Eval bug: GGML_ASSERT(hparams.n_embd_head_k % ggml_blck_size(type_k) == 0) failed

[AbdullahMPrograms]

Name and Version

ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no ggml_cuda_init: found 3 CUDA devices: Device 0: Tesla P40, compute capability 6.1, VMM: yes Device 1: Tesla P40, compute capability 6.1, VMM: yes Device 2: Tesla P40, compute capability 6.1, VMM: yes version: 4753 (51f311e0) built with cc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 for x86_64-linux-gnu

Operating systems

Linux

GGML backends

CUDA

Hardware

3x Telsa P40

Models

LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct

Problem description & steps to reproduce

Hello all, I am currently facing an issue with loading EXAONE-3.5-2.4B-Instruct on llama.cpp, log output is below, I similarly had issues when using split mode row with smaller models such as Llama3.2-3B, mostly likely due to the small tensor sizes. Any help is appreciated!

First Bad Commit

No response

Relevant log output

/home/ultimis/LLM/llama.cpp/build/bin/llama-server -m /home/ultimis/LLM/Models/EXAONE-3.5-2.4B-Instruct-Q8_0.gguf -c 32768 -ngl 99 --split-mode none --flash-attn --host 0.0.0.0 --port 8103 -ctv q8_0 -ctk q8_0
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 3 CUDA devices:
  Device 0: Tesla P40, compute capability 6.1, VMM: yes
  Device 1: Tesla P40, compute capability 6.1, VMM: yes
  Device 2: Tesla P40, compute capability 6.1, VMM: yes
build: 4753 (51f311e0) with cc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 for x86_64-linux-gnu
system info: n_threads = 16, n_threads_batch = 16, total_threads = 32

system_info: n_threads = 16 (n_threads_batch = 16) / 32 | CUDA : ARCHS = 520 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | AVX512 = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |

main: HTTP server is listening, hostname: 0.0.0.0, port: 8103, http threads: 31
main: loading model
srv    load_model: loading model '/home/ultimis/LLM/Models/EXAONE-3.5-2.4B-Instruct-Q8_0.gguf'
llama_model_load_from_file_impl: using device CUDA0 (Tesla P40) - 24007 MiB free
llama_model_loader: loaded meta data with 36 key-value pairs and 274 tensors from /home/ultimis/LLM/Models/EXAONE-3.5-2.4B-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              = exaone
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = EXAONE 3.5 2.4B Instruct
llama_model_loader: - kv   3:                           general.finetune str              = Instruct
llama_model_loader: - kv   4:                           general.basename str              = EXAONE-3.5
llama_model_loader: - kv   5:                         general.size_label str              = 2.4B
llama_model_loader: - kv   6:                            general.license str              = other
llama_model_loader: - kv   7:                       general.license.name str              = exaone
llama_model_loader: - kv   8:                       general.license.link str              = LICENSE
llama_model_loader: - kv   9:                               general.tags arr[str,4]       = ["lg-ai", "exaone", "exaone-3.5", "te...
llama_model_loader: - kv  10:                          general.languages arr[str,2]       = ["en", "ko"]
llama_model_loader: - kv  11:                    exaone.embedding_length u32              = 2560
llama_model_loader: - kv  12:                exaone.attention.head_count u32              = 32
llama_model_loader: - kv  13:             exaone.attention.head_count_kv u32              = 8
llama_model_loader: - kv  14:                      exaone.context_length u32              = 32768
llama_model_loader: - kv  15:    exaone.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  16:                 exaone.feed_forward_length u32              = 7168
llama_model_loader: - kv  17:                         exaone.block_count u32              = 30
llama_model_loader: - kv  18:                          general.file_type u32              = 7
llama_model_loader: - kv  19:                      exaone.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  20:                exaone.rope.dimension_count u32              = 80
llama_model_loader: - kv  21:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  22:                         tokenizer.ggml.pre str              = exaone
llama_model_loader: - kv  23:                      tokenizer.ggml.tokens arr[str,102400]  = ["[PAD]", "[BOS]", "[EOS]", "[UNK]", ...
llama_model_loader: - kv  24:                  tokenizer.ggml.token_type arr[i32,102400]  = [3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, ...
llama_model_loader: - kv  25:                      tokenizer.ggml.merges arr[str,101782]  = ["t h", "Ġ a", "Ġ í", "i n", "Ġ t...
llama_model_loader: - kv  26:                tokenizer.ggml.bos_token_id u32              = 1
llama_model_loader: - kv  27:                tokenizer.ggml.eos_token_id u32              = 361
llama_model_loader: - kv  28:            tokenizer.ggml.unknown_token_id u32              = 3
llama_model_loader: - kv  29:            tokenizer.ggml.padding_token_id u32              = 0
llama_model_loader: - kv  30:                    tokenizer.chat_template str              = {% for message in messages %}{% if lo...
llama_model_loader: - kv  31:               general.quantization_version u32              = 2
llama_model_loader: - kv  32:                      quantize.imatrix.file str              = /models_out/EXAONE-3.5-2.4B-Instruct-...
llama_model_loader: - kv  33:                   quantize.imatrix.dataset str              = /training_dir/calibration_datav3.txt
llama_model_loader: - kv  34:             quantize.imatrix.entries_count i32              = 210
llama_model_loader: - kv  35:              quantize.imatrix.chunks_count i32              = 137
llama_model_loader: - type  f32:   62 tensors
llama_model_loader: - type q8_0:  212 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 2.64 GiB (8.50 BPW)
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 362
load: token to piece cache size = 0.6622 MB
print_info: arch             = exaone
print_info: vocab_only       = 0
print_info: n_ctx_train      = 32768
print_info: n_embd           = 2560
print_info: n_layer          = 30
print_info: n_head           = 32
print_info: n_head_kv        = 8
print_info: n_rot            = 80
print_info: n_swa            = 0
print_info: n_embd_head_k    = 80
print_info: n_embd_head_v    = 80
print_info: n_gqa            = 4
print_info: n_embd_k_gqa     = 640
print_info: n_embd_v_gqa     = 640
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             = 7168
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       = ?B
print_info: model params     = 2.67 B
print_info: general.name     = EXAONE 3.5 2.4B Instruct
print_info: vocab type       = BPE
print_info: n_vocab          = 102400
print_info: n_merges         = 101782
print_info: BOS token        = 1 '[BOS]'
print_info: EOS token        = 361 '[|endofturn|]'
print_info: EOT token        = 42 '<|endoftext|>'
print_info: UNK token        = 3 '[UNK]'
print_info: PAD token        = 0 '[PAD]'
print_info: LF token         = 560 'Ċ'
print_info: EOG token        = 42 '<|endoftext|>'
print_info: EOG token        = 361 '[|endofturn|]'
print_info: max token length = 48
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 30 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 31/31 layers to GPU
load_tensors:        CUDA0 model buffer size =  2437.71 MiB
load_tensors:   CPU_Mapped model buffer size =   265.62 MiB
...................................................................................
/home/ultimis/LLM/llama.cpp/src/llama.cpp:9695: GGML_ASSERT(hparams.n_embd_head_k % ggml_blck_size(type_k) == 0) failed
llama_init_from_model: n_seq_max     = 1
llama_init_from_model: n_ctx         = 32768
llama_init_from_model: n_ctx_per_seq = 32768
llama_init_from_model: n_batch       = 2048
llama_init_from_model: n_ubatch      = 512
llama_init_from_model: flash_attn    = 1
llama_init_from_model: freq_base     = 1000000.0
llama_init_from_model: freq_scale    = 1
Could not attach to process.  If your uid matches the uid of the target
process, check the setting of /proc/sys/kernel/yama/ptrace_scope, or try
again as the root user.  For more details, see /etc/sysctl.d/10-ptrace.conf
ptrace: Operation not permitted.
No stack.
The program is not being run.