Misc. bug: GGML_ASSERT(view_src == NULL || data_size == 0 || data_size + view_offs <= ggml_nbytes(view_src)) failed

[Sv-Dmitry]

Name and Version

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 3080 Laptop GPU, compute capability 8.6, VMM: yes
load_backend: loaded CUDA backend from c:\D_Drive\Copies\ML_MODELS\llama-b5293-bin-win-cuda-cu12.4-x64\ggml-cuda.dll
load_backend: loaded RPC backend from c:\D_Drive\Copies\ML_MODELS\llama-b5293-bin-win-cuda-cu12.4-x64\ggml-rpc.dll
load_backend: loaded CPU backend from c:\D_Drive\Copies\ML_MODELS\llama-b5293-bin-win-cuda-cu12.4-x64\ggml-cpu-haswell.dll
version: 5293 (1e333d5)
built with MSVC 19.29.30159.0 for Windows AMD64

Operating systems

Windows

Which llama.cpp modules do you know to be affected?

llama-server

Command line

llama-server.exe --jinja -fa -m .\LLM\Qwen3-1.7B-UD-Q8_K_XL.gguf -c 4096 -ngl 100 -t 8 --host 192.168.1.64 --port 5000 --log-timestamps -b 2048 -ub 2048 --temp 0.0

Problem description & steps to reproduce

It seems that when the total number of tokens (parsing + generation) exceeds or approaches the limit n_ctx_slot - llama-server instead of interrupting the response, it crashes with the following error:

ggml\src\ggml.c:1554: GGML_ASSERT(view_src == NULL || data_size == 0 || data_size + view_offs <= ggml_nbytes(view_src)) failed

First Bad Commit

No response

Relevant log output

.\llama-b5293-bin-win-cuda-cu12.4-x64\llama-server.exe --jinja -fa -m .\LLM\Qwen3-1.7B-UD-Q8_K_XL.gguf -c 4096 -ngl 100 -t 8 --host 192.168.1.64 --port 5000 --log-timestamps -b 2048 -ub 2048 --temp 0.0
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 3080 Laptop GPU, compute capability 8.6, VMM: yes
load_backend: loaded CUDA backend from c:\D_Drive\Copies\ML_MODELS\llama-b5293-bin-win-cuda-cu12.4-x64\ggml-cuda.dll
load_backend: loaded RPC backend from c:\D_Drive\Copies\ML_MODELS\llama-b5293-bin-win-cuda-cu12.4-x64\ggml-rpc.dll
load_backend: loaded CPU backend from c:\D_Drive\Copies\ML_MODELS\llama-b5293-bin-win-cuda-cu12.4-x64\ggml-cpu-haswell.dll
build: 5293 (1e333d5b) with MSVC 19.29.30159.0 for Windows AMD64
system info: n_threads = 8, n_threads_batch = 8, total_threads = 16

system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CUDA : ARCHS = 500,610,700,750,800 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |

