Misc. bug: llama-server --ctx-size is divided by --parallel and cannot be increased?

[alasdairforsythe]

Name and Version

./llama-server --version
version: 4621 (6eecde3)
built with cc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 for x86_64-linux-gnu

Operating systems

Linux

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

llama-server

Command line

./llama-server \
    --model models/deepseek-v3-q2_k_xs-00001-of-00005.gguf --alias full \
    --host 0.0.0.0 \
    --port 55055 \
    --ctx-size 0 \
    --cache-type-k q5_0 \
    --slot-save-path "saved_slots.kvc" \
    --threads 94 \
    --threads-http 12 \
    --parallel 6 \
    --mirostat 2 \
    --mirostat-ent 5.7 \
    --mirostat-lr 0.14

Problem description & steps to reproduce

Context size is being divided by number of parallel inferences allowed to run on llama-server. This does not make sense. Attempting to rectify it by increasing --ctx-size to compensate for this results in an assertion fail. In order to not break previous behavior, I suggest adding a new command line argument --ctz-size-per-seq to override this behavior.

If llama-server is run with --ctx-size 0 --parallel 6 or --ctx-size 163840 --parallel 6, it divides the total ctx by 6.

print_info: file format = GGUF V3 (latest)
print_info: file type   = Q2_K - Medium
print_info: file size   = 206.05 GiB (2.64 BPW)
[...]
print_info: arch             = deepseek2
print_info: n_ctx_train      = 163840
[...]
llama_init_from_model: n_seq_max     = 6
llama_init_from_model: n_ctx         = 163840
llama_init_from_model: n_ctx_per_seq = 27306
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     = 10000.0
llama_init_from_model: freq_scale    = 0.025
llama_init_from_model: n_ctx_per_seq (27306) < n_ctx_train (163840) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 163840, offload = 1, type_k = 'q5_0', type_v = 'f16', n_layer = 61, can_shift = 0

Okay, so I increase the ctx to 327680 so it gives more to each parallel. That almost works, but then fails on GGML_ASSERT.

llama_init_from_model: n_seq_max     = 6
llama_init_from_model: n_ctx         = 327680
llama_init_from_model: n_ctx_per_seq = 54613
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     = 10000.0
llama_init_from_model: freq_scale    = 0.025
llama_init_from_model: n_ctx_per_seq (54613) < n_ctx_train (163840) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 327680, offload = 1, type_k = 'q5_0', type_v = 'f16', n_layer = 61, can_shift = 0
llama_kv_cache_init:        CPU KV buffer size = 275232.00 MiB
llama_init_from_model: KV self size  = 275232.00 MiB, K (q5_0): 150304.00 MiB, V (f16): 124928.00 MiB
llama_init_from_model:        CPU  output buffer size =     2.96 MiB
/home/main/llama.cpp/ggml/src/ggml.c:1578: GGML_ASSERT(view_src == NULL || data_size == 0 || data_size + view_offs <= ggml_nbytes(view_src)) failed

So it's impossible to use more than 27306 ctx size, even though I have 1.1TB of RAM.

First Bad Commit

No response

Relevant log output

./llama-server \
>     --model models/deepseek-v3-q2_k_xs-00001-of-00005.gguf --alias full \
>     --host 0.0.0.0 \
>     --port 55055 \
>     --ctx-size 327680 \
>     --cache-type-k q5_0 \
>     --slot-save-path "saved_slots.kvc" \
>     --threads 94 \
>     --threads-http 12 \
>     --parallel 6 \
>     --mirostat 2 \
>     --mirostat-ent 5.7 \
>     --mirostat-lr 0.14
build: 4621 (6eecde3c) with cc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 for x86_64-linux-gnu
system info: n_threads = 94, n_threads_batch = 94, total_threads = 96

system_info: n_threads = 94 (n_threads_batch = 94) / 96 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |

