Eval bug: llama-speculative core dump with Qwen3, GGML_ASSERT(batch.n_tokens > 0) failed

[jagusztinl]

Name and Version

version: 5307 (814f795)
built with cc (Ubuntu 14.2.0-4ubuntu2~24.04) 14.2.0 for aarch64-linux-gnu

Operating systems

Linux

GGML backends

BLAS

Hardware

Cobalt-100 Azure ARM

Models

Qwen3-4B-128K-Q4* tried all

Problem description & steps to reproduce

../build/bin/llama-speculative -m Qwen3-4B-128K-IQ4_XS.gguf -md Qwen3-4B-128K-Q4_0.gguf

First Bad Commit

No response

Relevant log output

build: 5307 (814f795e) with cc (Ubuntu 14.2.0-4ubuntu2~24.04) 14.2.0 for aarch64-linux-gnu
llama_model_loader: loaded meta data with 44 key-value pairs and 398 tensors from Qwen3-4B-128K-IQ4_XS.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              = qwen3
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Qwen3-4B-128K
llama_model_loader: - kv   3:                           general.finetune str              = 128k
llama_model_loader: - kv   4:                           general.basename str              = Qwen3-4B-128K
llama_model_loader: - kv   5:                       general.quantized_by str              = Unsloth
llama_model_loader: - kv   6:                         general.size_label str              = 4B
llama_model_loader: - kv   7:                            general.license str              = apache-2.0
llama_model_loader: - kv   8:                       general.license.link str              = https://huggingface.co/Qwen/Qwen3-4B/...
llama_model_loader: - kv   9:                           general.repo_url str              = https://huggingface.co/unsloth
llama_model_loader: - kv  10:                   general.base_model.count u32              = 1
llama_model_loader: - kv  11:                  general.base_model.0.name str              = Qwen3 4B
llama_model_loader: - kv  12:          general.base_model.0.organization str              = Qwen
llama_model_loader: - kv  13:              general.base_model.0.repo_url str              = https://huggingface.co/Qwen/Qwen3-4B
llama_model_loader: - kv  14:                               general.tags arr[str,2]       = ["unsloth", "text-generation"]
llama_model_loader: - kv  15:                          qwen3.block_count u32              = 36
llama_model_loader: - kv  16:                       qwen3.context_length u32              = 131072
llama_model_loader: - kv  17:                     qwen3.embedding_length u32              = 2560
llama_model_loader: - kv  18:                  qwen3.feed_forward_length u32              = 9728
llama_model_loader: - kv  19:                 qwen3.attention.head_count u32              = 32
llama_model_loader: - kv  20:              qwen3.attention.head_count_kv u32              = 8
llama_model_loader: - kv  21:                       qwen3.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  22:     qwen3.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  23:                 qwen3.attention.key_length u32              = 128
llama_model_loader: - kv  24:               qwen3.attention.value_length u32              = 128
llama_model_loader: - kv  25:                    qwen3.rope.scaling.type str              = yarn
llama_model_loader: - kv  26:                  qwen3.rope.scaling.factor f32              = 4.000000
llama_model_loader: - kv  27: qwen3.rope.scaling.original_context_length u32              = 32768
llama_model_loader: - kv  28:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  29:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  30:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  31:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  32:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  33:                tokenizer.ggml.eos_token_id u32              = 151645
llama_model_loader: - kv  34:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  35:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  36:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  37:                    tokenizer.chat_template str              = {%- if tools %}\n    {{- '<|im_start|>...
llama_model_loader: - kv  38:               general.quantization_version u32              = 2
llama_model_loader: - kv  39:                          general.file_type u32              = 30
llama_model_loader: - kv  40:                      quantize.imatrix.file str              = Qwen3-4B-128K-GGUF/imatrix_unsloth.dat
