DeepSeek-R1-UD-Q2_K_XL model inference by llama.cpp can't use flash-attention with n_embd_head_k!=n_embd_head_v
In llama.cpp, we can find
if (params.flash_attn && model->hparams.n_embd_head_k != model->hparams.n_embd_head_v) {
LLAMA_LOG_WARN("%s: flash_attn requires n_embd_head_k == n_embd_head_v - forcing off\n", __func__);
params.flash_attn = false;
}
but when using DeepSeek-R1-UD-Q2_K_XL , the log show
print_info: n_embd_head_k = 192
print_info: n_embd_head_v = 128.
how to solve this problem?
full log
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: yes
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 8 CUDA devices:
Device 0: NVIDIA L20, compute capability 8.9, VMM: yes
Device 1: NVIDIA L20, compute capability 8.9, VMM: yes
Device 2: NVIDIA L20, compute capability 8.9, VMM: yes
Device 3: NVIDIA L20, compute capability 8.9, VMM: yes
Device 4: NVIDIA L20, compute capability 8.9, VMM: yes
Device 5: NVIDIA L20, compute capability 8.9, VMM: yes
Device 6: NVIDIA L20, compute capability 8.9, VMM: yes
Device 7: NVIDIA L20, compute capability 8.9, VMM: yes
build: 4840 (3ffbbd5c) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu
system info: n_threads = 32, n_threads_batch = 32, total_threads = 96
system_info: n_threads = 32 (n_threads_batch = 32) / 96 | CUDA : ARCHS = 500,610,700,750,800 | FORCE_MMQ = 1 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | AMX_INT8 = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |
main: HTTP server is listening, hostname: 0.0.0.0, port: 9096, http threads: 95
main: loading model
srv load_model: loading model '../../../DeepSeek-R1-GGUF/DeepSeek-R1-UD-Q2_K_XL-00001-of-00005.gguf'
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA L20) - 48243 MiB free
llama_model_load_from_file_impl: using device CUDA1 (NVIDIA L20) - 45192 MiB free
llama_model_load_from_file_impl: using device CUDA2 (NVIDIA L20) - 48243 MiB free
llama_model_load_from_file_impl: using device CUDA3 (NVIDIA L20) - 45192 MiB free
llama_model_load_from_file_impl: using device CUDA4 (NVIDIA L20) - 48243 MiB free
llama_model_load_from_file_impl: using device CUDA5 (NVIDIA L20) - 45192 MiB free
llama_model_load_from_file_impl: using device CUDA6 (NVIDIA L20) - 45192 MiB free
llama_model_load_from_file_impl: using device CUDA7 (NVIDIA L20) - 48243 MiB free
llama_model_loader: additional 4 GGUFs metadata loaded.
llama_model_loader: loaded meta data with 48 key-value pairs and 1025 tensors from ../../../DeepSeek-R1-GGUF/DeepSeek-R1-UD-Q2_K_XL-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 R1 BF16
llama_model_loader: - kv 3: general.quantized_by str = Unsloth
llama_model_loader: - kv 4: general.size_label str = 256x20B
llama_model_loader: - kv 5: general.repo_url str = https://huggingface.co/unsloth
llama_model_loader: - kv 6: deepseek2.block_count u32 = 61
llama_model_loader: - kv 7: deepseek2.context_length u32 = 163840
llama_model_loader: - kv 8: deepseek2.embedding_length u32 = 7168
llama_model_loader: - kv 9: deepseek2.feed_forward_length u32 = 18432
llama_model_loader: - kv 10: deepseek2.attention.head_count u32 = 128
llama_model_loader: - kv 11: deepseek2.attention.head_count_kv u32 = 128
llama_model_loader: - kv 12: deepseek2.rope.freq_base f32 = 10000.000000
llama_model_loader: - kv 13: deepseek2.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 14: deepseek2.expert_used_count u32 = 8
llama_model_loader: - kv 15: deepseek2.leading_dense_block_count u32 = 3
llama_model_loader: - kv 16: deepseek2.vocab_size u32 = 129280
llama_model_loader: - kv 17: deepseek2.attention.q_lora_rank u32 = 1536
llama_model_loader: - kv 18: deepseek2.attention.kv_lora_rank u32 = 512
llama_model_loader: - kv 19: deepseek2.attention.key_length u32 = 192
llama_model_loader: - kv 20: deepseek2.attention.value_length u32 = 128
llama_model_loader: - kv 21: deepseek2.expert_feed_forward_length u32 = 2048
llama_model_loader: - kv 22: deepseek2.expert_count u32 = 256
llama_model_loader: - kv 23: deepseek2.expert_shared_count u32 = 1
llama_model_loader: - kv 24: deepseek2.expert_weights_scale f32 = 2.500000
llama_model_loader: - kv 25: deepseek2.expert_weights_norm bool = true
llama_model_loader: - kv 26: deepseek2.expert_gating_func u32 = 2
llama_model_loader: - kv 27: deepseek2.rope.dimension_count u32 = 64
llama_model_loader: - kv 28: deepseek2.rope.scaling.type str = yarn
llama_model_loader: - kv 29: deepseek2.rope.scaling.factor f32 = 40.000000
llama_model_loader: - kv 30: deepseek2.rope.scaling.original_context_length u32 = 4096
llama_model_loader: - kv 31: deepseek2.rope.scaling.yarn_log_multiplier f32 = 0.100000
llama_model_loader: - kv 32: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 33: tokenizer.ggml.pre str = deepseek-v3
llama_model_loader: - kv 34: tokenizer.ggml.tokens arr[str,129280] = ["<｜begin▁of▁sentence｜>", "<�...