main: binding port with default address family
main: HTTP server is listening, hostname: 192.168.1.64, port: 5000, http threads: 15
main: loading model
srv    load_model: loading model '.\LLM\Qwen3-1.7B-UD-Q8_K_XL.gguf'
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce RTX 3080 Laptop GPU) - 15253 MiB free
llama_model_loader: loaded meta data with 32 key-value pairs and 310 tensors from .\LLM\Qwen3-1.7B-UD-Q8_K_XL.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              = Qwen3-1.7B
llama_model_loader: - kv   3:                           general.basename str              = Qwen3-1.7B
llama_model_loader: - kv   4:                       general.quantized_by str              = Unsloth
llama_model_loader: - kv   5:                         general.size_label str              = 1.7B
llama_model_loader: - kv   6:                           general.repo_url str              = https://huggingface.co/unsloth
llama_model_loader: - kv   7:                          qwen3.block_count u32              = 28
llama_model_loader: - kv   8:                       qwen3.context_length u32              = 40960
llama_model_loader: - kv   9:                     qwen3.embedding_length u32              = 2048
llama_model_loader: - kv  10:                  qwen3.feed_forward_length u32              = 6144
llama_model_loader: - kv  11:                 qwen3.attention.head_count u32              = 16
llama_model_loader: - kv  12:              qwen3.attention.head_count_kv u32              = 8
llama_model_loader: - kv  13:                       qwen3.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  14:     qwen3.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  15:                 qwen3.attention.key_length u32              = 128
llama_model_loader: - kv  16:               qwen3.attention.value_length u32              = 128
llama_model_loader: - kv  17:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  18:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  19:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  20:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  21:                      tokenizer.ggml.merges arr[str,151387]  = ["─а ─а", "─а─а ─а─а", "i n", "─а t",...
llama_model_loader: - kv  22:                tokenizer.ggml.eos_token_id u32              = 151645
llama_model_loader: - kv  23:            tokenizer.ggml.padding_token_id u32              = 151654
llama_model_loader: - kv  24:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  25:                    tokenizer.chat_template str              = {%- if tools %}\n    {{- '<|im_start|>...
llama_model_loader: - kv  26:               general.quantization_version u32              = 2
llama_model_loader: - kv  27:                          general.file_type u32              = 7
llama_model_loader: - kv  28:                      quantize.imatrix.file str              = Qwen3-1.7B-GGUF/imatrix_unsloth.dat
llama_model_loader: - kv  29:                   quantize.imatrix.dataset str              = unsloth_calibration_Qwen3-1.7B.txt
llama_model_loader: - kv  30:             quantize.imatrix.entries_count i32              = 196
llama_model_loader: - kv  31:              quantize.imatrix.chunks_count i32              = 32
llama_model_loader: - type  f32:  113 tensors
llama_model_loader: - type q8_0:  171 tensors
llama_model_loader: - type bf16:   26 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 2.17 GiB (10.82 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      = 40960
print_info: n_embd           = 2048
print_info: n_layer          = 28
print_info: n_head           = 16
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            = 2
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             = 6144
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  = 40960
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       = 1.7B
print_info: model params     = 1.72 B
print_info: general.name     = Qwen3-1.7B
print_info: vocab type       = BPE
print_info: n_vocab          = 151936
print_info: n_merges         = 151387
print_info: BOS token        = 11 ','
print_info: EOS token        = 151645 '<|im_end|>'
print_info: EOT token        = 151645 '<|im_end|>'
print_info: PAD token        = 151654 '<|vision_pad|>'
print_info: LF token         = 198 '─К'
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 28 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 29/29 layers to GPU
load_tensors:        CUDA0 model buffer size =  2218.85 MiB
load_tensors:   CPU_Mapped model buffer size =   593.50 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      = 2048
llama_context: causal_attn   = 1
llama_context: flash_attn    = 1
llama_context: freq_base     = 1000000.0
llama_context: freq_scale    = 1
llama_context: n_ctx_per_seq (4096) < n_ctx_train (40960) -- the full capacity of the model will not be utilized
llama_context:  CUDA_Host  output buffer size =     0.58 MiB
llama_kv_cache_unified: kv_size = 4096, type_k = 'f16', type_v = 'f16', n_layer = 28, can_shift = 1, padding = 256
llama_kv_cache_unified:      CUDA0 KV buffer size =   448.00 MiB
llama_kv_cache_unified: KV self size  =  448.00 MiB, K (f16):  224.00 MiB, V (f16):  224.00 MiB
llama_context:      CUDA0 compute buffer size =  1203.00 MiB
llama_context:  CUDA_Host compute buffer size =    48.02 MiB
llama_context: graph nodes  = 959
llama_context: graph splits = 2
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)
srv          init: initializing slots, n_slots = 1
slot         init: id  0 | task -1 | new slot n_ctx_slot = 4096
main: model loaded
main: chat template, chat_template: {%- if tools %}
    {{- '<|im_start|>system\n' }}
    {%- if messages[0].role == 'system' %}
        {{- messages[0].content + '\n\n' }}
    {%- endif %}
    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
    {%- for tool in tools %}
        {{- "\n" }}
        {{- tool | tojson }}
    {%- endfor %}
    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
{%- else %}
    {%- if messages[0].role == 'system' %}
        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
    {%- endif %}
{%- endif %}
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
{%- for forward_message in messages %}
    {%- set index = (messages|length - 1) - loop.index0 %}
    {%- set message = messages[index] %}
    {%- set tool_start = '<tool_response>' %}
    {%- set tool_start_length = tool_start|length %}
    {%- set start_of_message = message.content[:tool_start_length] %}
    {%- set tool_end = '</tool_response>' %}
    {%- set tool_end_length = tool_end|length %}
    {%- set start_pos = (message.content|length) - tool_end_length %}
    {%- if start_pos < 0 %}
        {%- set start_pos = 0 %}
    {%- endif %}
    {%- set end_of_message = message.content[start_pos:] %}
