ããã¯ããªã«ãããããŠæžãããã®ïŒ
llama-cpp-pythonã䜿ããšOpenAI APIäºæã®ãµãŒããŒãç«ãŠãããããšãç¥ã£ãã®ã§ãã¡ãã£ãšåãããŠã¿ãŸããã
llama-cpp-python
llama-cpp-pythonã®GitHubãªããžããªãŒã¯ãã¡ãã
GitHub - abetlen/llama-cpp-python: Python bindings for llama.cpp
ããã¥ã¡ã³ãã¯ãã¡ãã§ãã
llama-cpp-pythonã¯ãllama.cppã®Pythonãã€ã³ãã£ã³ã°ã§ãã
GitHub - ggerganov/llama.cpp: Port of Facebook's LLaMA model in C/C++
Llama
ã䜿ã£ãã¢ã¯ã»ã¹ãã§ããã®ã§ããã
OpenAI APIäºæã®WebãµãŒããŒãå®è£ ããŠããŸãã
ä»åã¯ãã¡ããè©ŠããŠã¿ãŸãã
ãªããèšèãããããããããªãã£ããããã®ã§ãä»åã¯ãšããããåãã â çšèªãèŠãŠããããšããæµãã«ããããšæããŸãã
ç°å¢
ä»åã®ç°å¢ã¯ãã¡ãã
Ubuntu Linux 22.04 LTSã
$ lsb_release -a No LSB modules are available. Distributor ID: Ubuntu Description: Ubuntu 22.04.3 LTS Release: 22.04 Codename: jammy $ uname -srvmpio Linux 5.15.0-89-generic #99-Ubuntu SMP Mon Oct 30 20:42:41 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux
Pythonã
$ python3 -V Python 3.10.12 $ pip3 -V pip 22.0.2 from /usr/lib/python3/dist-packages/pip (python 3.10)
ã¹ããã¯ã
$ cat /proc/cpuinfo | grep 'model name' model name : Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz model name : Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz model name : Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz model name : Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz model name : Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz model name : Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz model name : Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz model name : Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz $ free -h total used free shared buff/cache available Mem: 15Gi 1.0Gi 12Gi 120Mi 1.8Gi 14Gi Swap: 2.0Gi 0B 2.0Gi
GPUã¯ãããŸããã
llama-cpp-pythonã§OpenAI APIäºæã®ãµãŒããŒãç«ãŠã
ã§ã¯ããŸãã¯llama-cpp-pythonã®ãµãŒããŒã¢ãžã¥ãŒã«ãã€ã³ã¹ããŒã«ããŸãã
$ pip3 install llama-cpp-python[server]
ããŒãžã§ã³ã
$ pip3 freeze | grep llama_cpp_python llama_cpp_python==0.2.19
å®è¡ã«ã¯ã¢ãã«ãå¿
èŠã§ããä»åã¯llama-2-7b-chat.Q4_K_M.gguf
ã䜿ãããšã«ããŸãã
TheBloke/Llama-2-7B-Chat-GGUF · Hugging Face
ã¢ãã«ã®ããŠã³ããŒãã4GBãããŸãã
$ curl -LO https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF/resolve/main/llama-2-7b-chat.Q4_K_M.gguf
ã¢ãã«ãããŠã³ããŒããããããµãŒããŒãèµ·åã
$ python3 -m llama_cpp.server --model llama-2-7b-chat.Q4_K_M.gguf
以äžã®ãããªè¡šç€ºãåºãããæºåå®äºã§ããååèµ·åã¯ãã£ããåŸ ã€ã¿ããã§ãã
INFO: Started server process [11251] INFO: Waiting for application startup. INFO: Application startup complete. INFO: Uvicorn running on http://localhost:8000 (Press CTRL+C to quit)
