Bug: Failed to run qwen2-57b-a14b-instruct-fp16.

[tang-t21]

What happened?

I am trying to run Qwen2-57B-A14B-instruct, and I used llama-gguf-split to merge the gguf files from Qwen/Qwen2-57B-A14B-Instruct-GGUF. But it's aborted with terminate called after throwing an instance of 'std::length_error' what(): vector::_M_default_append Aborted (core dumped)。

Name and Version

./build/bin/llama-cli --version
version: 3808 (699a0dc)
built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu

What operating system are you seeing the problem on?

Linux

Relevant log output

`(llama) root@201edf3683be:/home/llama.cpp# ./build/bin/llama-cli -m ./models/qwen2-57b-a14b-instruct-fp16.gguf -p "Beijing is the capital of" -n 64 -c 4096
build: 3808 (699a0dc1) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu (debug)
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_loader: loaded meta data with 28 key-value pairs and 479 tensors from ./models/qwen2-57b-a14b-instruct-fp16.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              = qwen2moe
llama_model_loader: - kv   1:                               general.name str              = Qwen2-MoE-A14.2B-Chat
llama_model_loader: - kv   2:                       qwen2moe.block_count u32              = 28
llama_model_loader: - kv   3:                    qwen2moe.context_length u32              = 32768
llama_model_loader: - kv   4:                  qwen2moe.embedding_length u32              = 3584
llama_model_loader: - kv   5:              qwen2moe.attention.head_count u32              = 28
llama_model_loader: - kv   6:           qwen2moe.attention.head_count_kv u32              = 4
llama_model_loader: - kv   7:                    qwen2moe.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv   8:  qwen2moe.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv   9:                 qwen2moe.expert_used_count u32              = 8
llama_model_loader: - kv  10:                      qwen2moe.expert_count u32              = 64
llama_model_loader: - kv  11:        qwen2moe.expert_feed_forward_length u32              = 2560
llama_model_loader: - kv  12:               qwen2moe.feed_forward_length u32              = 20480
llama_model_loader: - kv  13:                          general.file_type u32              = 1
llama_model_loader: - kv  14:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  15:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  16:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  17:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  18:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  19:                tokenizer.ggml.eos_token_id u32              = 151643
llama_model_loader: - kv  20:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  21:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  22:                    tokenizer.chat_template str              = {% for message in messages %}{{'<|im_...
llama_model_loader: - kv  23:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  24:               general.quantization_version u32              = 2
llama_model_loader: - kv  25:                                   split.no u16              = 0
llama_model_loader: - kv  26:                                split.count u16              = 0
llama_model_loader: - kv  27:                        split.tensors.count i32              = 479
llama_model_loader: - type  f32:  197 tensors
llama_model_loader: - type  f16:  282 tensors
llm_load_vocab: special tokens cache size = 293
llm_load_vocab: token to piece cache size = 0.9338 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = qwen2moe
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 151936
llm_load_print_meta: n_merges         = 151387
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 32768
llm_load_print_meta: n_embd           = 3584
llm_load_print_meta: n_layer          = 28
llm_load_print_meta: n_head           = 28
llm_load_print_meta: n_head_kv        = 4
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            = 7
llm_load_print_meta: n_embd_k_gqa     = 512
llm_load_print_meta: n_embd_v_gqa     = 512
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-06
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             = 20480
llm_load_print_meta: n_expert         = 64
llm_load_print_meta: n_expert_used    = 8
llm_load_print_meta: causal attn      = 1
llm_load_print_meta: pooling type     = 0
llm_load_print_meta: rope type        = 2
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 1000000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 32768
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       = 57B.A14B
llm_load_print_meta: model ftype      = F16
llm_load_print_meta: model params     = 57.41 B
llm_load_print_meta: model size       = 106.94 GiB (16.00 BPW) 
llm_load_print_meta: general.name     = Qwen2-MoE-A14.2B-Chat
llm_load_print_meta: BOS token        = 151643 '<|endoftext|>'
llm_load_print_meta: EOS token        = 151643 '<|endoftext|>'
llm_load_print_meta: PAD token        = 151643 '<|endoftext|>'
llm_load_print_meta: LF token         = 148848 'ÄĬ'
llm_load_print_meta: EOT token        = 151645 '<|im_end|>'
llm_load_print_meta: max token length = 256
llm_load_print_meta: n_ff_exp         = 2560
llm_load_print_meta: n_ff_shexp       = 0
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 4 CUDA devices:
  Device 0: NVIDIA H100 80GB HBM3, compute capability 9.0, VMM: yes
  Device 1: NVIDIA H100 80GB HBM3, compute capability 9.0, VMM: yes
  Device 2: NVIDIA H100 80GB HBM3, compute capability 9.0, VMM: yes
  Device 3: NVIDIA H100 80GB HBM3, compute capability 9.0, VMM: yes
llm_load_tensors: ggml ctx size =    0.20 MiB
llm_load_tensors: offloading 0 repeating layers to GPU
llm_load_tensors: offloaded 0/29 layers to GPU
llm_load_tensors:        CPU buffer size = 109511.40 MiB
.............................................................................................
llama_new_context_with_model: n_ctx      = 4096
llama_new_context_with_model: n_batch    = 2048
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: flash_attn = 0
llama_new_context_with_model: freq_base  = 1000000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init:  CUDA_Host KV buffer size =   224.00 MiB
llama_new_context_with_model: KV self size  =  224.00 MiB, K (f16):  112.00 MiB, V (f16):  112.00 MiB
llama_new_context_with_model:  CUDA_Host  output buffer size =     0.58 MiB
ggml_gallocr_reserve_n: reallocating CUDA0 buffer from size 0.00 MiB to 1349.38 MiB
ggml_gallocr_reserve_n: reallocating CUDA1 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA2 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA3 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA_Host buffer from size 0.00 MiB to 15.01 MiB
llama_new_context_with_model:      CUDA0 compute buffer size =  1349.38 MiB
llama_new_context_with_model:  CUDA_Host compute buffer size =    15.01 MiB
llama_new_context_with_model: graph nodes  = 1910
llama_new_context_with_model: graph splits = 536
llama_init_from_gpt_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
ggml_backend_sched_alloc_splits: failed to allocate graph, reserving (backend_ids_changed = 1)
main: llama threadpool init, n_threads = 128

