Loading up the model like this (it's an unsloth q8_0 quant) results in a visible cuda malloc failure on warmup, but the server keeps starting and then enters a seemingly normal state, but when you give it a request to process it then dies with another malloc failure.
$ ./build/bin/llama-server -m /usr/local/ai/models/qwen3-vl-235B-A22B-Instruct/Q8_0/Qwen3-VL-235B-A22B-Instruct-Q8_0-00001-of-00006.gguf --mmproj /usr/local/ai/models/qwen3-vl-235B-A22B-Instruct/mmproj-F32.gguf --port 8081 --host 0.0.0.0 --no-mmap
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 5 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 4090, compute capability 8.9, VMM: yes
Device 3: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes
Device 4: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes
main: n_parallel is set to auto, using n_parallel = 4 and kv_unified = true
build: 7474 (f9ec8858e) with GNU 15.2.1 for Linux x86_64
system info: n_threads = 56, n_threads_batch = 56, total_threads = 112
system_info: n_threads = 56 (n_threads_batch = 56) / 112 | CUDA : ARCHS = 890,1200 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
init: using 111 threads for HTTP server
start: binding port with default address family
main: loading model
srv load_model: loading model '/usr/local/ai/models/qwen3-vl-235B-A22B-Instruct/Q8_0/Qwen3-VL-235B-A22B-Instruct-Q8_0-00001-of-00006.gguf'
common_init_result: fitting params to device memory, for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on
llama_params_fit_impl: projected memory use with initial parameters [MiB]:
llama_params_fit_impl: - CUDA0 (NVIDIA GeForce RTX 4090) : 24082 total, 28124 used, 4436 deficit
llama_params_fit_impl: - CUDA1 (NVIDIA GeForce RTX 4090) : 24082 total, 27717 used, 4029 deficit
llama_params_fit_impl: - CUDA2 (NVIDIA GeForce RTX 4090) : 24082 total, 24682 used, 995 deficit
llama_params_fit_impl: - CUDA3 (NVIDIA RTX PRO 6000 Blackwell Workstation Edition): 97250 total, 106608 used, 9919 deficit
llama_params_fit_impl: - CUDA4 (NVIDIA RTX PRO 6000 Blackwell Workstation Edition): 97250 total, 101170 used, 4481 deficit
llama_params_fit_impl: projected to use 288303 MiB of device memory vs. 266748 MiB of free device memory
llama_params_fit_impl: cannot fulfill margin of 1024 MiB on all devices, need to use 28982 MiB less in total
llama_params_fit_impl: context size reduced from 262144 to 90368 -> need 31537 MiB less memory in total
llama_params_fit_impl: entire model should be fit across devices by reducing context
llama_params_fit_impl: with only dense weights in device memory there is a total surplus of 232762 MiB
llama_params_fit_impl: filling dense-only layers back-to-front:
llama_params_fit_impl: - CUDA4 (NVIDIA RTX PRO 6000 Blackwell Workstation Edition): 95 layers, 26964 MiB used, 69723 MiB free
llama_params_fit_impl: - CUDA3 (NVIDIA RTX PRO 6000 Blackwell Workstation Edition): 0 layers, 0 MiB used, 96688 MiB free