ã§ã¯ãOpenAIã®APIãåãããŠã¿ãŸããOpenAI APIã®ããã¥ã¡ã³ãã¯ãã¡ãã§ããã
ä»åã¯ã以äžãè©ŠããŠã¿ãŸãã
OpenAI / API reference / ENDPOINTS / Chat / Create chat completion
èªå·±çŽ¹ä»ããé¡ãããŠã¿ãŸãã
â»OpenAI APIã®å®çŸ©ãšç°ãªããmodel
ãã©ã¡ãŒã¿ãŒããªããŠãè¯ãã¿ããã§ã
$ curl -s -XPOST -H 'Content-Type: application/json' localhost:8000/v1/chat/completions -d \ '{"messages": [{"role": "user", "content": "Could you introduce yourself?"}]}' | jq
çµæã
{ "id": "chatcmpl-6754401d-4666-4362-aeb0-d76fb0a17a38", "object": "chat.completion", "created": 1700978076, "model": "llama-2-7b-chat.Q4_K_M.gguf", "choices": [ { "index": 0, "message": { "content": " Hello! I'm LLaMA, I'm a large language model trained by a team of researcher at Meta AI.\nMy primary function is to understand and respond to human input in a helpful and engaging manner. I can answer questions, provide information, and even generate text based on a given prompt or topic. My knowledge was built from a massive corpus of text data, including books, articles, and websites, which I use to learn patterns and relationships in language.\nI'm here to help you with any questions or topics you'd like to discuss, from simple queries to more in-depth conversations. Please feel free to ask me anything!", "role": "assistant" }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": 16, "completion_tokens": 140, "total_tokens": 156 } }
ããã£ãœãã®ãè¿ã£ãŠããŸããã
èªåã®ç°å¢ã ãšãè¿ã£ãŠãããŸã§ã«40ç§è¿ãããããŸãããã
llama_print_timings: load time = 3105.93 ms llama_print_timings: sample time = 83.56 ms / 141 runs ( 0.59 ms per token, 1687.35 tokens per second) llama_print_timings: prompt eval time = 3105.08 ms / 16 tokens ( 194.07 ms per token, 5.15 tokens per second) llama_print_timings: eval time = 38292.23 ms / 140 runs ( 273.52 ms per token, 3.66 tokens per second) llama_print_timings: total time = 42196.07 ms INFO: 127.0.0.1:59146 - "POST /v1/chat/completions HTTP/1.1" 200 OK
質åãå€ãããšé床ã倧ããå€ããã®ã§ããã®åçºã®é床ã¯åèã«ãªããŸãããã1åãè¶ ããããšããããŸãã
ä»å䜿ã£ãã¢ãã«ã ãšã䜿ãã®ã¯åºæ¬çã«è±èªã«ãªããŸããã
ãšãããããåããããšã¯ã§ããŸããã
é床ã«é£ã¯ãããŸãããæå ã®ç°å¢ã ãšã©ãã«ããªããªãã®ã§ã§ããç¯å²ã§äœ¿ãããšæããŸãã
llama-cpp-pythonã§äœ¿ããOpenAI APIäºæã®API
llama-cpp-pythonã§å®è£ ãããŠããAPIã¯ãOpenAI APIã®ãã®ãã¹ãŠã§ã¯ãããŸããã
ãµãŒããŒãèµ·åãããç¶æ
ã§http://localhost:8000/docs
ãèŠããšãOpenAPIå®çŸ©ã確èªã§ããã®ã§ããã¡ããåèã«ãããšããã§ãããã
ãµãŒããŒã¯ãFastAPIã§å®è£ ãããŠããã¿ããã§ããã
