loader: loaded meta data with 27 key-value pairs and 396 tensors from C:\Users\User\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 (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 = mllama llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.name str = Model llama_model_loader: - kv 3: general.size_label str = 10B llama_model_loader: - kv 4: mllama.block_count u32 = 40 llama_model_loader: - kv 5: mllama.context_length u32 = 131072 llama_model_loader: - kv 6: mllama.embedding_length u32 = 4096 llama_model_loader: - kv 7: mllama.feed_forward_length u32 = 14336 llama_model_loader: - kv 8: mllama.attention.head_count u32 = 32 llama_model_loader: - kv 9: mllama.attention.head_count_kv u32 = 8 llama_model_loader: - kv 10: mllama.rope.freq_base f32 = 500000.000000 llama_model_loader: - kv 11: mllama.attention.layer_norm_rms_epsilon f32 = 0.000010 llama_model_loader: - kv 12: general.file_type u32 = 15 llama_model_loader: - kv 13: mllama.vocab_size u32 = 128256 llama_model_loader: - kv 14: mllama.rope.dimension_count u32 = 128 llama_model_loader: - kv 15: mllama.attention.cross_attention_layers arr[i32,8] = [3, 8, 13, 18, 23, 28, 33, 38] llama_model_loader: - kv 16: tokenizer.ggml.add_bos_token bool = true llama_model_loader: - kv 17: tokenizer.ggml.model str = gpt2 llama_model_loader: - kv 18: tokenizer.ggml.pre str = llama-bpe llama_model_loader: - kv 19: tokenizer.ggml.tokens arr[str,128257] = ["!", "\"", "#", "$", "%", "&", "'", ... llama_model_loader: - kv 20: tokenizer.ggml.token_type arr[i32,128257] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 21: tokenizer.ggml.merges arr[str,280147] = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "... llama_model_loader: - kv 22: tokenizer.ggml.bos_token_id u32 = 128000 llama_model_loader: - kv 23: tokenizer.ggml.eos_token_id u32 = 128009 llama_model_loader: - kv 24: tokenizer.ggml.padding_token_id u32 = 128004 llama_model_loader: - kv 25: tokenizer.chat_template str = {{- bos_token }}\n{%- if custom_tools ... llama_model_loader: - kv 26: general.quantization_version u32 = 2 llama_model_loader: - type f32: 114 tensors llama_model_loader: - type q4_K: 245 tensors llama_model_loader: - type q6_K: 37 tensors llm_load_vocab: special tokens cache size = 257 llm_load_vocab: token to piece cache size = 0.7999 MB llm_load_print_meta: format = GGUF V3 (latest) llm_load_print_meta: arch = mllama llm_load_print_meta: vocab type = BPE llm_load_print_meta: n_vocab = 128256 llm_load_print_meta: n_merges = 280147 llm_load_print_meta: vocab_only = 0 llm_load_print_meta: n_ctx_train = 131072 llm_load_print_meta: n_embd = 4096 llm_load_print_meta: n_layer = 40 llm_load_print_meta: n_head = 32 llm_load_print_meta: n_head_kv = 8 llm_load_print_meta: n_rot = 128 llm_load_print_meta: n_swa = 0 llm_load_print_meta: n_embd_head_k = 128 llm_load_print_meta: n_embd_head_v = 128 llm_load_print_meta: n_gqa = 4 llm_load_print_meta: n_embd_k_gqa = 1024 llm_load_print_meta: n_embd_v_gqa = 1024 llm_load_print_meta: f_norm_eps = 0.0e+00 llm_load_print_meta: f_norm_rms_eps = 1.0e-05 llm_load_print_meta: f_clamp_kqv = 0.0e+00 llm_load_print_meta: f_max_alibi_bias = 0.0e+00 llm_load_print_meta: f_logit_scale = 0.0e+00 llm_load_print_meta: n_ff = 14336 llm_load_print_meta: n_expert = 0 llm_load_print_meta: n_expert_used = 0 llm_load_print_meta: causal attn = 1 llm_load_print_meta: pooling type = 0 llm_load_print_meta: rope type = 0 llm_load_print_meta: rope scaling = linear llm_load_print_meta: freq_base_train = 500000.0 llm_load_print_meta: freq_scale_train = 1 llm_load_print_meta: n_ctx_orig_yarn = 131072 llm_load_print_meta: rope_finetuned = unknown llm_load_print_meta: ssm_d_conv = 0 llm_load_print_meta: ssm_d_inner = 0 llm_load_print_meta: ssm_d_state = 0 llm_load_print_meta: ssm_dt_rank = 0 llm_load_print_meta: ssm_dt_b_c_rms = 0 llm_load_print_meta: model type = 11B llm_load_print_meta: model ftype = Q4_K - Medium llm_load_print_meta: model params = 9.78 B llm_load_print_meta: model size = 5.55 GiB (4.87 BPW) llm_load_print_meta: general.name = Model llm_load_print_meta: BOS token = 128000 '<|begin_of_text|>' llm_load_print_meta: EOS token = 128009 '<|eot_id|>' llm_load_print_meta: EOT token = 128009 '<|eot_id|>' llm_load_print_meta: EOM token = 128008 '<|eom_id|>' llm_load_print_meta: PAD token = 128004 '<|finetune_right_pad_id|>' llm_load_print_meta: LF token = 128 'Ä' llm_load_print_meta: EOG token = 128008 '<|eom_id|>' llm_load_print_meta: EOG token = 128009 '<|eot_id|>' llm_load_print_meta: max token length = 256 llama_model_load: vocab mismatch 128256 !- 128257 ... llm_load_tensors: offloading 40 repeating layers to GPU llm_load_tensors: offloading output layer to GPU llm_load_tensors: offloaded 41/41 layers to GPU llm_load_tensors: CUDA0 model buffer size = 5397.50 MiB llm_load_tensors: CPU model buffer size = 281.83 MiB llama_new_context_with_model: n_seq_max = 1 llama_new_context_with_model: n_ctx = 2048 llama_new_context_with_model: n_ctx_per_seq = 2048 llama_new_context_with_model: n_batch = 512 llama_new_context_with_model: n_ubatch = 512 llama_new_context_with_model: flash_attn = 0 llama_new_context_with_model: freq_base = 500000.0 llama_new_context_with_model: freq_scale = 1 llama_new_context_with_model: n_ctx_per_seq (2048) < n_ctx_train (131072) -- the full capacity of the model will not be utilized llama_kv_cache_init: kv_size = 2048, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 40, can_shift = 1 llama_kv_cache_init: CUDA0 KV buffer size = 656.25 MiB llama_new_context_with_model: KV self size = 656.25 MiB, K (f16): 328.12 MiB, V (f16): 328.12 MiB llama_new_context_with_model: CUDA_Host output buffer size = 0.50 MiB llama_new_context_with_model: CUDA0 compute buffer size = 258.50 MiB llama_new_context_with_model: CUDA_Host compute buffer size = 12.01 MiB llama_new_context_with_model: graph nodes = 1030 llama_new_context_with_model: graph splits = 2 mllama_model_load: model name: Llama-3.2-11B-Vision-Instruct mllama_model_load: description: vision encoder for Mllama mllama_model_load: GGUF version: 3 mllama_model_load: alignment: 32 mllama_model_load: n_tensors: 512 mllama_model_load: n_kv: 17 mllama_model_load: ftype: f16 mllama_model_load: mllama_model_load: mllama_model_load: using CUDA0 backend mllama_model_load: compute allocated memory: 2853.34 MB time=2025-02-23T22:21:42.170-08:00 level=INFO source=server.go:596 msg="llama runner started in 12.32 seconds" [GIN] 2025/02/23 - 22:21:42 | 200 | 12.6969004s | 127.0.0.1 | POST "/api/generate" [GIN] 2025/02/23 - 22:22:19 | 200 | 118.8µs | 127.0.0.1 | HEAD "/" [GIN] 2025/02/23 - 22:22:19 | 200 | 10.6673ms | 127.0.0.1 | GET "/api/tags" [GIN] 2025/02/23 - 22:22:21 | 200 | 86µs | 127.0.0.1 | HEAD "/" [GIN] 2025/02/23 - 22:22:21 | 200 | 629.9µs | 127.0.0.1 | GET "/api/ps" time=2025-02-23T22:23:38.035-08:00 level=WARN source=sched.go:137 msg="mllama doesn't support parallel requests yet" [GIN] 2025/02/23 - 22:23:48 | 200 | 1m2s | 10.0.0.216 | POST "/api/chat" ``` ### OS Windows 10 ### GPU Nvidia ### CPU Intel sandybridge ### Ollama version 0.5.11
GiteaMirror added the bug label 2026-04-12 17:24:18 -05:00
GiteaMirror commented 2026-04-12 17:24:19 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 24, 2025):