llama_model_loader: - kv  41:                   quantize.imatrix.dataset str              = unsloth_calibration_Qwen3-4B-128K.txt
llama_model_loader: - kv  42:             quantize.imatrix.entries_count i32              = 252
llama_model_loader: - kv  43:              quantize.imatrix.chunks_count i32              = 10
llama_model_loader: - type  f32:  145 tensors
llama_model_loader: - type q5_K:   36 tensors
llama_model_loader: - type q6_K:    1 tensors
llama_model_loader: - type iq4_xs:  216 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = IQ4_XS - 4.25 bpw
print_info: file size   = 2.11 GiB (4.50 BPW)
load: special tokens cache size = 26
load: token to piece cache size = 0.9311 MB
print_info: arch             = qwen3
print_info: vocab_only       = 0
print_info: n_ctx_train      = 131072
print_info: n_embd           = 2560
print_info: n_layer          = 36
print_info: n_head           = 32
print_info: n_head_kv        = 8
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            = 4
print_info: n_embd_k_gqa     = 1024
print_info: n_embd_v_gqa     = 1024
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             = 9728
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     = yarn
print_info: freq_base_train  = 1000000.0
print_info: freq_scale_train = 0.25
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       = 4B
print_info: model params     = 4.02 B
print_info: general.name     = Qwen3-4B-128K
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        = 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:   CPU_Mapped model buffer size =  2159.88 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    = 0.25
llama_context: n_ctx_per_seq (4096) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_context:        CPU  output buffer size =     0.58 MiB
llama_kv_cache_unified: kv_size = 4096, type_k = 'f16', type_v = 'f16', n_layer = 36, can_shift = 1, padding = 32
llama_kv_cache_unified:        CPU KV buffer size =   576.00 MiB
llama_kv_cache_unified: KV self size  =  576.00 MiB, K (f16):  288.00 MiB, V (f16):  288.00 MiB
llama_context:        CPU compute buffer size =   301.75 MiB
llama_context: graph nodes  = 1374
llama_context: graph splits = 578 (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)
llama_model_loader: loaded meta data with 44 key-value pairs and 398 tensors from Qwen3-4B-128K-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              = qwen3
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Qwen3-4B-128K
llama_model_loader: - kv   3:                           general.finetune str              = 128k
llama_model_loader: - kv   4:                           general.basename str              = Qwen3-4B-128K
llama_model_loader: - kv   5:                       general.quantized_by str              = Unsloth
llama_model_loader: - kv   6:                         general.size_label str              = 4B
llama_model_loader: - kv   7:                            general.license str              = apache-2.0
llama_model_loader: - kv   8:                       general.license.link str              = https://huggingface.co/Qwen/Qwen3-4B/...
llama_model_loader: - kv   9:                           general.repo_url str              = https://huggingface.co/unsloth
llama_model_loader: - kv  10:                   general.base_model.count u32              = 1
llama_model_loader: - kv  11:                  general.base_model.0.name str              = Qwen3 4B
llama_model_loader: - kv  12:          general.base_model.0.organization str              = Qwen
llama_model_loader: - kv  13:              general.base_model.0.repo_url str              = https://huggingface.co/Qwen/Qwen3-4B
llama_model_loader: - kv  14:                               general.tags arr[str,2]       = ["unsloth", "text-generation"]
llama_model_loader: - kv  15:                          qwen3.block_count u32              = 36
llama_model_loader: - kv  16:                       qwen3.context_length u32              = 131072
llama_model_loader: - kv  17:                     qwen3.embedding_length u32              = 2560
llama_model_loader: - kv  18:                  qwen3.feed_forward_length u32              = 9728
llama_model_loader: - kv  19:                 qwen3.attention.head_count u32              = 32
llama_model_loader: - kv  20:              qwen3.attention.head_count_kv u32              = 8
