yusiwen/llama.cpp Docker 镜像 - 轩辕镜像

My docker image of llama.cpp.

It is a minimal build which can run on CPU/GPU for small LLM models.

Basic usages

For CPU inferencing:

bash
# check version
$ docker run --rm yusiwen/llama.cpp:latest /main --version
version: 1879 (3e5ca79)
built with cc (GCC) 9.5.0 for x86_64-linux-gnu

# main
$ docker run --rm -v /opt/data/ai/models:/models yusiwen/llama.cpp:latest /llama-cli -m /models/mistral-7b-v0.1.Q4_K_M.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e
Log start
main: build = 1879 (3e5ca79)
main: built with cc (GCC) 9.5.0 for x86_64-linux-gnu
main: seed  = ***
llama_model_loader: loaded meta data with 20 key-value pairs and 291 tensors from /models/mistral-7b-v0.1.Q4_K_M.gguf (version GGUF V2)
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = llama
llama_model_loader: - kv   1:                               general.name str              = mistralai_mistral-7b-v0.1
llama_model_loader: - kv   2:                       llama.context_length u32              = 32768
llama_model_loader: - kv   3:                     llama.embedding_length u32              = 4096
llama_model_loader: - kv   4:                          llama.block_count u32              = 32
llama_model_loader: - kv   5:                  llama.feed_forward_length u32              = ***
llama_model_loader: - kv   6:                 llama.rope.dimension_count u32              = 128
llama_model_loader: - kv   7:                 llama.attention.head_count u32              = 32
llama_model_loader: - kv   8:              llama.attention.head_count_kv u32              = 8
llama_model_loader: - kv   9:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  10:                       llama.rope.freq_base f32              = ***.000000
llama_model_loader: - kv  11:                          general.file_type u32              = 15
llama_model_loader: - kv  12:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  13:                      tokenizer.ggml.tokens arr[str,32000]   = ["<unk>", "<s>", "</s>", "<0x00>", "<...
llama_model_loader: - kv  14:                      tokenizer.ggml.scores arr[f32,32000]   = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv  15:                  tokenizer.ggml.token_type arr[i32,32000]   = [2, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...
llama_model_loader: - kv  16:                tokenizer.ggml.bos_token_id u32              = 1
llama_model_loader: - kv  17:                tokenizer.ggml.eos_token_id u32              = 2
llama_model_loader: - kv  18:            tokenizer.ggml.unknown_token_id u32              = 0
llama_model_loader: - kv  19:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:   65 tensors
llama_model_loader: - type q4_K:  193 tensors
llama_model_loader: - type q6_K:   33 tensors
llm_load_vocab: special tokens definition check successful ( 259/32000 ).
llm_load_print_meta: format           = GGUF V2
llm_load_print_meta: arch             = llama
llm_load_print_meta: vocab type       = SPM
llm_load_print_meta: n_vocab          = 32000
llm_load_print_meta: n_merges         = 0
llm_load_print_meta: n_ctx_train      = 32768
llm_load_print_meta: n_embd           = 4096
llm_load_print_meta: n_head           = 32
llm_load_print_meta: n_head_kv        = 8
llm_load_print_meta: n_layer          = 32
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            = 4
llm_load_print_meta: n_embd_k_gqa     = 1024
llm_load_print_meta: n_embd_v_gqa     = 1024
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
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: n_ff             = ***
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = ***.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: model type       = 7B
llm_load_print_meta: model ftype      = Q4_K - Medium
llm_load_print_meta: model params     = 7.24 B
llm_load_print_meta: model size       = 4.07 GiB (4.83 BPW)
llm_load_print_meta: general.name     = mistralai_mistral-7b-v0.1
llm_load_print_meta: BOS token        = 1 '<s>'
llm_load_print_meta: EOS token        = 2 '</s>'
llm_load_print_meta: UNK token        = 0 '<unk>'
llm_load_print_meta: LF token         = 13 '<0x0A>'
llm_load_tensors: ggml ctx size =    0.11 MiB
llm_load_tensors: offloading 0 repeating layers to GPU
llm_load_tensors: offloaded 0/33 layers to GPU
llm_load_tensors:        CPU buffer size =  4165.37 MiB
...............................................................................................
llama_new_context_with_model: n_ctx      = 512
llama_new_context_with_model: freq_base  = ***.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init:        CPU KV buffer size =    64.00 MiB
llama_new_context_with_model: KV self size  =   64.00 MiB, K (f16):   32.00 MiB, V (f16):   32.00 MiB
llama_new_context_with_model: graph splits (measure): 1
llama_new_context_with_model:        CPU compute buffer size =    73.00 MiB

system_info: n_threads = 6 / 12 | AVX = 1 | AVX_VNNI = 0 | 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 |
sampling:
        repeat_last_n = 64, repeat_penalty = 1.100, frequency_penalty = 0.000, presence_penalty = 0.000
        top_k = 40, tfs_z = 1.000, top_p = 0.950, min_p = 0.050, typical_p = 1.000, temp = 0.800
        mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
sampling order:
CFG -> Penalties -> top_k -> tfs_z -> typical_p -> top_p -> min_p -> temp
generate: n_ctx = 512, n_batch = 512, n_predict = 400, n_keep = 0


 Building a website can be done in 10 simple steps:
Step 1: Pick your website name

The first step of building any website is to pick the website name you want. This is also known as a URL or domain. The most common URLs are .com, .net and .org. If you’re looking for something specific like a restaurant, then try using their local extension such as .ca for Canada.

Step 2: Set up your hosting account with the right amount of bandwidth and disk space

In order to set up your website on a server, you will need a hosting account. This is where all the files that make up your site live (images, videos, etc.). You can find many different companies online who offer these services at varying prices depending upon what features they offer. Some examples include GoDaddy or BlueHost.

Step 3: Designing Your Site Layout – Choose Themes & Plugins To Install On WordPress Website

Now that we have our hosting set up, it’s time to start designing our site layout! There are two main ways of doing this: using themes or building custom templates from scratch.

Themes provide pre-made designs for you to choose from while custom template builders allow complete control over how things look like on any given page/post within the site itself – think about it like programming languages versus HTML code. Both methods have their pros and cons; however, most people prefer using themes because they offer more flexibility when changing layouts without having any coding knowledge at all!

Step 4: Creating Pages For Your Website – Use WordPress Post Editor Or Create Custom Page Types On The Frontend With WooCommerce Plugin

Now that you’ve designed your site layout, it’s time to start creating pages for it. There are two main ways of doing this: using the default post editor or creating custom page types on the frontend with WooCommerce plugin (if you need e-commerce features

....

llama_print_timings:        load time =     448.09 ms
llama_print_timings:      sample time =      64.36 ms /   400 runs   (    0.16 ms per token,  6215.33 tokens per second)
llama_print_timings: prompt eval time =     965.08 ms /    19 tokens (   50.79 ms per token,    19.69 tokens per second)
llama_print_timings:        eval time =   42130.65 ms /   399 runs   (  105.59 ms per token,     9.47 tokens per second)
llama_print_timings:       total time =   43288.23 ms /   418 tokens
Log end

For GPU inferencing, use the image tagged with -cuda:

bash
$ docker run --rm -v /opt/data/ai/models:/models yusiwen/llama.cpp:latest-cuda /llama-cli -m /models/mistral-7b-v0.1.Q4_K_M.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e -ngl 50
...

Disclaimer

This image is builded only for my personal purpose of testing LLM inference on difference CPUs and GPUs in my own automation pipelines.

Use at your own risks.