llama_params_fit_impl: - CUDA2 (NVIDIA GeForce RTX 4090) : 0 layers, 0 MiB used, 23687 MiB free
llama_params_fit_impl: - CUDA1 (NVIDIA GeForce RTX 4090) : 0 layers, 0 MiB used, 23687 MiB free
llama_params_fit_impl: - CUDA0 (NVIDIA GeForce RTX 4090) : 0 layers, 1640 MiB used, 22047 MiB free
llama_params_fit_impl: converting dense-only layers to full layers and filling them front-to-back with overflow to next device/system memory:
llama_params_fit_impl: - CUDA0 (NVIDIA GeForce RTX 4090) : 8 layers ( 1 overflowing), 22414 MiB used, 1273 MiB free
llama_params_fit_impl: - CUDA1 (NVIDIA GeForce RTX 4090) : 8 layers ( 0 overflowing), 22662 MiB used, 1024 MiB free
llama_params_fit_impl: - CUDA2 (NVIDIA GeForce RTX 4090) : 9 layers ( 1 overflowing), 22079 MiB used, 1607 MiB free
llama_params_fit_impl: - CUDA3 (NVIDIA RTX PRO 6000 Blackwell Workstation Edition): 35 layers ( 1 overflowing), 95521 MiB used, 1167 MiB free
llama_params_fit_impl: - CUDA4 (NVIDIA RTX PRO 6000 Blackwell Workstation Edition): 35 layers ( 0 overflowing), 94327 MiB used, 2360 MiB free
llama_params_fit: successfully fit params to free device memory
llama_params_fit: fitting params to free memory took 9.25 seconds
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce RTX 4090) (0000:82:00.0) - 23687 MiB free
llama_model_load_from_file_impl: using device CUDA1 (NVIDIA GeForce RTX 4090) (0000:c1:00.0) - 23687 MiB free
llama_model_load_from_file_impl: using device CUDA2 (NVIDIA GeForce RTX 4090) (0000:c2:00.0) - 23687 MiB free
llama_model_load_from_file_impl: using device CUDA3 (NVIDIA RTX PRO 6000 Blackwell Workstation Edition) (0000:01:00.0) - 96688 MiB free
llama_model_load_from_file_impl: using device CUDA4 (NVIDIA RTX PRO 6000 Blackwell Workstation Edition) (0000:81:00.0) - 96688 MiB free
llama_model_loader: additional 5 GGUFs metadata loaded.
llama_model_loader: loaded meta data with 44 key-value pairs and 1131 tensors from /usr/local/ai/models/qwen3-vl-235B-A22B-Instruct/Q8_0/Qwen3-VL-235B-A22B-Instruct-Q8_0-00001-of-00006.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 = qwen3vlmoe
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.name str = Qwen3-Vl-235B-A22B-Instruct
llama_model_loader: - kv 3: general.finetune str = Instruct
llama_model_loader: - kv 4: general.basename str = Qwen3-Vl-235B-A22B-Instruct
llama_model_loader: - kv 5: general.quantized_by str = Unsloth
llama_model_loader: - kv 6: general.size_label str = 235B-A22B
llama_model_loader: - kv 7: general.license str = apache-2.0
llama_model_loader: - kv 8: general.repo_url str = https://huggingface.co/unsloth
llama_model_loader: - kv 9: general.base_model.count u32 = 1
llama_model_loader: - kv 10: general.base_model.0.name str = Qwen3 VL 235B A22B Instruct
llama_model_loader: - kv 11: general.base_model.0.organization str = Qwen
llama_model_loader: - kv 12: general.base_model.0.repo_url str = https://huggingface.co/Qwen/Qwen3-VL-...
