【大模型】大模型 CPU 推理之 llama.cpp

llama.cpp

• 描述

  The main goal of llama.cpp is to enable LLM inference with minimal setup and state-of-the-art performance on a wide variety of hardware - locally and in the cloud.

  • Plain C/C++ implementation without any dependencies
  • Apple silicon is a first-class citizen - optimized via ARM NEON, Accelerate and Metal frameworks
  • AVX, AVX2 and AVX512 support for x86 architectures
  • 1.5-bit, 2-bit, 3-bit, 4-bit, 5-bit, 6-bit, and 8-bit integer quantization for faster inference and reduced memory use
  • Custom CUDA kernels for running LLMs on NVIDIA GPUs (support for AMD GPUs via HIP)
  • Vulkan, SYCL, and (partial) OpenCL backend support
  • CPU+GPU hybrid inference to partially accelerate models larger than the total VRAM capacity

• 官网
  https://github.com/ggerganov/llama.cpp

• Supported platforms:

   Mac OSLinuxWindows (via CMake)DockerFreeBSD
  

• Supported models:

  • Typically finetunes of the base models below are supported as well.

  LLaMA 🦙
  LLaMA 2 🦙🦙
  Mistral 7B
  Mixtral MoE
  Falcon
  Chinese LLaMA / Alpaca and Chinese LLaMA-2 / Alpaca-2
  Vigogne (French)
  Koala
  Baichuan 1 & 2 + derivations
  Aquila 1 & 2
  Starcoder models
  Refact
  Persimmon 8B
  MPT
  Bloom
  Yi models
  StableLM models
  Deepseek models
  Qwen models
  PLaMo-13B
  Phi models
  GPT-2
  Orion 14B
  InternLM2
  CodeShell
  Gemma
  Mamba
  Xverse
  Command-R

  • Multimodal models:

  LLaVA 1.5 models, LLaVA 1.6 models
  BakLLaVA
  Obsidian
  ShareGPT4V
  MobileVLM 1.7B/3B models
  Yi-VL

安装llama.cpp

• 下载代码

  git clone https://github.com/ggerganov/llama.cpp

• Build
  On Linux or MacOS:

  cd llama.cppmake
  

  其他编译方法参考官网https://github.com/ggerganov/llama.cpp

Memory/Disk Requirements

Quantization

测试推理

下载模型

快速下载模型，参考： 无需 VPN 即可急速下载 huggingface 上的 LLM 模型
我这里下 qwen/Qwen1.5-1.8B-Chat-GGUF 进行测试

huggingface-cli download --resume-download  qwen/Qwen1.5-1.8B-Chat-GGUF  --local-dir  qwen/Qwen1.5-1.8B-Chat-GGUF

测试

cd ./llama.cpp./main -m /your/path/qwen/Qwen1.5-1.8B-Chat-GGUF/qwen1_5-1_8b-chat-q4_k_m.gguf -n 512 --color -i -cml -f ./prompts/chat-with-qwen.txt

