Eval bug: Static build failing to initialize Metal backend on macOS

[johnbean393]

Name and Version

myusername@Mac bin % ./llama-server --version
version: 4641 (9f4cc8f)
built with Apple clang version 16.0.0 (clang-1600.0.26.4) for arm64-apple-darwin24.3.0

Operating systems

Mac

GGML backends

Metal

Hardware

Apple M2 Max

Models

All models

Problem description & steps to reproduce

I needed a static build of llama-server, so I cloned the repository and built it locally via cmake using this command: cmake -B build -DBUILD_SHARED_LIBS=OFF; cmake --build build --config Release -j 12 -t "llama-server".

I then proceeded to test the binary with a number of models, but the server never started successfully, citing a failure initializing the Metal backend. The output can be found below in the relevant log output section.

First Bad Commit

No response

Relevant log output

(base) bj@Mac bin % ./llama-server --model /Users/bj/Library/Application\ Support/Magic\ Sorter/Sorted\ Land/Computer\ DN/AI/Text\ Generation/Models/LLaMa\ 3.2/gguf/Meta-Llama-3.2-1B-Instruct-Q8_0.gguf
build: 4641 (9f4cc8f8) with Apple clang version 16.0.0 (clang-1600.0.26.4) for arm64-apple-darwin24.3.0
system info: n_threads = 8, n_threads_batch = 8, total_threads = 12

system_info: n_threads = 8 (n_threads_batch = 8) / 12 | Metal : EMBED_LIBRARY = 1 | CPU : NEON = 1 | ARM_FMA = 1 | FP16_VA = 1 | MATMUL_INT8 = 1 | DOTPROD = 1 | MATMUL_INT8 = 1 | ACCELERATE = 1 | AARCH64_REPACK = 1 | 

