llama_model_load: error loading model: unable to allocate backend buffer

[phaelon74]

OS: Windows 11, running Text Generation WebUI, up to date on all releases.
Processor: Intel Core i5-8500 3GHz (6 Cores - no HT)
Memory: 16GB System Memory
GPUs: Five nVidia RTX 3600 - 12GB VRAM versions (First iteration during Covid)

Model: Coomand-R-35B-v1-OLD_Q4_K_M.gguf

Model Parameters:

• n-gpu-layers: 41 (41 of 41, loading FULLY into VRAM)
• n_ctx: 8192
• tensor split: 10,10,10,10,10
• flash-attn: Checked
• tensorcores: checked
• no-mmap: checked

Output from Model Load:

08:02:50-987413 INFO     Loading "Coomand-R-35B-v1-OLD_Q4_K_M.gguf"
08:02:51-580810 INFO     llama.cpp weights detected: "models\Coomand-R-35B-v1-OLD_Q4_K_M.gguf"
llama_model_loader: loaded meta data with 23 key-value pairs and 322 tensors from models\Coomand-R-35B-v1-OLD_Q4_K_M.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              = command-r
llama_model_loader: - kv   1:                               general.name str              = workspace
llama_model_loader: - kv   2:                      command-r.block_count u32              = 40
llama_model_loader: - kv   3:                   command-r.context_length u32              = 131072
llama_model_loader: - kv   4:                 command-r.embedding_length u32              = 8192
llama_model_loader: - kv   5:              command-r.feed_forward_length u32              = 22528
llama_model_loader: - kv   6:             command-r.attention.head_count u32              = 64
llama_model_loader: - kv   7:          command-r.attention.head_count_kv u32              = 64
llama_model_loader: - kv   8:                   command-r.rope.freq_base f32              = 8000000.000000
llama_model_loader: - kv   9:     command-r.attention.layer_norm_epsilon f32              = 0.000010
llama_model_loader: - kv  10:                          general.file_type u32              = 15
llama_model_loader: - kv  11:                      command-r.logit_scale f32              = 0.062500
llama_model_loader: - kv  12:                command-r.rope.scaling.type str              = none
llama_model_loader: - kv  13:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  14:                      tokenizer.ggml.tokens arr[str,256000]  = ["<PAD>", "<UNK>", "<CLS>", "<SEP>", ...
llama_model_loader: - kv  15:                  tokenizer.ggml.token_type arr[i32,256000]  = [3, 3, 3, 3, 3, 3, 3, 3, 1, 1, 1, 1, ...
llama_model_loader: - kv  16:                      tokenizer.ggml.merges arr[str,253333]  = ["─á ─á", "─á t", "e r", "i n", "─á a...
llama_model_loader: - kv  17:                tokenizer.ggml.bos_token_id u32              = 5
llama_model_loader: - kv  18:                tokenizer.ggml.eos_token_id u32              = 255001
llama_model_loader: - kv  19:            tokenizer.ggml.padding_token_id u32              = 0
llama_model_loader: - kv  20:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  21:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  22:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:   41 tensors
llama_model_loader: - type q4_K:  240 tensors
llama_model_loader: - type q6_K:   41 tensors
llm_load_vocab: missing pre-tokenizer type, using: 'default'
llm_load_vocab:
llm_load_vocab: ************************************
llm_load_vocab: GENERATION QUALITY WILL BE DEGRADED!
llm_load_vocab: CONSIDER REGENERATING THE MODEL
llm_load_vocab: ************************************
llm_load_vocab:
llm_load_vocab: special tokens definition check successful ( 1008/256000 ).
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = command-r
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 256000
llm_load_print_meta: n_merges         = 253333
llm_load_print_meta: n_ctx_train      = 131072
llm_load_print_meta: n_embd           = 8192
llm_load_print_meta: n_head           = 64
llm_load_print_meta: n_head_kv        = 64
llm_load_print_meta: n_layer          = 40
llm_load_print_meta: n_rot            = 128
llm_load_print_meta: n_embd_head_k    = 128
llm_load_print_meta: n_embd_head_v    = 128
llm_load_print_meta: n_gqa            = 1
llm_load_print_meta: n_embd_k_gqa     = 8192
llm_load_print_meta: n_embd_v_gqa     = 8192
llm_load_print_meta: f_norm_eps       = 1.0e-05
llm_load_print_meta: f_norm_rms_eps   = 0.0e+00
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    = 6.2e-02
llm_load_print_meta: n_ff             = 22528
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     = none
llm_load_print_meta: freq_base_train  = 8000000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_yarn_orig_ctx  = 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: model type       = 35B
llm_load_print_meta: model ftype      = Q4_K - Medium
llm_load_print_meta: model params     = 34.98 B
llm_load_print_meta: model size       = 20.04 GiB (4.92 BPW)
llm_load_print_meta: general.name     = workspace
llm_load_print_meta: BOS token        = 5 '<BOS_TOKEN>'
llm_load_print_meta: EOS token        = 255001 '<|END_OF_TURN_TOKEN|>'
llm_load_print_meta: PAD token        = 0 '<PAD>'
llm_load_print_meta: LF token         = 136 'Ä'
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:   no
ggml_cuda_init: CUDA_USE_TENSOR_CORES: yes
ggml_cuda_init: found 5 CUDA devices:
  Device 0: NVIDIA GeForce RTX 3060, compute capability 8.6, VMM: yes
  Device 1: NVIDIA GeForce RTX 3060, compute capability 8.6, VMM: yes
  Device 2: NVIDIA GeForce RTX 3060, compute capability 8.6, VMM: yes
  Device 3: NVIDIA GeForce RTX 3060, compute capability 8.6, VMM: yes
  Device 4: NVIDIA GeForce RTX 3060, compute capability 8.6, VMM: yes
llm_load_tensors: ggml ctx size =    1.01 MiB
ggml_backend_cuda_buffer_type_alloc_buffer: allocating 5099.12 MiB on device 4: cudaMalloc failed: out of memory
llama_model_load: error loading model: unable to allocate backend buffer
llama_load_model_from_file: failed to load model
CUDA error: out of memory
  current device: 4, in function ggml_backend_cuda_host_buffer_free_buffer at D:\a\llama-cpp-python-cuBLAS-wheels\llama-cpp-python-cuBLAS-wheels\vendor\llama.cpp\ggml-cuda.cu:993
  cudaFreeHost(buffer->context)
GGML_ASSERT: D:\a\llama-cpp-python-cuBLAS-wheels\llama-cpp-python-cuBLAS-wheels\vendor\llama.cpp\ggml-cuda.cu:61: !"CUDA error"

This really doesn't make any sense to me, as a 35B paramter at Q4 should load into 50GB VRAM without issue.

[slaren]

The error is exactly what it says, a call to cudaMalloc failed with an "out of memory" error, which is something outside the control of llama.cpp. NVIDIA may be able to help you with that, but I suspect that increasing the amount of system memory would fix the issue, I have found that CUDA allocations can fail when low on system memory.

[phaelon74]

I moved all cards to a new system, with 128GB of system ram, same issue is occuring. The model loads without issue on transformers in 4bit, 8 bit, and full load in vram. It also loads in Q8 as a GGUF on the same system. Something about command-R is not happy with llama.cpp. @slaren it uses ~77GB of system ram and 96GB of vram before bombing when I choose llama.cpp. That's not normal. A 35B model should fit just fine in 96GB vram and 128GB system memory at 8192 context.