llama_model_loader: - kv  21:                       qwen3.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  22:     qwen3.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  23:                 qwen3.attention.key_length u32              = 128
llama_model_loader: - kv  24:               qwen3.attention.value_length u32              = 128
llama_model_loader: - kv  25:                    qwen3.rope.scaling.type str              = yarn
llama_model_loader: - kv  26:                  qwen3.rope.scaling.factor f32              = 4.000000
llama_model_loader: - kv  27: qwen3.rope.scaling.original_context_length u32              = 32768
llama_model_loader: - kv  28:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  29:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  30:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  31:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  32:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  33:                tokenizer.ggml.eos_token_id u32              = 151645
llama_model_loader: - kv  34:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  35:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  36:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  37:                    tokenizer.chat_template str              = {%- if tools %}\n    {{- '<|im_start|>...
llama_model_loader: - kv  38:               general.quantization_version u32              = 2
llama_model_loader: - kv  39:                          general.file_type u32              = 2
llama_model_loader: - kv  40:                      quantize.imatrix.file str              = Qwen3-4B-128K-GGUF/imatrix_unsloth.dat
llama_model_loader: - kv  41:                   quantize.imatrix.dataset str              = unsloth_calibration_Qwen3-4B-128K.txt
llama_model_loader: - kv  42:             quantize.imatrix.entries_count i32              = 252
llama_model_loader: - kv  43:              quantize.imatrix.chunks_count i32              = 10
llama_model_loader: - type  f32:  145 tensors
llama_model_loader: - type q4_0:  248 tensors
llama_model_loader: - type q4_1:    4 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   = 2.21 GiB (4.71 BPW)
load: special tokens cache size = 26
load: token to piece cache size = 0.9311 MB
print_info: arch             = qwen3
print_info: vocab_only       = 0
print_info: n_ctx_train      = 131072
print_info: n_embd           = 2560
print_info: n_layer          = 36
print_info: n_head           = 32
print_info: n_head_kv        = 8
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            = 4
print_info: n_embd_k_gqa     = 1024
print_info: n_embd_v_gqa     = 1024
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             = 9728
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     = yarn
print_info: freq_base_train  = 1000000.0
print_info: freq_scale_train = 0.25
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       = 4B
print_info: model params     = 4.02 B
print_info: general.name     = Qwen3-4B-128K
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        = 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:   CPU_Mapped model buffer size =  2246.67 MiB
load_tensors: CPU_KLEIDIAI model buffer size =  1895.62 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    = 0.25
llama_context: n_ctx_per_seq (4096) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_context:        CPU  output buffer size =     0.58 MiB
llama_kv_cache_unified: kv_size = 4096, type_k = 'f16', type_v = 'f16', n_layer = 36, can_shift = 1, padding = 32
llama_kv_cache_unified:        CPU KV buffer size =   576.00 MiB
llama_kv_cache_unified: KV self size  =  576.00 MiB, K (f16):  288.00 MiB, V (f16):  288.00 MiB
llama_context:        CPU compute buffer size =   301.75 MiB
llama_context: graph nodes  = 1374
llama_context: graph splits = 82 (with bs=512), 1 (with bs=1)
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
/home/alerant/llama.cpp/src/llama-batch.cpp:282: GGML_ASSERT(batch.n_tokens > 0) failed


../build/bin/llama-speculative(+0x1bfcd4)[0xc69700a3fcd4]
../build/bin/llama-speculative(+0x1bfe9c)[0xc69700a3fe9c]
../build/bin/llama-speculative(+0xf15bc)[0xc697009715bc]
../build/bin/llama-speculative(+0xfc234)[0xc6970097c234]
../build/bin/llama-speculative(+0x22976c)[0xc69700aa976c]
../build/bin/llama-speculative(+0x224f0)[0xc697008a24f0]
/lib/aarch64-linux-gnu/libc.so.6(+0x284c4)[0xf5652a3d84c4]
/lib/aarch64-linux-gnu/libc.so.6(__libc_start_main+0x98)[0xf5652a3d8598]
../build/bin/llama-speculative(+0x2b3f0)[0xc697008ab3f0]
Aborted (core dumped)