system_info: n_threads = 128 (n_threads_batch = 128) / 255 | AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | FMA = 1 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | RISCV_VECT = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 | 

terminate called after throwing an instance of 'std::length_error'
  what():  vector::_M_default_append
Aborted (core dumped)`

[ggerganov]

Does it work if you add -ngl 99?

[tang-t21]

No, still the same error

[tang-t21]

Also the same error in running Deepseek-V2-Lite model.

[phymbert]

Cannot reproduce on a single GPU:

llama-cli --hf-repo Qwen/Qwen2-57B-A14B-Instruct-GGUF --hf-file qwen2-57b-a14b-instruct-q3_k_m.gguf -p "Beijing is the capital of" -n 64 -c 4096

Output

/home/phymbert/workspaces/llama.cpp/cmake-build-debug/bin/llama-cli --hf-repo Qwen/Qwen2-57B-A14B-Instruct-GGUF --hf-file qwen2-57b-a14b-instruct-q3_k_m.gguf -p "Beijing is the capital of" -n 64 -c 4096
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: NVIDIA GeForce RTX 3050 Laptop GPU, compute capability 8.6, VMM: yes
register_backend: registered backend CUDA (1 devices)
register_device: registered device CUDA0 (NVIDIA GeForce RTX 3050 Laptop GPU)
register_backend: registered backend CPU (1 devices)
register_device: registered device CPU (11th Gen Intel(R) Core(TM) i5-11400H @ 2.70GHz)
build: 4393 (d79d8f39) with cc (Ubuntu 13.2.0-23ubuntu4) 13.2.0 for x86_64-linux-gnu (debug)
main: llama backend init
main: load the model and apply lora adapter, if any
common_download_file: no previous model file found /home/phymbert/.cache/llama.cpp/Qwen_Qwen2-57B-A14B-Instruct-GGUF_qwen2-57b-a14b-instruct-q3_k_m.gguf
curl_perform_with_retry: Trying to download from https://huggingface.co/Qwen/Qwen2-57B-A14B-Instruct-GGUF/resolve/main/qwen2-57b-a14b-instruct-q3_k_m.gguf (attempt 1 of 3)...
common_download_file: trying to download model from https://huggingface.co/Qwen/Qwen2-57B-A14B-Instruct-GGUF/resolve/main/qwen2-57b-a14b-instruct-q3_k_m.gguf to /home/phymbert/.cache/llama.cpp/Qwen_Qwen2-57B-A14B-Instruct-GGUF_qwen2-57b-a14b-instruct-q3_k_m.gguf (server_etag:"0d637675f1adf9b07557cf8312b98396-1000", server_last_modified:Mon, 17 Jun 2024 04:40:23 GMT)...
curl_perform_with_retry: Trying to download from https://huggingface.co/Qwen/Qwen2-57B-A14B-Instruct-GGUF/resolve/main/qwen2-57b-a14b-instruct-q3_k_m.gguf (attempt 1 of 3)...
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  1146  100  1146    0     0  10544      0 --:--:-- --:--:-- --:--:--     0
100 25.6G  100 25.6G    0     0  39.9M      0  0:10:56  0:10:56 --:--:-- 37.8M
common_download_file: file metadata saved: /home/phymbert/.cache/llama.cpp/Qwen_Qwen2-57B-A14B-Instruct-GGUF_qwen2-57b-a14b-instruct-q3_k_m.gguf.json
llama_load_model_from_file: using device CUDA0 (NVIDIA GeForce RTX 3050 Laptop GPU) - 3765 MiB free
llama_model_loader: loaded meta data with 25 key-value pairs and 479 tensors from /home/phymbert/.cache/llama.cpp/Qwen_Qwen2-57B-A14B-Instruct-GGUF_qwen2-57b-a14b-instruct-q3_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              = qwen2moe
llama_model_loader: - kv   1:                               general.name str              = Qwen2-MoE-A14.2B-Chat
llama_model_loader: - kv   2:                       qwen2moe.block_count u32              = 28
llama_model_loader: - kv   3:                    qwen2moe.context_length u32              = 32768
llama_model_loader: - kv   4:                  qwen2moe.embedding_length u32              = 3584
llama_model_loader: - kv   5:              qwen2moe.attention.head_count u32              = 28
llama_model_loader: - kv   6:           qwen2moe.attention.head_count_kv u32              = 4
llama_model_loader: - kv   7:                    qwen2moe.rope.freq_base f32              = 1000000,000000
llama_model_loader: - kv   8:  qwen2moe.attention.layer_norm_rms_epsilon f32              = 0,000001
llama_model_loader: - kv   9:                 qwen2moe.expert_used_count u32              = 8
llama_model_loader: - kv  10:                      qwen2moe.expert_count u32              = 64
llama_model_loader: - kv  11:        qwen2moe.expert_feed_forward_length u32              = 2560
llama_model_loader: - kv  12:               qwen2moe.feed_forward_length u32              = 20480
llama_model_loader: - kv  13:                          general.file_type u32              = 12