I also see the logs reporting 8 cores but the GPU has 3072 cores - no idea if relevant.

CPU cores.

{'total_duration': 57086256700, 'load_duration': 54109228500, 'prompt_eval_count': 383, 'prompt_eval_duration': 15000000, 'eval_count': 65, 'eval_duration': 2958000000}
{'total_duration': 91264439700, 'load_duration': 87957377600, 'prompt_eval_count': 469, 'prompt_eval_duration': 234000000, 'eval_count': 64, 'eval_duration': 3063000000}

There's no indication of wall clock time, is it possible that the model was unloaded/reloaded between requests? See keep_alive.

<!-- gh-comment-id:2678028285 --> @rick-github commented on GitHub (Feb 24, 2025): > I also see the logs reporting 8 cores but the GPU has 3072 cores - no idea if relevant. CPU cores. ``` {'total_duration': 57086256700, 'load_duration': 54109228500, 'prompt_eval_count': 383, 'prompt_eval_duration': 15000000, 'eval_count': 65, 'eval_duration': 2958000000} {'total_duration': 91264439700, 'load_duration': 87957377600, 'prompt_eval_count': 469, 'prompt_eval_duration': 234000000, 'eval_count': 64, 'eval_duration': 3063000000} ``` There's no indication of wall clock time, is it possible that the model was unloaded/reloaded between requests? See [`keep_alive`](https://github.com/ollama/ollama/blob/main/docs/faq.md#how-do-i-keep-a-model-loaded-in-memory-or-make-it-unload-immediately).
GiteaMirror commented 2026-04-12 17:24:20 -05:00
Author
Owner
Copy Link

@ribbles commented on GitHub (Feb 24, 2025):

keep_alive is set to 20m (1200) and the requests are serial and immediate - when the first request completed the next request was submitted.

<!-- gh-comment-id:2678920229 --> @ribbles commented on GitHub (Feb 24, 2025): `keep_alive` is set to 20m (1200) and the requests are serial and immediate - when the first request completed the next request was submitted.
GiteaMirror commented 2026-04-12 17:24:20 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 24, 2025):

Can you post logs for two requests?

<!-- gh-comment-id:2678926153 --> @rick-github commented on GitHub (Feb 24, 2025): Can you post logs for two requests?
GiteaMirror commented 2026-04-12 17:24:21 -05:00
Author
Owner
Copy Link

@ribbles commented on GitHub (Feb 24, 2025):

Requests:

2025-02-24 08:34:34.988 Chat request: {"model": "llama3.2-vision", "messages": [{"role": "user", "content": "truncated"}], "stream": false, "options": {"temperature": 1.0}, "keep_alive": 1200, "format": {"type": "object", "properties": {"message": {"type": "string"}, "tool_calls": {"type": "array", "items": {"type": "object", "properties": {"servo_id": {"type": "integer", "minimum": 1, "maximum": 6}, "position": {"type": "integer", "minimum": 500, "maximum": 2500}}, "required": ["servo_id", "position"]}}}}, "images": ["truncated"]}
2025-02-24 08:35:59.800 Chat response: {"model": "llama3.2-vision", "created_at": "2025-02-24T16:35:59.7807453Z", "message": {"role": "assistant", "content": "truncated"}, "done_reason": "stop", "done": true, "total_duration": 84673957800, "load_duration": 79065586400, "prompt_eval_count": 383, "prompt_eval_duration": 1521000000, "eval_count": 71, "eval_duration": 4069000000}       