main: HTTP server is listening, hostname: 0.0.0.0, port: 55055, http threads: 12
main: loading model
srv    load_model: loading model 'models/deepseek-v3-q2_k_xs-00001-of-00005.gguf'
llama_model_loader: additional 4 GGUFs metadata loaded.
llama_model_loader: loaded meta data with 46 key-value pairs and 1025 tensors from models/deepseek-v3-q2_k_xs-00001-of-00005.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              = deepseek2
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = DeepSeek V3 BF16
llama_model_loader: - kv   3:                         general.size_label str              = 256x20B
llama_model_loader: - kv   4:                      deepseek2.block_count u32              = 61
llama_model_loader: - kv   5:                   deepseek2.context_length u32              = 163840
llama_model_loader: - kv   6:                 deepseek2.embedding_length u32              = 7168
llama_model_loader: - kv   7:              deepseek2.feed_forward_length u32              = 18432
llama_model_loader: - kv   8:             deepseek2.attention.head_count u32              = 128
llama_model_loader: - kv   9:          deepseek2.attention.head_count_kv u32              = 128
llama_model_loader: - kv  10:                   deepseek2.rope.freq_base f32              = 10000.000000
llama_model_loader: - kv  11: deepseek2.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  12:                deepseek2.expert_used_count u32              = 8
llama_model_loader: - kv  13:                          general.file_type u32              = 10
llama_model_loader: - kv  14:        deepseek2.leading_dense_block_count u32              = 3
llama_model_loader: - kv  15:                       deepseek2.vocab_size u32              = 129280
llama_model_loader: - kv  16:            deepseek2.attention.q_lora_rank u32              = 1536
llama_model_loader: - kv  17:           deepseek2.attention.kv_lora_rank u32              = 512
llama_model_loader: - kv  18:             deepseek2.attention.key_length u32              = 192
llama_model_loader: - kv  19:           deepseek2.attention.value_length u32              = 128
llama_model_loader: - kv  20:       deepseek2.expert_feed_forward_length u32              = 2048
llama_model_loader: - kv  21:                     deepseek2.expert_count u32              = 256
llama_model_loader: - kv  22:              deepseek2.expert_shared_count u32              = 1
llama_model_loader: - kv  23:             deepseek2.expert_weights_scale f32              = 2.500000
llama_model_loader: - kv  24:              deepseek2.expert_weights_norm bool             = true
llama_model_loader: - kv  25:               deepseek2.expert_gating_func u32              = 2
llama_model_loader: - kv  26:             deepseek2.rope.dimension_count u32              = 64
llama_model_loader: - kv  27:                deepseek2.rope.scaling.type str              = yarn
llama_model_loader: - kv  28:              deepseek2.rope.scaling.factor f32              = 40.000000
llama_model_loader: - kv  29: deepseek2.rope.scaling.original_context_length u32              = 4096
llama_model_loader: - kv  30: deepseek2.rope.scaling.yarn_log_multiplier f32              = 0.100000
llama_model_loader: - kv  31:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  32:                         tokenizer.ggml.pre str              = deepseek-v3
llama_model_loader: - kv  33:                      tokenizer.ggml.tokens arr[str,129280]  = ["<｜begin▁of▁sentence｜>", "<�...
llama_model_loader: - kv  34:                  tokenizer.ggml.token_type arr[i32,129280]  = [3, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  35:                      tokenizer.ggml.merges arr[str,127741]  = ["Ġ t", "Ġ a", "i n", "Ġ Ġ", "h e...
llama_model_loader: - kv  36:                tokenizer.ggml.bos_token_id u32              = 0
llama_model_loader: - kv  37:                tokenizer.ggml.eos_token_id u32              = 1
llama_model_loader: - kv  38:            tokenizer.ggml.padding_token_id u32              = 1
llama_model_loader: - kv  39:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  40:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  41:                    tokenizer.chat_template str              = {% if not add_generation_prompt is de...