llama_model_loader: - kv  14:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  15:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  16:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  17:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  18:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  19:                tokenizer.ggml.eos_token_id u32              = 151643
llama_model_loader: - kv  20:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  21:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  22:                    tokenizer.chat_template str              = {% for message in messages %}{{'<|im_...
llama_model_loader: - kv  23:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  24:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  197 tensors
llama_model_loader: - type q3_K:  169 tensors
llama_model_loader: - type q4_K:  108 tensors
llama_model_loader: - type q5_K:    4 tensors
llama_model_loader: - type q6_K:    1 tensors
llm_load_vocab: special tokens cache size = 293
llm_load_vocab: token to piece cache size = 0,9338 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = qwen2moe
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 151936
llm_load_print_meta: n_merges         = 151387
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 32768
llm_load_print_meta: n_embd           = 3584
llm_load_print_meta: n_layer          = 28
llm_load_print_meta: n_head           = 28
llm_load_print_meta: n_head_kv        = 4
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            = 7
llm_load_print_meta: n_embd_k_gqa     = 512
llm_load_print_meta: n_embd_v_gqa     = 512
llm_load_print_meta: f_norm_eps       = 0,0e+00
llm_load_print_meta: f_norm_rms_eps   = 1,0e-06
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             = 20480
llm_load_print_meta: n_expert         = 64
llm_load_print_meta: n_expert_used    = 8
llm_load_print_meta: causal attn      = 1
llm_load_print_meta: pooling type     = 0
llm_load_print_meta: rope type        = 2
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 1000000,0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 32768
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       = 57B.A14B
llm_load_print_meta: model ftype      = Q3_K - Medium
llm_load_print_meta: model params     = 57,41 B
llm_load_print_meta: model size       = 25,61 GiB (3,83 BPW) 
llm_load_print_meta: general.name     = Qwen2-MoE-A14.2B-Chat
llm_load_print_meta: BOS token        = 151643 '<|endoftext|>'
llm_load_print_meta: EOS token        = 151643 '<|endoftext|>'
llm_load_print_meta: EOT token        = 151645 '<|im_end|>'
llm_load_print_meta: PAD token        = 151643 '<|endoftext|>'
llm_load_print_meta: LF token         = 148848 'ÄĬ'
llm_load_print_meta: EOG token        = 151643 '<|endoftext|>'
llm_load_print_meta: EOG token        = 151645 '<|im_end|>'
llm_load_print_meta: max token length = 256
llm_load_print_meta: n_ff_exp         = 2560
llm_load_print_meta: n_ff_shexp       = 0
llm_load_tensors: offloading 0 repeating layers to GPU
llm_load_tensors: offloaded 0/29 layers to GPU
llm_load_tensors:   CPU_Mapped model buffer size = 26225,29 MiB
................................................................................................
llama_new_context_with_model: n_seq_max     = 1
llama_new_context_with_model: n_ctx         = 4096
llama_new_context_with_model: n_ctx_per_seq = 4096
llama_new_context_with_model: n_batch       = 2048
llama_new_context_with_model: n_ubatch      = 512
llama_new_context_with_model: flash_attn    = 0
llama_new_context_with_model: freq_base     = 1000000,0
llama_new_context_with_model: freq_scale    = 1
llama_new_context_with_model: n_ctx_per_seq (4096) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 4096, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 28
llama_kv_cache_init:        CPU KV buffer size =   224,00 MiB
llama_new_context_with_model: KV self size  =  224,00 MiB, K (f16):  112,00 MiB, V (f16):  112,00 MiB
llama_new_context_with_model:        CPU  output buffer size =     0,58 MiB
llama_new_context_with_model:      CUDA0 compute buffer size =   806,64 MiB
llama_new_context_with_model:  CUDA_Host compute buffer size =    15,01 MiB
llama_new_context_with_model: graph nodes  = 1910
llama_new_context_with_model: graph splits = 536 (with bs=512), 1 (with bs=1)
common_init_from_params: setting dry_penalty_last_n to ctx_size = 4096
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
main: llama threadpool init, n_threads = 6