llama_model_loader: - kv 13: general.tags arr[str,1] = ["unsloth"]
llama_model_loader: - kv 14: qwen3vlmoe.block_count u32 = 94
llama_model_loader: - kv 15: qwen3vlmoe.context_length u32 = 262144
llama_model_loader: - kv 16: qwen3vlmoe.embedding_length u32 = 4096
llama_model_loader: - kv 17: qwen3vlmoe.feed_forward_length u32 = 12288
llama_model_loader: - kv 18: qwen3vlmoe.attention.head_count u32 = 64
llama_model_loader: - kv 19: qwen3vlmoe.attention.head_count_kv u32 = 4
llama_model_loader: - kv 20: qwen3vlmoe.rope.freq_base f32 = 5000000.000000
llama_model_loader: - kv 21: qwen3vlmoe.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 22: qwen3vlmoe.expert_used_count u32 = 8
llama_model_loader: - kv 23: qwen3vlmoe.attention.key_length u32 = 128
llama_model_loader: - kv 24: qwen3vlmoe.attention.value_length u32 = 128
llama_model_loader: - kv 25: general.file_type u32 = 7
llama_model_loader: - kv 26: qwen3vlmoe.expert_count u32 = 128
llama_model_loader: - kv 27: qwen3vlmoe.expert_feed_forward_length u32 = 1536
llama_model_loader: - kv 28: qwen3vlmoe.rope.dimension_sections arr[i32,4] = [24, 20, 20, 0]
llama_model_loader: - kv 29: qwen3vlmoe.n_deepstack_layers u32 = 3
llama_model_loader: - kv 30: general.quantization_version u32 = 2
llama_model_loader: - kv 31: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 32: tokenizer.ggml.pre str = qwen2
llama_model_loader: - kv 33: tokenizer.ggml.tokens arr[str,151936] = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 34: tokenizer.ggml.token_type arr[i32,151936] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 35: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv 36: tokenizer.ggml.eos_token_id u32 = 151645
llama_model_loader: - kv 37: tokenizer.ggml.padding_token_id u32 = 151654
llama_model_loader: - kv 38: tokenizer.ggml.bos_token_id u32 = 151643
llama_model_loader: - kv 39: tokenizer.ggml.add_bos_token bool = false
llama_model_loader: - kv 40: tokenizer.chat_template str = {%- if tools %}\n {{- '<|im_start|>...
llama_model_loader: - kv 41: split.no u16 = 0
llama_model_loader: - kv 42: split.count u16 = 6
llama_model_loader: - kv 43: split.tensors.count i32 = 1131
llama_model_loader: - type f32: 471 tensors
llama_model_loader: - type q8_0: 660 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q8_0
print_info: file size = 232.77 GiB (8.51 BPW)
load: printing all EOG tokens:
load: - 151643 ('<|endoftext|>')
load: - 151645 ('<|im_end|>')
load: - 151662 ('<|fim_pad|>')
load: - 151663 ('<|repo_name|>')
load: - 151664 ('<|file_sep|>')
load: special tokens cache size = 26
load: token to piece cache size = 0.9311 MB
print_info: arch = qwen3vlmoe
print_info: vocab_only = 0
print_info: no_alloc = 0
print_info: n_ctx_train = 262144
print_info: n_embd = 4096
print_info: n_embd_inp = 16384
print_info: n_layer = 94
print_info: n_head = 64
print_info: n_head_kv = 4
print_info: n_rot = 128
print_info: n_swa = 0
print_info: is_swa_any = 0
print_info: n_embd_head_k = 128
print_info: n_embd_head_v = 128
print_info: n_gqa = 16
print_info: n_embd_k_gqa = 512
print_info: n_embd_v_gqa = 512
print_info: f_norm_eps = 0.0e+00
print_info: f_norm_rms_eps = 1.0e-06
print_info: f_clamp_kqv = 0.0e+00
print_info: f_max_alibi_bias = 0.0e+00
print_info: f_logit_scale = 0.0e+00
print_info: f_attn_scale = 0.0e+00
print_info: n_ff = 12288
print_info: n_expert = 128
print_info: n_expert_used = 8
print_info: n_expert_groups = 0
print_info: n_group_used = 0
print_info: causal attn = 1
print_info: pooling type = 0
print_info: rope type = 40
print_info: rope scaling = linear
print_info: freq_base_train = 5000000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn = 262144
print_info: rope_yarn_log_mul= 0.0000