2025-02-24 08:36:10.736 Chat request: {"model": "llama3.2-vision", "messages": [{"role": "user", "content": "truncated"}, {"role": "assistant", "content": "truncated"}, {"role": "user", "content": "truncated"}], "stream": false, "options": {"temperature": 1.0}, "keep_alive": 1200, "format": {"type": "object", "properties": {"message": {"type": "string"}, "tool_calls": {"type": "array", "items": {"type": "object", "properties": {"servo_id": {"type": "integer", "minimum": 1, "maximum": 6}, "position": {"type": "integer", "minimum": 500, "maximum": 2500}}, "required": ["servo_id", "position"]}}}}, "images": ["truncated"]}
2025-02-24 08:37:31.445 Chat response: {"model": "llama3.2-vision", "created_at": "2025-02-24T16:37:31.4638142Z", "message": {"role": "assistant", "content": "truncated"}, "done_reason": "stop", "done": true, "total_duration": 80591115900, "load_duration": 75593078600, "prompt_eval_count": 475, "prompt_eval_duration": 259000000, "eval_count": 67, "eval_duration": 3281000000}

Logs for both requests (after ollama restart)

2025/02/24 08:33:27 routes.go:1186: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://0.0.0.0:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:C:\\Users\\luser\\.ollama\\models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[* http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:]"
time=2025-02-24T08:33:27.983-08:00 level=INFO source=images.go:432 msg="total blobs: 19"
time=2025-02-24T08:33:27.990-08:00 level=INFO source=images.go:439 msg="total unused blobs removed: 0"
time=2025-02-24T08:33:27.997-08:00 level=INFO source=routes.go:1237 msg="Listening on [::]:11434 (version 0.5.11)"
time=2025-02-24T08:33:27.997-08:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs"
time=2025-02-24T08:33:27.997-08:00 level=INFO source=gpu_windows.go:167 msg=packages count=2
time=2025-02-24T08:33:27.997-08:00 level=INFO source=gpu_windows.go:214 msg="" package=0 cores=8 efficiency=0 threads=16
time=2025-02-24T08:33:27.998-08:00 level=INFO source=gpu_windows.go:214 msg="" package=1 cores=8 efficiency=0 threads=16
time=2025-02-24T08:33:28.370-08:00 level=INFO source=gpu.go:319 msg="detected OS VRAM overhead" id=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b library=cuda compute=5.2 driver=12.7 name="Tesla M40 24GB" overhead="671.2 MiB"
time=2025-02-24T08:33:28.377-08:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b library=cuda variant=v11 compute=5.2 driver=12.7 name="Tesla M40 24GB" total="22.5 GiB" available="21.6 GiB"
time=2025-02-24T08:35:41.082-08:00 level=WARN source=sched.go:137 msg="mllama doesn't support parallel requests yet"
time=2025-02-24T08:35:41.290-08:00 level=INFO source=sched.go:714 msg="new model will fit in available VRAM in single GPU, loading" model=C:\Users\luser\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 gpu=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b parallel=1 available=23195418624 required="11.3 GiB"
time=2025-02-24T08:35:41.307-08:00 level=INFO source=server.go:100 msg="system memory" total="63.9 GiB" free="58.7 GiB" free_swap="67.9 GiB"
time=2025-02-24T08:35:41.322-08:00 level=INFO source=memory.go:356 msg="offload to cuda" projector.weights="1.8 GiB" projector.graph="2.8 GiB" layers.requested=-1 layers.model=41 layers.offload=41 layers.split="" memory.available="[21.6 GiB]" memory.gpu_overhead="0 B" memory.required.full="11.3 GiB" memory.required.partial="11.3 GiB" memory.required.kv="656.2 MiB" memory.required.allocations="[11.3 GiB]" memory.weights.total="5.5 GiB" memory.weights.repeating="5.1 GiB" memory.weights.nonrepeating="411.0 MiB" memory.graph.full="258.5 MiB" memory.graph.partial="669.5 MiB"
time=2025-02-24T08:35:41.354-08:00 level=INFO source=server.go:380 msg="starting llama server" cmd="C:\\Users\\luser\\AppData\\Local\\Programs\\Ollama\\ollama.exe runner --model C:\\Users\\luser\\.ollama\\models\\blobs\\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 --ctx-size 2048 --batch-size 512 --n-gpu-layers 41 --mmproj C:\\Users\\luser\\.ollama\\models\\blobs\\sha256-ece5e659647a20a5c28ab9eea1c12a1ad430bc0f2a27021d00ad103b3bf5206f --threads 16 --no-mmap --parallel 1 --port 59625"
time=2025-02-24T08:35:41.366-08:00 level=INFO source=sched.go:449 msg="loaded runners" count=1
time=2025-02-24T08:35:41.366-08:00 level=INFO source=server.go:557 msg="waiting for llama runner to start responding"
time=2025-02-24T08:35:41.368-08:00 level=INFO source=server.go:591 msg="waiting for server to become available" status="llm server error"
time=2025-02-24T08:35:41.453-08:00 level=INFO source=runner.go:936 msg="starting go runner"
time=2025-02-24T08:35:41.453-08:00 level=INFO source=runner.go:937 msg=system info="CPU : LLAMAFILE = 1 | CPU : LLAMAFILE = 1 | cgo(clang)" threads=16
time=2025-02-24T08:35:41.464-08:00 level=INFO source=runner.go:995 msg="Server listening on 127.0.0.1:59625"
time=2025-02-24T08:35:41.621-08:00 level=INFO source=server.go:591 msg="waiting for server to become available" status="llm server loading model"
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: Tesla M40 24GB, compute capability 5.2, VMM: yes
load_backend: loaded CUDA backend from C:\Users\luser\AppData\Local\Programs\Ollama\lib\ollama\cuda_v11\ggml-cuda.dll
load_backend: loaded CPU backend from C:\Users\luser\AppData\Local\Programs\Ollama\lib\ollama\ggml-cpu-sandybridge.dll
llama_load_model_from_file: using device CUDA0 (Tesla M40 24GB) - 22081 MiB free
llama_model_loader: loaded meta data with 27 key-value pairs and 396 tensors from C:\Users\luser\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 (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              = mllama
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Model
llama_model_loader: - kv   3:                         general.size_label str              = 10B
llama_model_loader: - kv   4:                         mllama.block_count u32              = 40
llama_model_loader: - kv   5:                      mllama.context_length u32              = 131072
llama_model_loader: - kv   6:                    mllama.embedding_length u32              = 4096
llama_model_loader: - kv   7:                 mllama.feed_forward_length u32              = 14336