system_info: n_threads = 6 (n_threads_batch = 6) / 12 | CUDA : ARCHS = 860 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 | 

sampler seed: 1924651909
sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = 4096
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000
sampler chain: logits -> logit-bias -> penalties -> dry -> top-k -> typical -> top-p -> min-p -> xtc -> temp-ext -> dist 
generate: n_ctx = 4096, n_batch = 2048, n_predict = 64, n_keep = 0

Beijing is the capital of China, and it is a city with a long history and rich culture. It is also the political, economic, and cultural center of China. Beijing is located in the north of China, and it is a megacity with a population of more than 20 million people. The city is famous for its ancient architecture

llama_perf_sampler_print:    sampling time =     138,62 ms /    70 runs   (    1,98 ms per token,   504,97 tokens per second)
llama_perf_context_print:        load time =   20535,79 ms
llama_perf_context_print: prompt eval time =    9858,11 ms /     6 tokens ( 1643,02 ms per token,     0,61 tokens per second)
llama_perf_context_print:        eval time =  186457,86 ms /    63 runs   ( 2959,65 ms per token,     0,34 tokens per second)
llama_perf_context_print:       total time =  196604,93 ms /    69 tokens

Process finished with exit code 0

So if the issue still exists, may it be related to split on multiple H100 on your output ?:

ggml_gallocr_reserve_n: reallocating CUDA0 buffer from size 0.00 MiB to 1349.38 MiB
ggml_gallocr_reserve_n: reallocating CUDA1 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA2 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA3 buffer from size 0.00 MiB to 0.00 MiB

@tang-t21 Could you dump the stack trace with gdb ? just run gdb --args ./llama-cli ... and type bt to get the call stack when it crashes.