print_info: rope_finetuned = unknown
print_info: mrope sections = [24, 20, 20, 0]
print_info: model type = 235B.A22B
print_info: model params = 235.09 B
print_info: general.name = Qwen3-Vl-235B-A22B-Instruct
print_info: n_ff_exp = 1536
print_info: vocab type = BPE
print_info: n_vocab = 151936
print_info: n_merges = 151387
print_info: BOS token = 151643 '<|endoftext|>'
print_info: EOS token = 151645 '<|im_end|>'
print_info: EOT token = 151645 '<|im_end|>'
print_info: PAD token = 151654 '<|vision_pad|>'
print_info: LF token = 198 'Ċ'
print_info: FIM PRE token = 151659 '<|fim_prefix|>'
print_info: FIM SUF token = 151661 '<|fim_suffix|>'
print_info: FIM MID token = 151660 '<|fim_middle|>'
print_info: FIM PAD token = 151662 '<|fim_pad|>'
print_info: FIM REP token = 151663 '<|repo_name|>'
print_info: FIM SEP token = 151664 '<|file_sep|>'
print_info: EOG token = 151643 '<|endoftext|>'
print_info: EOG token = 151645 '<|im_end|>'
print_info: EOG token = 151662 '<|fim_pad|>'
print_info: EOG token = 151663 '<|repo_name|>'
print_info: EOG token = 151664 '<|file_sep|>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = false)
load_tensors: offloading output layer to GPU
load_tensors: offloading 93 repeating layers to GPU
load_tensors: offloaded 95/95 layers to GPU
load_tensors: CUDA0 model buffer size = 19362.26 MiB
load_tensors: CUDA1 model buffer size = 20994.26 MiB
load_tensors: CUDA2 model buffer size = 20250.54 MiB
load_tensors: CUDA3 model buffer size = 89095.88 MiB
load_tensors: CUDA4 model buffer size = 88022.21 MiB
load_tensors: CUDA_Host model buffer size = 630.59 MiB
....................................................................................................
common_init_result: added <|endoftext|> logit bias = -inf
common_init_result: added <|im_end|> logit bias = -inf
common_init_result: added <|fim_pad|> logit bias = -inf
common_init_result: added <|repo_name|> logit bias = -inf
common_init_result: added <|file_sep|> logit bias = -inf
llama_context: constructing llama_context
llama_context: n_seq_max = 4
llama_context: n_ctx = 90368
llama_context: n_ctx_seq = 90368
llama_context: n_batch = 2048
llama_context: n_ubatch = 512
llama_context: causal_attn = 1
llama_context: flash_attn = auto
llama_context: kv_unified = true
llama_context: freq_base = 5000000.0
llama_context: freq_scale = 1
llama_context: n_ctx_seq (90368) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
llama_context: CUDA_Host output buffer size = 2.32 MiB
llama_kv_cache: CUDA0 KV buffer size = 1412.00 MiB
llama_kv_cache: CUDA1 KV buffer size = 1412.00 MiB
llama_kv_cache: CUDA2 KV buffer size = 1588.50 MiB
llama_kv_cache: CUDA3 KV buffer size = 6177.50 MiB
llama_kv_cache: CUDA4 KV buffer size = 6001.00 MiB
llama_kv_cache: size = 16591.00 MiB ( 90368 cells, 94 layers, 4/1 seqs), K (f16): 8295.50 MiB, V (f16): 8295.50 MiB
llama_context: Flash Attention was auto, set to enabled
llama_context: CUDA0 compute buffer size = 312.77 MiB
llama_context: CUDA1 compute buffer size = 256.53 MiB
llama_context: CUDA2 compute buffer size = 256.27 MiB
llama_context: CUDA3 compute buffer size = 248.27 MiB
llama_context: CUDA4 compute buffer size = 304.75 MiB
llama_context: CUDA_Host compute buffer size = 184.52 MiB
llama_context: graph nodes = 5929
llama_context: graph splits = 8
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
clip_model_loader: model name: Qwen3-Vl-235B-A22B-Instruct
clip_model_loader: description:
clip_model_loader: GGUF version: 3
clip_model_loader: alignment: 32
clip_model_loader: n_tensors: 352
clip_model_loader: n_kv: 31
clip_model_loader: has vision encoder
clip_ctx: CLIP using CUDA0 backend
load_hparams: Qwen-VL models require at minimum 1024 image tokens to function correctly on grounding tasks