main: HTTP server is listening, hostname: 127.0.0.1, port: 8080, http threads: 11
main: loading model
srv    load_model: loading model '/Users/bj/Library/Application Support/Magic Sorter/Sorted Land/Computer DN/AI/Text Generation/Models/LLaMa 3.2/gguf/Meta-Llama-3.2-1B-Instruct-Q8_0.gguf'
llama_model_load_from_file_impl: using device Metal (Apple M2 Max) - 24575 MiB free
llama_model_loader: loaded meta data with 30 key-value pairs and 147 tensors from /Users/bj/Library/Application Support/Magic Sorter/Sorted Land/Computer DN/AI/Text Generation/Models/LLaMa 3.2/gguf/Meta-Llama-3.2-1B-Instruct-Q8_0.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              = llama
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Llama 3.2 1B Instruct
llama_model_loader: - kv   3:                           general.finetune str              = Instruct
llama_model_loader: - kv   4:                           general.basename str              = Llama-3.2
llama_model_loader: - kv   5:                         general.size_label str              = 1B
llama_model_loader: - kv   6:                               general.tags arr[str,6]       = ["facebook", "meta", "pytorch", "llam...
llama_model_loader: - kv   7:                          general.languages arr[str,8]       = ["en", "de", "fr", "it", "pt", "hi", ...
llama_model_loader: - kv   8:                          llama.block_count u32              = 16
llama_model_loader: - kv   9:                       llama.context_length u32              = 131072
llama_model_loader: - kv  10:                     llama.embedding_length u32              = 2048
llama_model_loader: - kv  11:                  llama.feed_forward_length u32              = 8192
llama_model_loader: - kv  12:                 llama.attention.head_count u32              = 32
llama_model_loader: - kv  13:              llama.attention.head_count_kv u32              = 8
llama_model_loader: - kv  14:                       llama.rope.freq_base f32              = 500000.000000
llama_model_loader: - kv  15:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  16:                 llama.attention.key_length u32              = 64
llama_model_loader: - kv  17:               llama.attention.value_length u32              = 64
llama_model_loader: - kv  18:                          general.file_type u32              = 7
llama_model_loader: - kv  19:                           llama.vocab_size u32              = 128256
llama_model_loader: - kv  20:                 llama.rope.dimension_count u32              = 64
llama_model_loader: - kv  21:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  22:                         tokenizer.ggml.pre str              = llama-bpe
llama_model_loader: - kv  23:                      tokenizer.ggml.tokens arr[str,128256]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  24:                  tokenizer.ggml.token_type arr[i32,128256]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  25:                      tokenizer.ggml.merges arr[str,280147]  = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "...
llama_model_loader: - kv  26:                tokenizer.ggml.bos_token_id u32              = 128000
llama_model_loader: - kv  27:                tokenizer.ggml.eos_token_id u32              = 128009
llama_model_loader: - kv  28:                    tokenizer.chat_template str              = {% set loop_messages = messages %}{% ...
llama_model_loader: - kv  29:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:   34 tensors
llama_model_loader: - type q8_0:  113 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 1.22 GiB (8.50 BPW) 
load: special tokens cache size = 256
load: token to piece cache size = 0.7999 MB
print_info: arch             = llama
print_info: vocab_only       = 0
print_info: n_ctx_train      = 131072
print_info: n_embd           = 2048
print_info: n_layer          = 16
print_info: n_head           = 32
print_info: n_head_kv        = 8
print_info: n_rot            = 64
print_info: n_swa            = 0
print_info: n_embd_head_k    = 64
print_info: n_embd_head_v    = 64
print_info: n_gqa            = 4
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-05
print_info: f_clamp_kqv      = 0.0e+00
print_info: f_max_alibi_bias = 0.0e+00
print_info: f_logit_scale    = 0.0e+00
print_info: n_ff             = 8192
print_info: n_expert         = 0
print_info: n_expert_used    = 0
print_info: causal attn      = 1
print_info: pooling type     = 0
print_info: rope type        = 0
print_info: rope scaling     = linear
print_info: freq_base_train  = 500000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn  = 131072
print_info: rope_finetuned   = unknown
print_info: ssm_d_conv       = 0
print_info: ssm_d_inner      = 0
print_info: ssm_d_state      = 0
print_info: ssm_dt_rank      = 0
print_info: ssm_dt_b_c_rms   = 0
print_info: model type       = 1B
print_info: model params     = 1.24 B
print_info: general.name     = Llama 3.2 1B Instruct
print_info: vocab type       = BPE
print_info: n_vocab          = 128256
print_info: n_merges         = 280147
print_info: BOS token        = 128000 '<|begin_of_text|>'
print_info: EOS token        = 128009 '<|eot_id|>'
print_info: EOT token        = 128009 '<|eot_id|>'
print_info: EOM token        = 128008 '<|eom_id|>'
print_info: LF token         = 198 'Ċ'
print_info: EOG token        = 128008 '<|eom_id|>'
print_info: EOG token        = 128009 '<|eot_id|>'
print_info: max token length = 256
load_tensors: offloading 16 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 17/17 layers to GPU
load_tensors: Metal_Mapped model buffer size =  1252.43 MiB
load_tensors:   CPU_Mapped model buffer size =   266.16 MiB
llama_init_from_model: n_seq_max     = 1
llama_init_from_model: n_ctx         = 4096
llama_init_from_model: n_ctx_per_seq = 4096
llama_init_from_model: n_batch       = 2048
llama_init_from_model: n_ubatch      = 512
llama_init_from_model: flash_attn    = 0
llama_init_from_model: freq_base     = 500000.0
llama_init_from_model: freq_scale    = 1