需要修改提示语，可以编辑 ./prompts/chat-with-qwen.txt 进行修改。

加载模型输出信息：

llama.cpp# ./main -m /mnt/data/llm/Qwen1.5-1.8B-Chat-GGUF/qwen1_5-1_8b-chat-q4_k_m.gguf -n 512 --color -i -cml -f ./prompts/chat-with-qwen
.txt
Log start
main: build = 2527 (ad3a0505)
main: built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu
main: seed  = 1711760850
llama_model_loader: loaded meta data with 21 key-value pairs and 291 tensors from /mnt/data/llm/Qwen1.5-1.8B-Chat-GGUF/qwen1_5-1_8b-chat-q4_k_m.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              = qwen2
llama_model_loader: - kv   1:                               general.name str              = Qwen1.5-1.8B-Chat-AWQ-fp16
llama_model_loader: - kv   2:                          qwen2.block_count u32              = 24
llama_model_loader: - kv   3:                       qwen2.context_length u32              = 32768
llama_model_loader: - kv   4:                     qwen2.embedding_length u32              = 2048
llama_model_loader: - kv   5:                  qwen2.feed_forward_length u32              = 5504
llama_model_loader: - kv   6:                 qwen2.attention.head_count u32              = 16
llama_model_loader: - kv   7:              qwen2.attention.head_count_kv u32              = 16
llama_model_loader: - kv   8:     qwen2.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv   9:                       qwen2.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  10:                qwen2.use_parallel_residual bool             = true
llama_model_loader: - kv  11:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  12:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  13:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  14:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  15:                tokenizer.ggml.eos_token_id u32              = 151645
llama_model_loader: - kv  16:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  17:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  18:                    tokenizer.chat_template str              = {% for message in messages %}{{'<|im_...
llama_model_loader: - kv  19:               general.quantization_version u32              = 2
llama_model_loader: - kv  20:                          general.file_type u32              = 15
llama_model_loader: - type  f32:  121 tensors
llama_model_loader: - type q5_0:   12 tensors
llama_model_loader: - type q8_0:   12 tensors
llama_model_loader: - type q4_K:  133 tensors
llama_model_loader: - type q6_K:   13 tensors
llm_load_vocab: special tokens definition check successful ( 293/151936 ).
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = qwen2
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 151936
llm_load_print_meta: n_merges         = 151387
llm_load_print_meta: n_ctx_train      = 32768
llm_load_print_meta: n_embd           = 2048
llm_load_print_meta: n_head           = 16
llm_load_print_meta: n_head_kv        = 16
llm_load_print_meta: n_layer          = 24
llm_load_print_meta: n_rot            = 128
llm_load_print_meta: n_embd_head_k    = 128
llm_load_print_meta: n_embd_head_v    = 128
llm_load_print_meta: n_gqa            = 1
llm_load_print_meta: n_embd_k_gqa     = 2048
llm_load_print_meta: n_embd_v_gqa     = 2048
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-06
llm_load_print_meta: f_clamp_kqv      = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale    = 0.0e+00
llm_load_print_meta: n_ff             = 5504
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0
llm_load_print_meta: causal attn      = 1
llm_load_print_meta: pooling type     = 0
llm_load_print_meta: rope type        = 2
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 1000000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_yarn_orig_ctx  = 32768
llm_load_print_meta: rope_finetuned   = unknown
llm_load_print_meta: ssm_d_conv       = 0
llm_load_print_meta: ssm_d_inner      = 0
llm_load_print_meta: ssm_d_state      = 0
llm_load_print_meta: ssm_dt_rank      = 0
llm_load_print_meta: model type       = 1B
llm_load_print_meta: model ftype      = Q4_K - Medium
llm_load_print_meta: model params     = 1.84 B
llm_load_print_meta: model size       = 1.13 GiB (5.28 BPW)
llm_load_print_meta: general.name     = Qwen1.5-1.8B-Chat-AWQ-fp16
llm_load_print_meta: BOS token        = 151643 '<|endoftext|>'
llm_load_print_meta: EOS token        = 151645 '<|im_end|>'
llm_load_print_meta: PAD token        = 151643 '<|endoftext|>'
llm_load_print_meta: LF token         = 148848 'ÄĬ'
llm_load_tensors: ggml ctx size =    0.11 MiB
llm_load_tensors:        CPU buffer size =  1155.67 MiB
...................................................................
llama_new_context_with_model: n_ctx      = 512
llama_new_context_with_model: n_batch    = 512
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: freq_base  = 1000000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init:        CPU KV buffer size =    96.00 MiB
llama_new_context_with_model: KV self size  =   96.00 MiB, K (f16):   48.00 MiB, V (f16):   48.00 MiB
llama_new_context_with_model:        CPU  output buffer size =   296.75 MiB
llama_new_context_with_model:        CPU compute buffer size =   300.75 MiB
llama_new_context_with_model: graph nodes  = 868
llama_new_context_with_model: graph splits = 1system_info: n_threads = 4 / 4 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 |
main: interactive mode on.
Reverse prompt: '<|im_start|>user
'
sampling:repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000top_k = 40, tfs_z = 1.000, top_p = 0.950, min_p = 0.050, typical_p = 1.000, temp = 0.800mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
sampling order:
CFG -> Penalties -> top_k -> tfs_z -> typical_p -> top_p -> min_p -> temperature
generate: n_ctx = 512, n_batch = 2048, n_predict = 512, n_keep = 10== Running in interactive mode. ==- Press Ctrl+C to interject at any time.- Press Return to return control to LLaMa.- To return control without starting a new line, end your input with '/'.- If you want to submit another line, end your input with '\'.system
You are a helpful assistant.
user>

输入文本：What’s AI?

输出示例：

参考

• https://github.com/ggerganov/llama.cpp