llama_model_loader: - kv   8:                mllama.attention.head_count u32              = 32
llama_model_loader: - kv   9:             mllama.attention.head_count_kv u32              = 8
llama_model_loader: - kv  10:                      mllama.rope.freq_base f32              = 500000.000000
llama_model_loader: - kv  11:    mllama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  12:                          general.file_type u32              = 15
llama_model_loader: - kv  13:                          mllama.vocab_size u32              = 128256
llama_model_loader: - kv  14:                mllama.rope.dimension_count u32              = 128
llama_model_loader: - kv  15:    mllama.attention.cross_attention_layers arr[i32,8]       = [3, 8, 13, 18, 23, 28, 33, 38]
llama_model_loader: - kv  16:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  17:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  18:                         tokenizer.ggml.pre str              = llama-bpe
llama_model_loader: - kv  19:                      tokenizer.ggml.tokens arr[str,128257]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  20:                  tokenizer.ggml.token_type arr[i32,128257]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  21:                      tokenizer.ggml.merges arr[str,280147]  = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "...
llama_model_loader: - kv  22:                tokenizer.ggml.bos_token_id u32              = 128000
llama_model_loader: - kv  23:                tokenizer.ggml.eos_token_id u32              = 128009
llama_model_loader: - kv  24:            tokenizer.ggml.padding_token_id u32              = 128004
llama_model_loader: - kv  25:                    tokenizer.chat_template str              = {{- bos_token }}\n{%- if custom_tools ...
llama_model_loader: - kv  26:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  114 tensors
llama_model_loader: - type q4_K:  245 tensors
llama_model_loader: - type q6_K:   37 tensors
llm_load_vocab: special tokens cache size = 257
llm_load_vocab: token to piece cache size = 0.7999 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = mllama
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 128256
llm_load_print_meta: n_merges         = 280147
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 131072
llm_load_print_meta: n_embd           = 4096
llm_load_print_meta: n_layer          = 40
llm_load_print_meta: n_head           = 32
llm_load_print_meta: n_head_kv        = 8
llm_load_print_meta: n_rot            = 128
llm_load_print_meta: n_swa            = 0
llm_load_print_meta: n_embd_head_k    = 128
llm_load_print_meta: n_embd_head_v    = 128
llm_load_print_meta: n_gqa            = 4
llm_load_print_meta: n_embd_k_gqa     = 1024
llm_load_print_meta: n_embd_v_gqa     = 1024
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
llm_load_print_meta: f_clamp_kqv      = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale    = 0.0e+00
llm_load_print_meta: n_ff             = 14336
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0
llm_load_print_meta: causal attn      = 1
llm_load_print_meta: pooling type     = 0
llm_load_print_meta: rope type        = 0
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 500000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 131072
llm_load_print_meta: rope_finetuned   = unknown
llm_load_print_meta: ssm_d_conv       = 0
llm_load_print_meta: ssm_d_inner      = 0
llm_load_print_meta: ssm_d_state      = 0
llm_load_print_meta: ssm_dt_rank      = 0
llm_load_print_meta: ssm_dt_b_c_rms   = 0
llm_load_print_meta: model type       = 11B
llm_load_print_meta: model ftype      = Q4_K - Medium
llm_load_print_meta: model params     = 9.78 B
llm_load_print_meta: model size       = 5.55 GiB (4.87 BPW) 
llm_load_print_meta: general.name     = Model
llm_load_print_meta: BOS token        = 128000 '<|begin_of_text|>'
llm_load_print_meta: EOS token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOT token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOM token        = 128008 '<|eom_id|>'
llm_load_print_meta: PAD token        = 128004 '<|finetune_right_pad_id|>'
llm_load_print_meta: LF token         = 128 'Ä'
llm_load_print_meta: EOG token        = 128008 '<|eom_id|>'
llm_load_print_meta: EOG token        = 128009 '<|eot_id|>'
llm_load_print_meta: max token length = 256
llama_model_load: vocab mismatch 128256 !- 128257 ...
llm_load_tensors: offloading 40 repeating layers to GPU
llm_load_tensors: offloading output layer to GPU
llm_load_tensors: offloaded 41/41 layers to GPU
llm_load_tensors:        CUDA0 model buffer size =  5397.50 MiB
llm_load_tensors:          CPU model buffer size =   281.83 MiB
llama_new_context_with_model: n_seq_max     = 1
llama_new_context_with_model: n_ctx         = 2048
llama_new_context_with_model: n_ctx_per_seq = 2048
llama_new_context_with_model: n_batch       = 512
llama_new_context_with_model: n_ubatch      = 512
llama_new_context_with_model: flash_attn    = 0
llama_new_context_with_model: freq_base     = 500000.0
llama_new_context_with_model: freq_scale    = 1
llama_new_context_with_model: n_ctx_per_seq (2048) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 2048, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 40, can_shift = 1
llama_kv_cache_init:      CUDA0 KV buffer size =   656.25 MiB
llama_new_context_with_model: KV self size  =  656.25 MiB, K (f16):  328.12 MiB, V (f16):  328.12 MiB
llama_new_context_with_model:  CUDA_Host  output buffer size =     0.50 MiB
llama_new_context_with_model:      CUDA0 compute buffer size =   258.50 MiB
llama_new_context_with_model:  CUDA_Host compute buffer size =    12.01 MiB
llama_new_context_with_model: graph nodes  = 1030
llama_new_context_with_model: graph splits = 2
mllama_model_load: model name:   Llama-3.2-11B-Vision-Instruct
mllama_model_load: description:  vision encoder for Mllama
mllama_model_load: GGUF version: 3
mllama_model_load: alignment:    32
mllama_model_load: n_tensors:    512
mllama_model_load: n_kv:         17
mllama_model_load: ftype:        f16
mllama_model_load: 
mllama_model_load: mllama_model_load: using CUDA0 backend