    {%- if ns.multi_step_tool and message.role == "user" and not(start_of_message == tool_start and end_of_message == tool_end) %}
        {%- set ns.multi_step_tool = false %}
        {%- set ns.last_query_index = index %}
    {%- endif %}
{%- endfor %}
{%- for message in messages %}
    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
    {%- elif message.role == "assistant" %}
        {%- set content = message.content %}
        {%- set reasoning_content = '' %}
        {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
            {%- set reasoning_content = message.reasoning_content %}
        {%- else %}
            {%- if '</think>' in message.content %}
                {%- set content = (message.content.split('</think>')|last).lstrip('\n') %}
                {%- set reasoning_content = (message.content.split('</think>')|first).rstrip('\n') %}
                {%- set reasoning_content = (reasoning_content.split('<think>')|last).lstrip('\n') %}
            {%- endif %}
        {%- endif %}
        {%- if loop.index0 > ns.last_query_index %}
            {%- if loop.last or (not loop.last and reasoning_content) %}
                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
            {%- else %}
                {{- '<|im_start|>' + message.role + '\n' + content }}
            {%- endif %}
        {%- else %}
            {{- '<|im_start|>' + message.role + '\n' + content }}
        {%- endif %}
        {%- if message.tool_calls %}
            {%- for tool_call in message.tool_calls %}
                {%- if (loop.first and content) or (not loop.first) %}
                    {{- '\n' }}
                {%- endif %}
                {%- if tool_call.function %}
                    {%- set tool_call = tool_call.function %}
                {%- endif %}
                {{- '<tool_call>\n{"name": "' }}
                {{- tool_call.name }}
                {{- '", "arguments": ' }}
                {%- if tool_call.arguments is string %}
                    {{- tool_call.arguments }}
                {%- else %}
                    {{- tool_call.arguments | tojson }}
                {%- endif %}
                {{- '}\n</tool_call>' }}
            {%- endfor %}
        {%- endif %}
        {{- '<|im_end|>\n' }}
    {%- elif message.role == "tool" %}
        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
            {{- '<|im_start|>user' }}
        {%- endif %}
        {{- '\n<tool_response>\n' }}
        {{- message.content }}
        {{- '\n</tool_response>' }}
        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
            {{- '<|im_end|>\n' }}
        {%- endif %}
    {%- endif %}
{%- endfor %}
{%- if add_generation_prompt %}
    {{- '<|im_start|>assistant\n' }}
    {%- if enable_thinking is defined and enable_thinking is false %}
        {{- '<think>\n\n</think>\n\n' }}
    {%- endif %}
{%- endif %}, example_format: '<|im_start|>system
You are a helpful assistant<|im_end|>
<|im_start|>user
Hello<|im_end|>
<|im_start|>assistant
Hi there<|im_end|>
<|im_start|>user
How are you?<|im_end|>
<|im_start|>assistant
'
main: server is listening on http://192.168.1.64:5000 - starting the main loop
srv  update_slots: all slots are idle
srv  params_from_: Chat format: Content-only
slot launch_slot_: id  0 | task 0 | processing task
slot update_slots: id  0 | task 0 | new prompt, n_ctx_slot = 4096, n_keep = 0, n_prompt_tokens = 543
slot update_slots: id  0 | task 0 | kv cache rm [0, end)
slot update_slots: id  0 | task 0 | prompt processing progress, n_past = 543, n_tokens = 543, progress = 1.000000
slot update_slots: id  0 | task 0 | prompt done, n_past = 543, n_tokens = 543
slot      release: id  0 | task 0 | stop processing: n_past = 2660, truncated = 0
slot print_timing: id  0 | task 0 |
prompt eval time =      96.55 ms /   543 tokens (    0.18 ms per token,  5624.15 tokens per second)
       eval time =   19154.37 ms /  2118 tokens (    9.04 ms per token,   110.58 tokens per second)
      total time =   19250.91 ms /  2661 tokens
srv  update_slots: all slots are idle
srv  log_server_r: request: POST /v1/chat/completions 192.168.1.64 200
srv  params_from_: Chat format: Content-only
slot launch_slot_: id  0 | task 2119 | processing task
slot update_slots: id  0 | task 2119 | new prompt, n_ctx_slot = 4096, n_keep = 0, n_prompt_tokens = 547
slot update_slots: id  0 | task 2119 | kv cache rm [538, end)
slot update_slots: id  0 | task 2119 | prompt processing progress, n_past = 547, n_tokens = 9, progress = 0.016453
slot update_slots: id  0 | task 2119 | prompt done, n_past = 547, n_tokens = 9
slot      release: id  0 | task 2119 | stop processing: n_past = 1161, truncated = 0
slot print_timing: id  0 | task 2119 |
prompt eval time =     188.43 ms /     9 tokens (   20.94 ms per token,    47.76 tokens per second)
       eval time =    5479.40 ms /   615 tokens (    8.91 ms per token,   112.24 tokens per second)
      total time =    5667.83 ms /   624 tokens
srv  update_slots: all slots are idle
srv  log_server_r: request: POST /v1/chat/completions 192.168.1.64 200
srv  params_from_: Chat format: Content-only
slot launch_slot_: id  0 | task 2735 | processing task
slot update_slots: id  0 | task 2735 | new prompt, n_ctx_slot = 4096, n_keep = 0, n_prompt_tokens = 547
slot update_slots: id  0 | task 2735 | need to evaluate at least 1 token to generate logits, n_past = 547, n_prompt_tokens = 547
slot update_slots: id  0 | task 2735 | kv cache rm [546, end)
slot update_slots: id  0 | task 2735 | prompt processing progress, n_past = 547, n_tokens = 1, progress = 0.001828
slot update_slots: id  0 | task 2735 | prompt done, n_past = 547, n_tokens = 1
slot      release: id  0 | task 2735 | stop processing: n_past = 2656, truncated = 0
slot print_timing: id  0 | task 2735 |
prompt eval time =     152.14 ms /     1 tokens (  152.14 ms per token,     6.57 tokens per second)
       eval time =   19106.85 ms /  2110 tokens (    9.06 ms per token,   110.43 tokens per second)
      total time =   19258.99 ms /  2111 tokens
srv  update_slots: all slots are idle
srv  log_server_r: request: POST /v1/chat/completions 192.168.1.64 200
srv  params_from_: Chat format: Content-only
slot launch_slot_: id  0 | task 4846 | processing task
slot update_slots: id  0 | task 4846 | new prompt, n_ctx_slot = 4096, n_keep = 0, n_prompt_tokens = 87
slot update_slots: id  0 | task 4846 | kv cache rm [1, end)
slot update_slots: id  0 | task 4846 | prompt processing progress, n_past = 87, n_tokens = 86, progress = 0.988506
slot update_slots: id  0 | task 4846 | prompt done, n_past = 87, n_tokens = 86
slot update_slots: id  0 | task 4846 | slot context shift, n_keep = 0, n_left = 4095, n_discard = 2047
D:\a\llama.cpp\llama.cpp\ggml\src\ggml.c:1554: GGML_ASSERT(view_src == NULL || data_size == 0 || data_size + view_offs <= ggml_nbytes(view_src)) failed