llama_init_from_model: n_ctx_per_seq (4096) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
ggml_metal_init: allocating
ggml_metal_init: found device: Apple M2 Max
ggml_metal_init: picking default device: Apple M2 Max
ggml_metal_init: using embedded metal library
ggml_metal_init: error: Error Domain=MTLLibraryErrorDomain Code=3 "program_source:61:35: error: unknown type name 'block_q4_0'
void dequantize_q4_0(device const block_q4_0 * xb, short il, thread type4x4 & reg) {
                                  ^
program_source:80:38: error: unknown type name 'block_q4_0'
void dequantize_q4_0_t4(device const block_q4_0 * xb, short il, thread type4 & reg) {
                                     ^
program_source:95:35: error: unknown type name 'block_q4_1'
void dequantize_q4_1(device const block_q4_1 * xb, short il, thread type4x4 & reg) {
                                  ^
program_source:114:38: error: unknown type name 'block_q4_1'
void dequantize_q4_1_t4(device const block_q4_1 * xb, short il, thread type4 & reg) {
                                     ^
program_source:129:35: error: unknown type name 'block_q5_0'
void dequantize_q5_0(device const block_q5_0 * xb, short il, thread type4x4 & reg) {
                                  ^
program_source:161:38: error: unknown type name 'block_q5_0'
void dequantize_q5_0_t4(device const block_q5_0 * xb, short il, thread type4 & reg) {
                                     ^
program_source:191:35: error: unknown type name 'block_q5_1'
void dequantize_q5_1(device const block_q5_1 * xb, short il, thread type4x4 & reg) {
                                  ^
program_source:223:38: error: unknown type name 'block_q5_1'
void dequantize_q5_1_t4(device const block_q5_1 * xb, short il, thread type4 & reg) {
                                     ^
program_source:253:35: error: unknown type name 'block_q8_0'
void dequantize_q8_0(device const block_q8_0 *xb, short il, thread type4x4 & reg) {
                                  ^
program_source:267:38: error: unknown type name 'block_q8_0'
void dequantize_q8_0_t4(device const block_q8_0 *xb, short il, thread type4 & reg) {
                                     ^
program_source:277:35: error: unknown type name 'block_q2_K'
void dequantize_q2_K(device const block_q2_K *xb, short il, thread type4x4 & reg) {
                                  ^
program_source:296:35: error: unknown type name 'block_q3_K'
void dequantize_q3_K(device const block_q3_K *xb, short il, thread type4x4 & reg) {
                                  ^
program_source:330:35: error: unknown type name 'block_q4_K'
void dequantize_q4_K(device const block_q4_K * xb, short il, thread type4x4 & reg) {
                                  ^
program_source:349:35: error: unknown type name 'block_q5_K'
void dequantize_q5_K(device const block_q5_K *xb, short il, thread type4x4 & reg) {
                                  ^
program_source:372:35: error: unknown type name 'block_q6_K'
void dequantize_q6_K(device const block_q6_K *xb, short il, thread type4x4 & reg) {
                                  ^
program_source:396:38: error: unknown type name 'block_iq2_xxs'
void dequantize_iq2_xxs(device const block_iq2_xxs * xb, short il, thread type4x4 & reg) {
                                     ^
program_source:408:52: error: use of undeclared identifier 'iq2xxs_grid'
    constant uint8_t * grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+0]);
                                                   ^
program_source:409:21: error: use of undeclared identifier 'ksigns_iq2xs'
    uint8_t signs = ksigns_iq2xs[(aux32_s >> 14*il) & 127];
                    ^
program_source:411:49: error: use of undeclared identifier 'kmask_iq2xs'
        reg[i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
                                                ^
program_source:413:33: error: use of undeclared identifier 'iq2xxs_grid'
    grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+1]);
                                ^
program_source:414:13: error: use of undeclared identifier 'ksigns_iq2xs'
    signs = ksigns_iq2xs[(aux32_s >> (14*il+7)) & 127];
            ^
program_source:416:51: error: use of undeclared identifier 'kmask_iq2xs'
        reg[2+i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
                                                  ^
program_source:421:37: error: unknown type name 'block_iq2_xs'
void dequantize_iq2_xs(device const block_iq2_xs * xb, short il, thread type4x4 & reg) {
                                    ^
program_source:429:52: error: use of undeclared identifier 'iq2xs_grid'
    constant uint8_t * grid = (constant uint8_t *)(iq2xs_grid + (q2[2*il+0] & 511));
                                                   ^
program_source:430:21: error: use of undeclared identifier 'ksigns_iq2xs'
    uint8_t signs = ksigns_iq2xs[q2[2*il+0] >> 9];
                    ^
.
.
.
More similar errors here (could not include it in the issue due to character limit)
.
.
.
            ^
program_source:6634:82: error: explicit instantiation of 'kernel_mul_mv_id' does not refer to a function template, variable template, member function, member class, or static data member
template [[host_name("kernel_mul_mv_id_iq4_xs_f32")]]  kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq4_xs_f32_impl>>;
                                                                                 ^
program_source:6547:13: note: explicit instantiation refers here
kernel void kernel_mul_mv_id(
            ^
}
ggml_backend_metal_device_init: error: failed to allocate context
llama_init_from_model: failed to initialize Metal backend
common_init_from_params: failed to create context with model '/Users/bj/Library/Application Support/Magic Sorter/Sorted Land/Computer DN/AI/Text Generation/Models/LLaMa 3.2/gguf/Meta-Llama-3.2-1B-Instruct-Q8_0.gguf'
srv    load_model: failed to load model, '/Users/bj/Library/Application Support/Magic Sorter/Sorted Land/Computer DN/AI/Text Generation/Models/LLaMa 3.2/gguf/Meta-Llama-3.2-1B-Instruct-Q8_0.gguf'
main: exiting due to model loading error