load_hparams: if you encounter problems with accuracy, try adding --image-min-tokens 1024
load_hparams: more info: https://github.com/ggml-org/llama.cpp/issues/16842
load_hparams: projector: qwen3vl_merger
load_hparams: n_embd: 1152
load_hparams: n_head: 16
load_hparams: n_ff: 4304
load_hparams: n_layer: 27
load_hparams: ffn_op: gelu
load_hparams: projection_dim: 4096
--- vision hparams ---
load_hparams: image_size: 768
load_hparams: patch_size: 16
load_hparams: has_llava_proj: 0
load_hparams: minicpmv_version: 0
load_hparams: n_merge: 2
load_hparams: n_wa_pattern: 0
load_hparams: image_min_pixels: 8192
load_hparams: image_max_pixels: 4194304
load_hparams: model size: 2198.75 MiB
load_hparams: metadata size: 0.12 MiB
warmup: warmup with image size = 1472 x 1472
ggml_backend_cuda_buffer_type_alloc_buffer: allocating 372.08 MiB on device 0: cudaMalloc failed: out of memory
ggml_gallocr_reserve_n_impl: failed to allocate CUDA0 buffer of size 390156544
alloc_compute_meta: graph splits = 1, nodes = 853
warmup: flash attention is enabled
srv load_model: loaded multimodal model, '/usr/local/ai/models/qwen3-vl-235B-A22B-Instruct/mmproj-F32.gguf'
srv init: initializing slots, n_slots = 4
slot init: id 0 | task -1 | new slot, n_ctx = 90368
slot init: id 1 | task -1 | new slot, n_ctx = 90368
slot init: id 2 | task -1 | new slot, n_ctx = 90368
slot init: id 3 | task -1 | new slot, n_ctx = 90368
srv init: prompt cache is enabled, size limit: 8192 MiB
srv init: use `--cache-ram 0` to disable the prompt cache
srv init: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
srv init: thinking = 0
init: chat template, chat_template: {%- if tools %}
{{- '<|im_start|>system\n' }}
{%- if messages[0].role == 'system' %}
{%- if messages[0].content is string %}
{{- messages[0].content }}
{%- else %}
{%- for content in messages[0].content %}
{%- if 'text' in content %}
{{- content.text }}
{%- endif %}
{%- endfor %}
{%- endif %}
{{- '\n\n' }}
{%- endif %}
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
{%- for tool in tools %}
{{- "\n" }}
{{- tool | tojson }}
{%- endfor %}
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
{%- else %}
{%- if messages[0].role == 'system' %}
{{- '<|im_start|>system\n' }}
{%- if messages[0].content is string %}
{{- messages[0].content }}
{%- else %}
{%- for content in messages[0].content %}
{%- if 'text' in content %}
{{- content.text }}
{%- endif %}
{%- endfor %}
{%- endif %}
{{- '<|im_end|>\n' }}
{%- endif %}
{%- endif %}
{%- set image_count = namespace(value=0) %}
{%- set video_count = namespace(value=0) %}
{%- for message in messages %}
{%- if message.role == "user" %}
{{- '<|im_start|>' + message.role + '\n' }}
{%- if message.content is string %}
{{- message.content }}
{%- else %}
{%- for content in message.content %}
{%- if content.type == 'image' or 'image' in content or 'image_url' in content %}
{%- set image_count.value = image_count.value + 1 %}
{%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}
<|vision_start|><|image_pad|><|vision_end|>
{%- elif content.type == 'video' or 'video' in content %}
{%- set video_count.value = video_count.value + 1 %}
{%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}
<|vision_start|><|video_pad|><|vision_end|>
{%- elif 'text' in content %}
{{- content.text }}
{%- endif %}
{%- endfor %}
{%- endif %}
{{- '<|im_end|>\n' }}
{%- elif message.role == "assistant" %}
{{- '<|im_start|>' + message.role + '\n' }}
{%- if message.content is string %}
{{- message.content }}
{%- else %}
{%- for content_item in message.content %}
{%- if 'text' in content_item %}
{{- content_item.text }}
{%- endif %}
{%- endfor %}
{%- endif %}
{%- if message.tool_calls %}