llama_model_loader: - kv  42:               general.quantization_version u32              = 2
llama_model_loader: - kv  43:                                   split.no u16              = 0
llama_model_loader: - kv  44:                                split.count u16              = 5
llama_model_loader: - kv  45:                        split.tensors.count i32              = 1025
llama_model_loader: - type  f32:  361 tensors
llama_model_loader: - type q2_K:  662 tensors
llama_model_loader: - type q4_K:    1 tensors
llama_model_loader: - type q6_K:    1 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q2_K - Medium
print_info: file size   = 206.05 GiB (2.64 BPW)
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 818
load: token to piece cache size = 0.8223 MB
print_info: arch             = deepseek2
print_info: vocab_only       = 0
print_info: n_ctx_train      = 163840
print_info: n_embd           = 7168
print_info: n_layer          = 61
print_info: n_head           = 128
print_info: n_head_kv        = 128
print_info: n_rot            = 64
print_info: n_swa            = 0
print_info: n_embd_head_k    = 192
print_info: n_embd_head_v    = 128
print_info: n_gqa            = 1
print_info: n_embd_k_gqa     = 24576
print_info: n_embd_v_gqa     = 16384
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: n_ff             = 18432
print_info: n_expert         = 256
print_info: n_expert_used    = 8
print_info: causal attn      = 1
print_info: pooling type     = 0
print_info: rope type        = 0
print_info: rope scaling     = yarn
print_info: freq_base_train  = 10000.0
print_info: freq_scale_train = 0.025
print_info: n_ctx_orig_yarn  = 4096
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       = 671B
print_info: model params     = 671.03 B
print_info: general.name     = DeepSeek V3 BF16
print_info: n_layer_dense_lead   = 3
print_info: n_lora_q             = 1536
print_info: n_lora_kv            = 512
print_info: n_ff_exp             = 2048
print_info: n_expert_shared      = 1
print_info: expert_weights_scale = 2.5
print_info: expert_weights_norm  = 1
print_info: expert_gating_func   = sigmoid
print_info: rope_yarn_log_mul    = 0.1000
print_info: vocab type       = BPE
print_info: n_vocab          = 129280
print_info: n_merges         = 127741
print_info: BOS token        = 0 '<｜begin▁of▁sentence｜>'
print_info: EOS token        = 1 '<｜end▁of▁sentence｜>'
print_info: EOT token        = 1 '<｜end▁of▁sentence｜>'
print_info: PAD token        = 1 '<｜end▁of▁sentence｜>'
print_info: LF token         = 201 'Ċ'
print_info: FIM PRE token    = 128801 '<｜fim▁begin｜>'
print_info: FIM SUF token    = 128800 '<｜fim▁hole｜>'
print_info: FIM MID token    = 128802 '<｜fim▁end｜>'
print_info: EOG token        = 1 '<｜end▁of▁sentence｜>'
print_info: max token length = 256
load_tensors:   CPU_Mapped model buffer size = 42690.54 MiB
load_tensors:   CPU_Mapped model buffer size = 42108.14 MiB
load_tensors:   CPU_Mapped model buffer size = 42049.55 MiB
load_tensors:   CPU_Mapped model buffer size = 42101.11 MiB
load_tensors:   CPU_Mapped model buffer size = 42049.55 MiB
llama_init_from_model: n_seq_max     = 6
llama_init_from_model: n_ctx         = 327680
llama_init_from_model: n_ctx_per_seq = 54613
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     = 10000.0
llama_init_from_model: freq_scale    = 0.025
llama_init_from_model: n_ctx_per_seq (54613) < n_ctx_train (163840) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 327680, offload = 1, type_k = 'q5_0', type_v = 'f16', n_layer = 61, can_shift = 0
llama_kv_cache_init:        CPU KV buffer size = 275232.00 MiB
llama_init_from_model: KV self size  = 275232.00 MiB, K (q5_0): 150304.00 MiB, V (f16): 124928.00 MiB
llama_init_from_model:        CPU  output buffer size =     2.96 MiB
/home/main/llama.cpp/ggml/src/ggml.c:1578: GGML_ASSERT(view_src == NULL || data_size == 0 || data_size + view_offs <= ggml_nbytes(view_src)) failed

[ggerganov]

Could be an integer overflow somewhere. Could you try to trace it down?

[alasdairforsythe]

It's line 1902 in server.cpp:

const int32_t n_ctx_slot = n_ctx / params_base.n_parallel;

I will add a new parameter to optionally override this value and push it.