[dannoll]

Hi @johnbean393 — did you ever resolve this issue? I’m seeing something very similar on M3 Max with Metal enabled and would love to know if you found a workaround or if this was a config issue on your end. Thanks in advance!

[damien-king]

I'm also hitting this, built

#main: binding port with default address family
main: HTTP server is listening, hostname: 127.0.0.1, port: 8080, http threads: 19
main: loading model
srv load_model: loading model '~/models/nomic-ai/nomic-embed-code-GGUF/nomic-embed-code.Q4_0.gguf'
llama_model_load_from_file_impl: using device Metal (Apple M1 Ultra) - 49151 MiB free
llama_model_loader: loaded meta data with 55 key-value pairs and 338 tensors from ~/models/nomic-ai/nomic-embed-code-GGUF/nomic-embed-code.Q4_0.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv 0: general.architecture str = qwen2
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.name str = nomic-embed-code
llama_model_loader: - kv 3: general.size_label str = 7.1B
llama_model_loader: - kv 4: general.license str = apache-2.0
llama_model_loader: - kv 5: general.base_model.count u32 = 1
llama_model_loader: - kv 6: general.base_model.0.name str = Qwen2.5 Coder 7B Instruct
llama_model_loader: - kv 7: general.base_model.0.organization str = Qwen
llama_model_loader: - kv 8: general.base_model.0.repo_url str = https://huggingface.co/Qwen/Qwen2.5-C...
llama_model_loader: - kv 9: general.dataset.count u32 = 6
llama_model_loader: - kv 10: general.dataset.0.name str = Cornstack Python v1
llama_model_loader: - kv 11: general.dataset.0.version str = v1
llama_model_loader: - kv 12: general.dataset.0.organization str = Nomic Ai
llama_model_loader: - kv 13: general.dataset.0.repo_url str = https://huggingface.co/nomic-ai/corns...
llama_model_loader: - kv 14: general.dataset.1.name str = Cornstack Javascript v1
llama_model_loader: - kv 15: general.dataset.1.version str = v1
llama_model_loader: - kv 16: general.dataset.1.organization str = Nomic Ai
llama_model_loader: - kv 17: general.dataset.1.repo_url str = https://huggingface.co/nomic-ai/corns...
llama_model_loader: - kv 18: general.dataset.2.name str = Cornstack Java v1
llama_model_loader: - kv 19: general.dataset.2.version str = v1
llama_model_loader: - kv 20: general.dataset.2.organization str = Nomic Ai
llama_model_loader: - kv 21: general.dataset.2.repo_url str = https://huggingface.co/nomic-ai/corns...
llama_model_loader: - kv 22: general.dataset.3.name str = Cornstack Go v1
llama_model_loader: - kv 23: general.dataset.3.version str = v1
llama_model_loader: - kv 24: general.dataset.3.organization str = Nomic Ai
llama_model_loader: - kv 25: general.dataset.3.repo_url str = https://huggingface.co/nomic-ai/corns...
llama_model_loader: - kv 26: general.dataset.4.name str = Cornstack Php v1
llama_model_loader: - kv 27: general.dataset.4.version str = v1
llama_model_loader: - kv 28: general.dataset.4.organization str = Nomic Ai
llama_model_loader: - kv 29: general.dataset.4.repo_url str = https://huggingface.co/nomic-ai/corns...
llama_model_loader: - kv 30: general.dataset.5.name str = Cornstack Ruby v1
llama_model_loader: - kv 31: general.dataset.5.version str = v1
llama_model_loader: - kv 32: general.dataset.5.organization str = Nomic Ai
llama_model_loader: - kv 33: general.dataset.5.repo_url str = https://huggingface.co/nomic-ai/corns...
llama_model_loader: - kv 34: general.tags arr[str,4] = ["sentence-transformers", "sentence-s...
llama_model_loader: - kv 35: qwen2.block_count u32 = 28
llama_model_loader: - kv 36: qwen2.context_length u32 = 32768
llama_model_loader: - kv 37: qwen2.embedding_length u32 = 3584
llama_model_loader: - kv 38: qwen2.feed_forward_length u32 = 18944
llama_model_loader: - kv 39: qwen2.attention.head_count u32 = 28
llama_model_loader: - kv 40: qwen2.attention.head_count_kv u32 = 4
llama_model_loader: - kv 41: qwen2.rope.freq_base f32 = 1000000.000000
llama_model_loader: - kv 42: qwen2.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 43: qwen2.pooling_type u32 = 3
llama_model_loader: - kv 44: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 45: tokenizer.ggml.pre str = qwen2