mllama_model_load: compute allocated memory: 2853.34 MB
time=2025-02-24T08:35:54.171-08:00 level=INFO source=server.go:596 msg="llama runner started in 12.81 seconds"
[GIN] 2025/02/24 - 08:35:59 | 200 |         1m24s |      10.0.0.216 | POST     "/api/chat"
time=2025-02-24T08:37:26.463-08:00 level=WARN source=sched.go:137 msg="mllama doesn't support parallel requests yet"
llama_model_loader: loaded meta data with 27 key-value pairs and 396 tensors from C:\Users\luser\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 (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              = mllama
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Model
llama_model_loader: - kv   3:                         general.size_label str              = 10B
llama_model_loader: - kv   4:                         mllama.block_count u32              = 40
llama_model_loader: - kv   5:                      mllama.context_length u32              = 131072
llama_model_loader: - kv   6:                    mllama.embedding_length u32              = 4096
llama_model_loader: - kv   7:                 mllama.feed_forward_length u32              = 14336
llama_model_loader: - kv   8:                mllama.attention.head_count u32              = 32
llama_model_loader: - kv   9:             mllama.attention.head_count_kv u32              = 8
llama_model_loader: - kv  10:                      mllama.rope.freq_base f32              = 500000.000000
llama_model_loader: - kv  11:    mllama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  12:                          general.file_type u32              = 15
llama_model_loader: - kv  13:                          mllama.vocab_size u32              = 128256
llama_model_loader: - kv  14:                mllama.rope.dimension_count u32              = 128
llama_model_loader: - kv  15:    mllama.attention.cross_attention_layers arr[i32,8]       = [3, 8, 13, 18, 23, 28, 33, 38]
llama_model_loader: - kv  16:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  17:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  18:                         tokenizer.ggml.pre str              = llama-bpe
llama_model_loader: - kv  19:                      tokenizer.ggml.tokens arr[str,128257]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  20:                  tokenizer.ggml.token_type arr[i32,128257]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  21:                      tokenizer.ggml.merges arr[str,280147]  = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "...
llama_model_loader: - kv  22:                tokenizer.ggml.bos_token_id u32              = 128000
llama_model_loader: - kv  23:                tokenizer.ggml.eos_token_id u32              = 128009
llama_model_loader: - kv  24:            tokenizer.ggml.padding_token_id u32              = 128004
llama_model_loader: - kv  25:                    tokenizer.chat_template str              = {{- bos_token }}\n{%- if custom_tools ...
llama_model_loader: - kv  26:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  114 tensors
llama_model_loader: - type q4_K:  245 tensors
llama_model_loader: - type q6_K:   37 tensors
llm_load_vocab: special tokens cache size = 257
llm_load_vocab: token to piece cache size = 0.7999 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = mllama
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 128256
llm_load_print_meta: n_merges         = 280147
llm_load_print_meta: vocab_only       = 1
llm_load_print_meta: model type       = ?B
llm_load_print_meta: model ftype      = all F32
llm_load_print_meta: model params     = 9.78 B
llm_load_print_meta: model size       = 5.55 GiB (4.87 BPW) 
llm_load_print_meta: general.name     = Model
llm_load_print_meta: BOS token        = 128000 '<|begin_of_text|>'
llm_load_print_meta: EOS token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOT token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOM token        = 128008 '<|eom_id|>'
llm_load_print_meta: PAD token        = 128004 '<|finetune_right_pad_id|>'
llm_load_print_meta: LF token         = 128 'Ä'
llm_load_print_meta: EOG token        = 128008 '<|eom_id|>'
llm_load_print_meta: EOG token        = 128009 '<|eot_id|>'
llm_load_print_meta: max token length = 256
llama_model_load: vocab mismatch 128256 !- 128257 ...
llama_model_load: vocab only - skipping tensors
[GIN] 2025/02/24 - 08:37:31 | 200 |         1m20s |      10.0.0.216 | POST     "/api/chat"
<!-- gh-comment-id:2679042011 --> @ribbles commented on GitHub (Feb 24, 2025): Requests: ``` 2025-02-24 08:34:34.988 Chat request: {"model": "llama3.2-vision", "messages": [{"role": "user", "content": "truncated"}], "stream": false, "options": {"temperature": 1.0}, "keep_alive": 1200, "format": {"type": "object", "properties": {"message": {"type": "string"}, "tool_calls": {"type": "array", "items": {"type": "object", "properties": {"servo_id": {"type": "integer", "minimum": 1, "maximum": 6}, "position": {"type": "integer", "minimum": 500, "maximum": 2500}}, "required": ["servo_id", "position"]}}}}, "images": ["truncated"]} 2025-02-24 08:35:59.800 Chat response: {"model": "llama3.2-vision", "created_at": "2025-02-24T16:35:59.7807453Z", "message": {"role": "assistant", "content": "truncated"}, "done_reason": "stop", "done": true, "total_duration": 84673957800, "load_duration": 79065586400, "prompt_eval_count": 383, "prompt_eval_duration": 1521000000, "eval_count": 71, "eval_duration": 4069000000} 2025-02-24 08:36:10.736 Chat request: {"model": "llama3.2-vision", "messages": [{"role": "user", "content": "truncated"}, {"role": "assistant", "content": "truncated"}, {"role": "user", "content": "truncated"}], "stream": false, "options": {"temperature": 1.0}, "keep_alive": 1200, "format": {"type": "object", "properties": {"message": {"type": "string"}, "tool_calls": {"type": "array", "items": {"type": "object", "properties": {"servo_id": {"type": "integer", "minimum": 1, "maximum": 6}, "position": {"type": "integer", "minimum": 500, "maximum": 2500}}, "required": ["servo_id", "position"]}}}}, "images": ["truncated"]} 2025-02-24 08:37:31.445 Chat response: {"model": "llama3.2-vision", "created_at": "2025-02-24T16:37:31.4638142Z", "message": {"role": "assistant", "content": "truncated"}, "done_reason": "stop", "done": true, "total_duration": 80591115900, "load_duration": 75593078600, "prompt_eval_count": 475, "prompt_eval_duration": 259000000, "eval_count": 67, "eval_duration": 3281000000} ``` Logs for both requests (after ollama restart) ``` 2025/02/24 08:33:27 routes.go:1186: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://0.0.0.0:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:C:\\Users\\luser\\.ollama\\models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[* http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:]" time=2025-02-24T08:33:27.983-08:00 level=INFO source=images.go:432 msg="total blobs: 19" time=2025-02-24T08:33:27.990-08:00 level=INFO source=images.go:439 msg="total unused blobs removed: 0" time=2025-02-24T08:33:27.997-08:00 