[Sv-Dmitry]

.\llama-b5293-bin-win-cuda-cu12.4-x64\llama-server.exe --jinja -fa -m .\LLM\Qwen3-1.7B-UD-Q8_K_XL.gguf -c 4096 -ngl 100 -t 8 --host 192.168.1.64 --port 5000 --log-timestamps -b 2048 -ub 2048 --temp 0.0

slot launch_slot_: id 0 | task 4846 | processing task
slot update_slots: id 0 | task 4846 | new prompt, n_ctx_slot = 4096, n_keep = 0, n_prompt_tokens = 87
slot update_slots: id 0 | task 4846 | kv cache rm [1, end)
slot update_slots: id 0 | task 4846 | prompt processing progress, n_past = 87, n_tokens = 86, progress = 0.988506
slot update_slots: id 0 | task 4846 | prompt done, n_past = 87, n_tokens = 86
slot update_slots: id 0 | task 4846 | slot context shift, n_keep = 0, n_left = 4095, n_discard = 2047
D:\a\llama.cpp\llama.cpp\ggml\src\ggml.c:1554: GGML_ASSERT(view_src == NULL || data_size == 0 || data_size + view_offs <= ggml_nbytes(view_src)) failed

I tried several models (old ones and "thinking") - the same results.

[xbodx]

This model doesn't work (bug):
https://huggingface.co/Qwen/QwQ-32B-GGUF/resolve/main/qwq-32b-q6_k.gguf?download=true

And this model works! (no bug):
https://huggingface.co/unsloth/Qwen3-32B-128K-GGUF/resolve/main/Qwen3-32B-128K-Q6_K.gguf?download=true

[slaren]

@ggerganov I have tracked this to kv_self_update when reserving a worst-case graph after a shift. It fails when creating the k_l view, kv_head = 2049 and n_tokens = 2048, but the context size is only 4096, so it results in an overflow. Not sure what would be the best way to handle this.
It can be reproduced with:
llama-cli -m models/Qwen3-1.7B-UD-Q8_K_XL.gguf -c 4096 -n 99999 --ignore-eos -ngl 100 -ub 2048 -fa --jinja --temp 0 -no-cnv -p 1
Model is from https://huggingface.co/unsloth/Qwen3-1.7B-GGUF

[ggerganov]

I think that this should be enough:

diff --git a/src/llama-kv-cache.cpp b/src/llama-kv-cache.cpp
index 3dcad65bb..8e160a193 100644
--- a/src/llama-kv-cache.cpp
+++ b/src/llama-kv-cache.cpp
@@ -441,6 +441,7 @@ void llama_kv_cache_unified::defrag_sched(float thold) {
 
 void llama_kv_cache_unified::set_full() {
     n = size;
+    head = 0;
 }
 
 llama_sbatch llama_kv_cache_unified::sbatch_init(

Haven't tested, just a guess atm. Will look into this further.

[slaren]

For more context, it happens after a shift + defrag, a shift alone is not enough, so it requires at least 2048 context to trigger this, and a batch size of at least half the context size. The proposed change fixes it from what I can tell.