llama_model_loader: - kv 35: tokenizer.ggml.token_type arr[i32,129280] = [3, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 36: tokenizer.ggml.merges arr[str,127741] = ["Ġ t", "Ġ a", "i n", "Ġ Ġ", "h e...
llama_model_loader: - kv 37: tokenizer.ggml.bos_token_id u32 = 0
llama_model_loader: - kv 38: tokenizer.ggml.eos_token_id u32 = 1
llama_model_loader: - kv 39: tokenizer.ggml.padding_token_id u32 = 128815
llama_model_loader: - kv 40: tokenizer.ggml.add_bos_token bool = true
llama_model_loader: - kv 41: tokenizer.ggml.add_eos_token bool = false
llama_model_loader: - kv 42: tokenizer.chat_template str = {% if not add_generation_prompt is de...
llama_model_loader: - kv 43: general.quantization_version u32 = 2
llama_model_loader: - kv 44: general.file_type u32 = 10
llama_model_loader: - kv 45: split.no u16 = 0
llama_model_loader: - kv 46: split.tensors.count i32 = 1025
llama_model_loader: - kv 47: split.count u16 = 5
llama_model_loader: - type f32: 361 tensors
llama_model_loader: - type q2_K: 171 tensors
llama_model_loader: - type q3_K: 3 tensors
llama_model_loader: - type q4_K: 306 tensors
llama_model_loader: - type q6_K: 184 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q2_K - Medium
print_info: file size = 211.03 GiB (2.70 BPW)
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 819
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 R1 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 = 128815 '<｜PAD▁TOKEN｜>'
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: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 61 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 62/62 layers to GPU
load_tensors: CPU_Mapped model buffer size = 497.11 MiB
load_tensors: CUDA0 model buffer size = 24150.90 MiB
load_tensors: CUDA1 model buffer size = 25674.66 MiB
load_tensors: CUDA2 model buffer size = 29342.47 MiB
load_tensors: CUDA3 model buffer size = 25674.66 MiB
load_tensors: CUDA4 model buffer size = 33010.28 MiB
load_tensors: CUDA5 model buffer size = 25674.66 MiB
load_tensors: CUDA6 model buffer size = 25674.66 MiB
load_tensors: CUDA7 model buffer size = 26399.64 MiB
....................................................................................................