{%- for tool_call in message.tool_calls %}
{%- if (loop.first and message.content) or (not loop.first) %}
{{- '\n' }}
{%- endif %}
{%- if tool_call.function %}
{%- set tool_call = tool_call.function %}
{%- endif %}
{{- '<tool_call>\n{"name": "' }}
{{- tool_call.name }}
{{- '", "arguments": ' }}
{%- if tool_call.arguments is string %}
{{- tool_call.arguments }}
{%- else %}
{{- tool_call.arguments | tojson }}
{%- endif %}
{{- '}\n</tool_call>' }}
{%- endfor %}
{%- endif %}
{{- '<|im_end|>\n' }}
{%- elif message.role == "tool" %}
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
{{- '<|im_start|>user' }}
{%- endif %}
{{- '\n<tool_response>\n' }}
{%- if message.content is string %}
{{- message.content }}
{%- else %}
{%- for content in message.content %}
{%- if content.type == 'image' or 'image' in content or 'image_url' in content %}
{%- set image_count.value = image_count.value + 1 %}
{%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}
<|vision_start|><|image_pad|><|vision_end|>
{%- elif content.type == 'video' or 'video' in content %}
{%- set video_count.value = video_count.value + 1 %}
{%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}
<|vision_start|><|video_pad|><|vision_end|>
{%- elif 'text' in content %}
{{- content.text }}
{%- endif %}
{%- endfor %}
{%- endif %}
{{- '\n</tool_response>' }}
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
{{- '<|im_end|>\n' }}
{%- endif %}
{%- endif %}
{%- endfor %}
{%- if add_generation_prompt %}
{{- '<|im_start|>assistant\n' }}
{%- endif %}
, example_format: '<|im_start|>system
You are a helpful assistant<|im_end|>
<|im_start|>user
Hello<|im_end|>
<|im_start|>assistant
Hi there<|im_end|>
<|im_start|>user
How are you?<|im_end|>
<|im_start|>assistant
'
main: model loaded
main: server is listening on http://0.0.0.0:8081
main: starting the main loop...
srv update_slots: all slots are idle
srv params_from_: Chat format: Hermes 2 Pro
slot get_availabl: id 3 | task -1 | selected slot by LRU, t_last = -1
slot launch_slot_: id 3 | task -1 | sampler chain: logits -> penalties -> dry -> top-n-sigma -> top-k -> typical -> top-p -> min-p -> xtc -> temp-ext -> dist
slot launch_slot_: id 3 | task 0 | processing task
slot update_slots: id 3 | task 0 | new prompt, n_ctx_slot = 90368, n_keep = 0, task.n_tokens = 8137
slot update_slots: id 3 | task 0 | n_tokens = 0, memory_seq_rm [0, end)
slot update_slots: id 3 | task 0 | prompt processing progress, n_tokens = 8, batch.n_tokens = 8, progress = 0.000983
slot update_slots: id 3 | task 0 | n_tokens = 8, memory_seq_rm [8, end)
srv process_chun: processing image...
encoding image slice...
ggml_backend_cuda_buffer_type_alloc_buffer: allocating 713.22 MiB on device 0: cudaMalloc failed: out of memory
ggml_gallocr_reserve_n_impl: failed to allocate CUDA0 buffer of size 747861504
Segmentation fault (core dumped)
Name and Version
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 5 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 4090, compute capability 8.9, VMM: yes
Device 3: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes
Device 4: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes
version: 7474 (f9ec885)
built with GNU 15.2.1 for Linux x86_64
Operating systems
Linux
Which llama.cpp modules do you know to be affected?
llama-server
Command line
Problem description & steps to reproduce
Loading up the model like this (it's an unsloth q8_0 quant) results in a visible cuda malloc failure on warmup, but the server keeps starting and then enters a seemingly normal state, but when you give it a request to process it then dies with another malloc failure.
First Bad Commit
b1f3a6e
I think it's the autofit params because without it nothing would even start?
Relevant log output