llama_model_loader: - kv 46: tokenizer.ggml.tokens arr[str,152064] = ["!", """, "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 47: tokenizer.ggml.token_type arr[i32,152064] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 48: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv 49: tokenizer.ggml.eos_token_id u32 = 151645
llama_model_loader: - kv 50: tokenizer.ggml.padding_token_id u32 = 151643
llama_model_loader: - kv 51: tokenizer.ggml.bos_token_id u32 = 151643
llama_model_loader: - kv 52: tokenizer.ggml.add_bos_token bool = false
llama_model_loader: - kv 53: general.quantization_version u32 = 2
llama_model_loader: - kv 54: general.file_type u32 = 2
llama_model_loader: - type f32: 141 tensors
llama_model_loader: - type q4_0: 196 tensors
llama_model_loader: - type q6_K: 1 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q4_0
print_info: file size = 3.84 GiB (4.66 BPW)
load: special tokens cache size = 22
load: token to piece cache size = 0.9310 MB
print_info: arch = qwen2
print_info: vocab_only = 0
print_info: n_ctx_train = 32768
print_info: n_embd = 3584
print_info: n_layer = 28
print_info: n_head = 28
print_info: n_head_kv = 4
print_info: n_rot = 128
print_info: n_swa = 0
print_info: n_swa_pattern = 1
print_info: n_embd_head_k = 128
print_info: n_embd_head_v = 128
print_info: n_gqa = 7
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 = 18944
print_info: n_expert = 0
print_info: n_expert_used = 0
print_info: causal attn = 1
print_info: pooling type = 0
print_info: rope type = 2
print_info: rope scaling = linear
print_info: freq_base_train = 1000000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn = 32768
print_info: rope_finetuned = unknown
print_info: ssm_d_conv = 0
print_info: ssm_d_inner = 0
print_info: ssm_d_state = 0
print_info: ssm_dt_rank = 0
print_info: ssm_dt_b_c_rms = 0
print_info: model type = 7B
print_info: model params = 7.07 B
print_info: general.name = nomic-embed-code
print_info: vocab type = BPE
print_info: n_vocab = 152064
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 = 151643 '<|endoftext|>'
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 = true)
load_tensors: offloading 1 repeating layers to GPU
load_tensors: offloaded 1/29 layers to GPU
load_tensors: Metal_Mapped model buffer size = 125.06 MiB
load_tensors: CPU_Mapped model buffer size = 3803.01 MiB
..........................................................................................
llama_context: constructing llama_context
llama_context: n_seq_max = 1
llama_context: n_ctx = 4096
llama_context: n_ctx_per_seq = 4096
llama_context: n_batch = 2048
llama_context: n_ubatch = 512
llama_context: causal_attn = 1
llama_context: flash_attn = 0
llama_context: freq_base = 1000000.0
llama_context: freq_scale = 1
llama_context: n_ctx_per_seq (4096) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
ggml_metal_init: allocating
ggml_metal_init: found device: Apple M1 Ultra
ggml_metal_init: picking default device: Apple M1 Ultra
ggml_metal_load_library: using embedded metal library
ggml_metal_load_library: error: Error Domain=MTLLibraryErrorDomain Code=3 "program_source:65:35: error: unknown type name 'block_q4_0'
void dequantize_q4_0(device const block_q4_0 * xb, short il, thread type4x4 & reg){

Built using

cmake -S . -B build -DGGML_METAL=on -DLLAMA_SERVER=on
cmake --build build --parallel

[dannoll]

@damien-king and others encountering similar errors on certain Macbook M1/M2/M3 configurations, you may want to try with the following flag in the configuration step: -DGGML_METAL_EMBED_LIBRARY=OFF. Metal will still be enabled and layers will still be offloaded to GPU, but the build process will use a Metal library external to the binary at runtime.

cmake -B build -DGGML_METAL=ON -DGGML_METAL_EMBED_LIBRARY=OFF
cmake --build build --config Release

[dannoll]

The root cause of this issue (for me, possibly for others): for embedded Metal builds, the build script was mishandling project directory paths that contained spaces. Here's the fix, now merged: #14126