llama_init_from_model: flash_attn requires n_embd_head_k == n_embd_head_v - forcing off
llama_init_from_model: n_seq_max = 10
llama_init_from_model: n_ctx = 2048
llama_init_from_model: n_ctx_per_seq = 204
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 (204) < n_ctx_train (163840) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 2048, offload = 1, type_k = 'q4_0', type_v = 'f16', n_layer = 61, can_shift = 0
llama_kv_cache_init: CUDA0 KV buffer size = 819.00 MiB
llama_kv_cache_init: CUDA1 KV buffer size = 637.00 MiB
llama_kv_cache_init: CUDA2 KV buffer size = 728.00 MiB
llama_kv_cache_init: CUDA3 KV buffer size = 637.00 MiB
llama_kv_cache_init: CUDA4 KV buffer size = 819.00 MiB
llama_kv_cache_init: CUDA5 KV buffer size = 637.00 MiB
llama_kv_cache_init: CUDA6 KV buffer size = 637.00 MiB
llama_kv_cache_init: CUDA7 KV buffer size = 637.00 MiB
llama_init_from_model: KV self size = 5551.00 MiB, K (q4_0): 1647.00 MiB, V (f16): 3904.00 MiB
llama_init_from_model: CUDA_Host output buffer size = 4.93 MiB
llama_init_from_model: pipeline parallelism enabled (n_copies=4)
llama_init_from_model: CUDA0 compute buffer size = 738.01 MiB
llama_init_from_model: CUDA1 compute buffer size = 738.01 MiB
llama_init_from_model: CUDA2 compute buffer size = 738.01 MiB
llama_init_from_model: CUDA3 compute buffer size = 738.01 MiB
llama_init_from_model: CUDA4 compute buffer size = 738.01 MiB
llama_init_from_model: CUDA5 compute buffer size = 738.01 MiB
llama_init_from_model: CUDA6 compute buffer size = 738.01 MiB
llama_init_from_model: CUDA7 compute buffer size = 738.02 MiB
llama_init_from_model: CUDA_Host compute buffer size = 30.02 MiB
llama_init_from_model: graph nodes = 5025
llama_init_from_model: graph splits = 9
common_init_from_params: KV cache shifting is not supported for this model, disabling KV cache shifting
common_init_from_params: setting dry_penalty_last_n to ctx_size = 2048
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
srv init: initializing slots, n_slots = 10
slot init: id 0 | task -1 | new slot n_ctx_slot = 204
slot init: id 1 | task -1 | new slot n_ctx_slot = 204
slot init: id 2 | task -1 | new slot n_ctx_slot = 204
slot init: id 3 | task -1 | new slot n_ctx_slot = 204
slot init: id 4 | task -1 | new slot n_ctx_slot = 204
slot init: id 5 | task -1 | new slot n_ctx_slot = 204
slot init: id 6 | task -1 | new slot n_ctx_slot = 204
slot init: id 7 | task -1 | new slot n_ctx_slot = 204
slot init: id 8 | task -1 | new slot n_ctx_slot = 204
slot init: id 9 | task -1 | new slot n_ctx_slot = 204
main: model loaded
main: chat template, chat_template: {% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='', is_first_sp=true) %}{%- for message in messages %}{%- if message['role'] == 'system' %}{%- if ns.is_first_sp %}{% set ns.system_prompt = ns.system_prompt + message['content'] %}{% set ns.is_first_sp = false %}{%- else %}{% set ns.system_prompt = ns.system_prompt + '\n\n' + message['content'] %}{%- endif %}{%- endif %}{%- endfor %}{{ bos_token }}{{ ns.system_prompt }}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and 'tool_calls' in message %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls'] %}{%- if not ns.is_first %}{%- if message['content'] is none %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + 'json' + '\n' + tool['function']['arguments'] + '\n' + '' + '<｜tool▁call▁end｜>'}}{%- else %}{{'<｜Assistant｜>' + message['content'] + '<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + 'json' + '\n' + tool['function']['arguments'] + '\n' + '' + '<｜tool▁call▁end｜>'}}{%- endif %}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + 'json' + '\n' + tool['function']['arguments'] + '\n' + '' + '<｜tool▁call▁end｜>'}}{%- endif %}{%- endfor %}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- if message['role'] == 'assistant' and 'tool_calls' not in message %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '' in content %}{% set content = content.split('')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜>'}}{% endif %}, example_format: 'You are a helpful assistant
<｜User｜>Hello<｜Assistant｜>Hi there<｜end▁of▁sentence｜><｜User｜>How are you?<｜Assistant｜>'
main: server is listening on http://0.0.0.0:9096 - starting the main loop
llama.cpp@main 目前尚未在 R1 上支持闪存注意力（flash attention）。相关 Pull Request 参考：https://github.com/ggml-org/llama.cpp/pull/11557
如需完整指南（包括如何通过 ktransformers 等工具运行模型），请参考我的文档：https://github.com/ubergarm/r1-ktransformers-guide/blob/main/README.zh.md (简体中文版指南)
llama.cpp@main does not support flash attention on R1 yet. https://github.com/ggml-org/llama.cpp/pull/11557
Refer to my guide for full information to run with ktransformers etc: https://github.com/ubergarm/r1-ktransformers-guide