level=INFO source=routes.go:1237 msg="Listening on [::]:11434 (version 0.5.11)" time=2025-02-24T08:33:27.997-08:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs" time=2025-02-24T08:33:27.997-08:00 level=INFO source=gpu_windows.go:167 msg=packages count=2 time=2025-02-24T08:33:27.997-08:00 level=INFO source=gpu_windows.go:214 msg="" package=0 cores=8 efficiency=0 threads=16 time=2025-02-24T08:33:27.998-08:00 level=INFO source=gpu_windows.go:214 msg="" package=1 cores=8 efficiency=0 threads=16 time=2025-02-24T08:33:28.370-08:00 level=INFO source=gpu.go:319 msg="detected OS VRAM overhead" id=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b library=cuda compute=5.2 driver=12.7 name="Tesla M40 24GB" overhead="671.2 MiB" time=2025-02-24T08:33:28.377-08:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b library=cuda variant=v11 compute=5.2 driver=12.7 name="Tesla M40 24GB" total="22.5 GiB" available="21.6 GiB" time=2025-02-24T08:35:41.082-08:00 level=WARN source=sched.go:137 msg="mllama doesn't support parallel requests yet" time=2025-02-24T08:35:41.290-08:00 level=INFO source=sched.go:714 msg="new model will fit in available VRAM in single GPU, loading" model=C:\Users\luser\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 gpu=GPU-538d5233-15ee-c83a-d8e6-e7881e199a8b parallel=1 available=23195418624 required="11.3 GiB" time=2025-02-24T08:35:41.307-08:00 level=INFO source=server.go:100 msg="system memory" total="63.9 GiB" free="58.7 GiB" free_swap="67.9 GiB" time=2025-02-24T08:35:41.322-08:00 level=INFO source=memory.go:356 msg="offload to cuda" projector.weights="1.8 GiB" projector.graph="2.8 GiB" layers.requested=-1 layers.model=41 layers.offload=41 layers.split="" memory.available="[21.6 GiB]" memory.gpu_overhead="0 B" memory.required.full="11.3 GiB" memory.required.partial="11.3 GiB" memory.required.kv="656.2 MiB" memory.required.allocations="[11.3 GiB]" memory.weights.total="5.5 GiB" memory.weights.repeating="5.1 GiB" memory.weights.nonrepeating="411.0 MiB" memory.graph.full="258.5 MiB" memory.graph.partial="669.5 MiB" time=2025-02-24T08:35:41.354-08:00 level=INFO source=server.go:380 msg="starting llama server" cmd="C:\\Users\\luser\\AppData\\Local\\Programs\\Ollama\\ollama.exe runner --model C:\\Users\\luser\\.ollama\\models\\blobs\\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 --ctx-size 2048 --batch-size 512 --n-gpu-layers 41 --mmproj C:\\Users\\luser\\.ollama\\models\\blobs\\sha256-ece5e659647a20a5c28ab9eea1c12a1ad430bc0f2a27021d00ad103b3bf5206f --threads 16 --no-mmap --parallel 1 --port 59625" time=2025-02-24T08:35:41.366-08:00 level=INFO source=sched.go:449 msg="loaded runners" count=1 time=2025-02-24T08:35:41.366-08:00 level=INFO source=server.go:557 msg="waiting for llama runner to start responding" time=2025-02-24T08:35:41.368-08:00 level=INFO source=server.go:591 msg="waiting for server to become available" status="llm server error" time=2025-02-24T08:35:41.453-08:00 level=INFO source=runner.go:936 msg="starting go runner" time=2025-02-24T08:35:41.453-08:00 level=INFO source=runner.go:937 msg=system info="CPU : LLAMAFILE = 1 | CPU : LLAMAFILE = 1 | cgo(clang)" threads=16 time=2025-02-24T08:35:41.464-08:00 level=INFO source=runner.go:995 msg="Server listening on 127.0.0.1:59625" time=2025-02-24T08:35:41.621-08:00 level=INFO source=server.go:591 msg="waiting for server to become available" status="llm server loading model" 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: Tesla M40 24GB, compute capability 5.2, VMM: yes load_backend: loaded CUDA backend from C:\Users\luser\AppData\Local\Programs\Ollama\lib\ollama\cuda_v11\ggml-cuda.dll load_backend: loaded CPU backend from C:\Users\luser\AppData\Local\Programs\Ollama\lib\ollama\ggml-cpu-sandybridge.dll llama_load_model_from_file: using device CUDA0 (Tesla M40 24GB) - 22081 MiB free llama_model_loader: loaded meta data with 27 key-value pairs and 396 tensors from C:\Users\luser\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 (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 = mllama llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.name str = Model llama_model_loader: - kv 3: general.size_label str = 10B llama_model_loader: - kv 4: mllama.block_count u32 = 40 llama_model_loader: - kv 5: mllama.context_length u32 = 131072 llama_model_loader: - kv 6: mllama.embedding_length u32 = 4096 llama_model_loader: - kv 7: mllama.feed_forward_length u32 = 14336 llama_model_loader: - kv 8: mllama.attention.head_count u32 = 32 llama_model_loader: - kv 9: mllama.attention.head_count_kv u32 = 8 llama_model_loader: - kv 10: mllama.rope.freq_base f32 = 500000.000000 llama_model_loader: - kv 11: mllama.attention.layer_norm_rms_epsilon f32 = 0.000010 llama_model_loader: - kv 12: general.file_type u32 = 15 llama_model_loader: - kv 13: mllama.vocab_size u32 = 128256 llama_model_loader: - kv 14: mllama.rope.dimension_count u32 = 128 llama_model_loader: - kv 15: mllama.attention.cross_attention_layers arr[i32,8] = [3, 8, 13, 18, 23, 28, 33, 38] llama_model_loader: - kv 16: tokenizer.ggml.add_bos_token bool = true llama_model_loader: - kv 17: tokenizer.ggml.model str = gpt2 llama_model_loader: - kv 18: tokenizer.ggml.pre str = llama-bpe llama_model_loader: - kv 19: tokenizer.ggml.tokens arr[str,128257] = ["!", "\"", "#", "$", "%", "&", "'", ... llama_model_loader: - kv 20: tokenizer.ggml.token_type arr[i32,128257] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 21: tokenizer.ggml.merges arr[str,280147] = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "... llama_model_loader: - kv 22: tokenizer.ggml.bos_token_id u32 = 128000 llama_model_loader: - kv 23: tokenizer.ggml.eos_token_id u32 = 128009 llama_model_loader: - kv 24: tokenizer.ggml.padding_token_id u32 = 128004 llama_model_loader: - kv 25: tokenizer.chat_template str = {{- bos_token }}\n{%- if custom_tools ... llama_model_loader: - kv 26: general.quantization_version u32 = 2 llama_model_loader: - type f32: 114 tensors llama_model_loader: - type q4_K: 245 tensors llama_model_loader: - type q6_K: 37 tensors llm_load_vocab: special tokens cache size = 257 llm_load_vocab: token to piece cache size = 0.7999 MB llm_load_print_meta: format = GGUF V3 (latest) llm_load_print_meta: arch = mllama llm_load_print_meta: vocab type = BPE llm_load_print_meta: n_vocab = 128256 llm_load_print_meta: n_merges = 280147 llm_load_print_meta: vocab_only = 0 llm_load_print_meta: n_ctx_train = 131072 llm_load_print_meta: n_embd = 4096 llm_load_print_meta: n_layer = 40 llm_load_print_meta: n_head = 32 llm_load_print_meta: n_head_kv = 8 llm_load_print_meta: n_rot = 128 llm_load_print_meta: n_swa = 0 llm_load_print_meta: n_embd_head_k = 128 llm_load_print_meta: n_embd_head_v = 128 llm_load_print_meta: n_gqa = 4 llm_load_print_meta: n_embd_k_gqa = 1024 llm_load_print_meta: n_embd_v_gqa = 1024 llm_load_print_meta: f_norm_eps = 0.0e+00 llm_load_print_meta: f_norm_rms_eps = 1.0e-05 llm_load_print_meta: f_clamp_kqv = 0.0e+00 llm_load_print_meta: f_max_alibi_bias = 0.0e+00 llm_load_print_meta: f_logit_scale = 0.0e+00 llm_load_print_meta: n_ff = 14336 llm_load_print_meta: n_expert = 0 llm_load_print_meta: n_expert_used = 0 llm_load_print_meta: causal attn = 1 llm_load_print_meta: pooling type = 0 llm_load_print_meta: rope type = 0 llm_load_print_meta: rope scaling = linear llm_load_print_meta: freq_base_train = 500000.0 llm_load_print_meta: freq_scale_train = 1 llm_load_print_meta: n_ctx_orig_yarn = 131072 llm_load_print_meta: rope_finetuned = unknown llm_load_print_meta: ssm_d_conv = 0 llm_load_print_meta: ssm_d_inner = 0 llm_load_print_meta: ssm_d_state = 0 llm_load_print_meta: ssm_dt_rank = 0 llm_load_print_meta: ssm_dt_b_c_rms = 0 llm_load_print_meta: model type = 11B llm_load_print_meta: model ftype = Q4_K - Medium llm_load_print_meta: model params = 9.78 B llm_load_print_meta: model size = 5.55 GiB (4.87 BPW) llm_load_print_meta: general.name = Model llm_load_print_meta: BOS token = 128000 '<|begin_of_text|>' llm_load_print_meta: EOS token = 128009 '<|eot_id|>' llm_load_print_meta: EOT token = 128009 '<|eot_id|>' llm_load_print_meta: EOM token = 128008 '<|eom_id|>' llm_load_print_meta: PAD token = 128004 '<|finetune_right_pad_id|>' llm_load_print_meta: LF token = 128 'Ä' llm_load_print_meta: EOG token = 128008 '<|eom_id|>' llm_load_print_meta: EOG token = 128009 '<|eot_id|>' llm_load_print_meta: max token length = 256 llama_model_load: vocab mismatch 128256 !- 128257 ... llm_load_tensors: offloading 40 repeating layers to GPU llm_load_tensors: offloading output layer to GPU llm_load_tensors: offloaded 41/41 layers to GPU llm_load_tensors: CUDA0 model buffer size = 5397.50 MiB llm_load_tensors: CPU model buffer size = 281.83 MiB llama_new_context_with_model: n_seq_max = 1 llama_new_context_with_model: n_ctx = 2048 llama_new_context_with_model: n_ctx_per_seq = 2048 llama_new_context_with_model: n_batch = 512 llama_new_context_with_model: n_ubatch = 512 llama_new_context_with_model: flash_attn = 0 llama_new_context_with_model: freq_base = 500000.0 llama_new_context_with_model: freq_scale = 1 llama_new_context_with_model: n_ctx_per_seq (2048) < n_ctx_train (131072) -- the full capacity of the model will not be utilized llama_kv_cache_init: kv_size = 2048, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 40, can_shift = 1 llama_kv_cache_init: CUDA0 KV buffer size = 656.25 MiB llama_new_context_with_model: KV self size = 656.25 MiB, K (f16): 328.12 MiB, V (f16): 328.12 MiB llama_new_context_with_model: CUDA_Host output buffer size = 0.50 MiB llama_new_context_with_model: CUDA0 compute buffer size = 258.50 MiB llama_new_context_with_model: CUDA_Host compute buffer size = 12.01 MiB llama_new_context_with_model: graph nodes = 1030 llama_new_context_with_model: graph splits = 2 mllama_model_load: model name: Llama-3.2-11B-Vision-Instruct mllama_model_load: description: vision encoder for Mllama mllama_model_load: GGUF version: 3 mllama_model_load: alignment: 32 mllama_model_load: n_tensors: 512 mllama_model_load: n_kv: 17 mllama_model_load: ftype: f16 mllama_model_load: mllama_model_load: mllama_model_load: using CUDA0 backend mllama_model_load: compute allocated memory: 2853.34 MB time=2025-02-24T08:35:54.171-08:00 level=INFO source=server.go:596 msg="llama runner started in 12.81 seconds" [GIN] 2025/02/24 - 08:35:59 | 200 | 1m24s | 10.0.0.216 | POST "/api/chat" time=2025-02-24T08:37:26.463-08:00 level=WARN source=sched.go:137 msg="mllama doesn't support parallel requests yet" llama_model_loader: loaded meta data with 27 key-value pairs and 396 tensors from C:\Users\luser\.ollama\models\blobs\sha256-11f274007f093fefeec994a5dbbb33d0733a4feb87f7ab66dcd7c1069fef0068 (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 = mllama llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.name str = Model llama_model_loader: - kv 3: general.size_label str = 10B llama_model_loader: - kv 4: mllama.block_count u32 = 40 llama_model_loader: - kv 5: mllama.context_length u32 = 131072 llama_model_loader: - kv 6: mllama.embedding_length u32 = 4096 llama_model_loader: - kv 7: mllama.feed_forward_length u32 = 14336 llama_model_loader: - kv 8: mllama.attention.head_count u32 = 32 llama_model_loader: - kv 9: mllama.attention.head_count_kv u32 = 8 llama_model_loader: - kv 10: mllama.rope.freq_base f32 = 500000.000000 llama_model_loader: - kv 11: mllama.attention.layer_norm_rms_epsilon f32 = 0.000010 llama_model_loader: - kv 12: general.file_type u32 = 15 llama_model_loader: - kv 13: mllama.vocab_size u32 = 128256 llama_model_loader: - kv 14: mllama.rope.dimension_count u32 = 128 llama_model_loader: - kv 15: mllama.attention.cross_attention_layers arr[i32,8] = [3, 8, 13, 18, 23, 28, 33, 38] llama_model_loader: - kv 16: tokenizer.ggml.add_bos_token bool = true llama_model_loader: - kv 17: tokenizer.ggml.model str = gpt2 llama_model_loader: - kv 18: tokenizer.ggml.pre str = llama-bpe llama_model_loader: - kv 19: tokenizer.ggml.tokens arr[str,128257] = ["!", "\"", "#", "$", "%", "&", "'", ... llama_model_loader: - kv 20: tokenizer.ggml.token_type arr[i32,128257] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 21: tokenizer.ggml.merges arr[str,280147] = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "... llama_model_loader: - kv 22: tokenizer.ggml.bos_token_id u32 = 128000 llama_model_loader: - kv 23: tokenizer.ggml.eos_token_id u32 = 128009 llama_model_loader: - kv 24: tokenizer.ggml.padding_token_id u32 = 128004 llama_model_loader: - kv 25: tokenizer.chat_template str = {{- bos_token }}\n{%- if custom_tools ... llama_model_loader: - kv 26: general.quantization_version u32 = 2 llama_model_loader: - type f32: 114 tensors llama_model_loader: - type q4_K: 245 tensors llama_model_loader: - type q6_K: 37 tensors llm_load_vocab: special tokens cache size = 257 llm_load_vocab: token to piece cache size = 0.7999 MB llm_load_print_meta: format = GGUF V3 (latest) llm_load_print_meta: arch = mllama llm_load_print_meta: vocab type = BPE llm_load_print_meta: n_vocab = 128256 llm_load_print_meta: n_merges = 280147 llm_load_print_meta: vocab_only = 1 llm_load_print_meta: model type = ?B llm_load_print_meta: model ftype = all F32 llm_load_print_meta: model params = 9.78 B llm_load_print_meta: model size = 5.55 GiB (4.87 BPW) llm_load_print_meta: general.name = Model llm_load_print_meta: BOS token = 128000 '<|begin_of_text|>' llm_load_print_meta: EOS token = 128009 '<|eot_id|>' llm_load_print_meta: EOT token = 128009 '<|eot_id|>' llm_load_print_meta: EOM token = 128008 '<|eom_id|>' llm_load_print_meta: PAD token = 128004 '<|finetune_right_pad_id|>' llm_load_print_meta: LF token = 128 'Ä' llm_load_print_meta: EOG token = 128008 '<|eom_id|>' llm_load_print_meta: EOG token = 128009 '<|eot_id|>' llm_load_print_meta: max token length = 256 llama_model_load: vocab mismatch 128256 !- 128257 ... llama_model_load: vocab only - skipping tensors [GIN] 2025/02/24 - 08:37:31 | 200 | 1m20s | 10.0.0.216 | POST "/api/chat" ```
GiteaMirror commented 2026-04-12 17:24:22 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 24, 2025):

I cannot replicate a long load_duration. What's the result of

curl http://localhost:11434/api/chat -d "{\"model\":\"llama3.2-vision\",\"messages\":[{\"role\":\"user\",\"content\":\"hello\"}],\"stream\":false}"
<!-- gh-comment-id:2679201822 --> @rick-github commented on GitHub (Feb 24, 2025): I cannot replicate a long load_duration. What's the result of ``` curl http://localhost:11434/api/chat -d "{\"model\":\"llama3.2-vision\",\"messages\":[{\"role\":\"user\",\"content\":\"hello\"}],\"stream\":false}" ```
GiteaMirror commented 2026-04-12 17:24:22 -05:00
Author
Owner
Copy Link

@ribbles commented on GitHub (Feb 24, 2025):

{
    "model": "llama3.2-vision",
    "created_at": "2025-02-24T18:05:44.3856736Z",
    "message": {
        "role": "assistant",
        "content": "Hello! How are you today? Is there something I can help you with or would you like to chat?"
    },
    "done_reason": "stop",
    "done": true,
    "total_duration": 16876372500,
    "load_duration": 13873535400,
    "prompt_eval_count": 11,
    "prompt_eval_duration": 1807000000,
    "eval_count": 23,
    "eval_duration": 1180000000
}
<!-- gh-comment-id:2679268577 --> @ribbles commented on GitHub (Feb 24, 2025): ``` { "model": "llama3.2-vision", "created_at": "2025-02-24T18:05:44.3856736Z", "message": { "role": "assistant", "content": "Hello! How are you today? Is there something I can help you with or would you like to chat?" }, "done_reason": "stop", "done": true, "total_duration": 16876372500, "load_duration": 13873535400, "prompt_eval_count": 11, "prompt_eval_duration": 1807000000, "eval_count": 23, "eval_duration": 1180000000 } ```
GiteaMirror commented 2026-04-12 17:24:23 -05:00
Author
Owner
Copy Link

@ribbles commented on GitHub (Feb 25, 2025):

It's the dimension of the image (3200x2400), even though it was only 200KB. I reduced the size to 1280x1024 (still 200KB) and responses are under 6 seconds. Makes sense that the compressed file size doesn't matter, just the pixel count and color depth.

I'm still uncertain why it shows up as "load_duration" in the telemetry.

<!-- gh-comment-id:2680789483 --> @ribbles commented on GitHub (Feb 25, 2025): It's the dimension of the image (3200x2400), even though it was only 200KB. I reduced the size to 1280x1024 (still 200KB) and responses are under 6 seconds. Makes sense that the compressed file size doesn't matter, just the pixel count and color depth. I'm still uncertain why it shows up as "load_duration" in the telemetry.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#6076
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?