https://github.com/abetlen/llama-cpp-python/blob/v0.2.19/llama_cpp/server/app.py
0.2.19æç¹ã§å®è£ ãããŠããã®ã¯ã以äžã®ããã§ãã
- ENDPOINTS / Chat / Create chat completion
- ENDPOINTS / Embeddings / Create embeddings
- ENDPOINTS / Models / List models
- LEGACY / Completions / Create completion
ãšããã§ãã²ãšã€æ¯è²ãç°ãªããã®ãæ··ãã£ãŠããŸãã
/v1/engines/copilot-codex/completions
ããã¯ãGitHub Copilotã®REST APIã§ã¯ãªãã§ããããïŒ
GitHub Copilotã«é¢ããæ©èœãæã£ãŠããã®ã§ããããïŒãšæããããå®è£ ãèŠããšLEGACYã®Completionsãšåããã®ã®ããã§ãã
https://github.com/abetlen/llama-cpp-python/blob/v0.2.19/llama_cpp/server/app.py#L652-L660
çšèªãããããªã
ãšãŸããåããã ããªãïŒéãã§ããïŒç°¡åã§ããããåºãŠããåèªãå šç¶ããããŸããã
ã²ãšã€ã²ãšã€ãèŠãŠããããšæããŸãã
LLMïŒå€§èŠæš¡èšèªã¢ãã«ïŒ
LLMïŒå€§èŠæš¡èšèªã¢ãã«ïŒèªäœã«ã€ããŠã¯ãä»ã®ãµã€ãããâŠã
大規模言語モデルとは何ですか? - LLM AI の説明 - AWS
Llama
Llamaã¯ãMeta瀟ããªãŒãã³ãœãŒã¹ã§å ¬éããŠããLLMã§ãã
çŸåšã¯Llama 2ã¿ããã§ãã
ãã©ã¡ãŒã¿ãŒæ°ã¯7BïŒ70åïŒã13Bã70Bã®3çš®é¡ã§ããã
ã¢ãã«ã«ã¯Llama Chatãã³ãŒãçæåãã®Code Llamaãããããã§ãã
- Llama 2 - Resource Overview - Meta AI
- Introducing Code Llama, a state-of-the-art large language model for coding
llama.cpp
llama.cppã¯ãLlamaã¢ãã«ãéåžžã®PCã§å®è¡ã§ããããã«ããC/C++å®è£ ã§ãã
GitHub - ggerganov/llama.cpp: Port of Facebook's LLaMA model in C/C++
mac OSåãã«äœãããããã§ããLinuxãWindowsã§ãå®è¡ã§ããDockerã€ã¡ãŒãžãããããã§ãã
ç¹åŸŽã¯ä»¥äžã§ããã
- Plain C/C++ implementation without dependencies
- Apple silicon first-class citizen - optimized via ARM NEON, Accelerate and Metal frameworks
- AVX, AVX2 and AVX512 support for x86 architectures
- Mixed F16 / F32 precision
- 2-bit, 3-bit, 4-bit, 5-bit, 6-bit and 8-bit integer quantization support
- CUDA, Metal and OpenCL GPU backend support
ãµããŒããããŠããã¢ãã«ãå€ãããã§ãLlamaãLlama 2以å€ã«ãå€æ°äžŠãã§ããŸãã
ã€ã³ã¹ããŒã«æã«ã¯GPUã®æç¡ãªã©ã§ãã³ã³ãã€ã«ãªãã·ã§ã³ã調æŽããããã§ãã
å¿
èŠãªã¡ã¢ãªãšãã£ã¹ã¯ã¯ã䜿çšããã¢ãã«ã«äŸåããŸããä»ã¯ã¢ãã«ãã¡ã¢ãªã«ããŒããããããã¢ãã«ã®ãã£ã¹ã¯ãµã€ãºããã®ãŸãŸ
å¿
èŠãªã¡ã¢ãªéã«ãªãããã§ãã
llama.cpp / Memory/Disk Requirements
ç®å®ãšããŠã7Bã®ã¢ãã«ãªã3.9GBã13Bã®ã¢ãã«ãªã7.8GBã30Bã®ã¢ãã«ãªã19.5GBã65Bã®ã¢ãã«ãªã38.5GBãšããããšã«ãªããŸãã
ã¡ãªã¿ã«ãllama.cppèªäœããµãŒããŒã®æ©èœã¯æã£ãŠããããã§ãã
ã¢ãã«ïŒGGUFïŒ
LLMã䜿ããšããããšã§ãã¢ãã«ãå¿ èŠã«ãªããŸãã
llama.cppã§ã¯ãã¢ãã«ã®ååŸå ãšããŠä»¥äžã玹ä»ãããŠããŸãã
- llama.cpp / Using OpenLLaMA
- llama.cpp / Using GPT4All
- llama.cpp / Using Pygmalion 7B & Metharme 7B
- llama.cpp / Obtaining the Facebook LLaMA original model and Stanford Alpaca model data
- llama.cpp / Obtaining and using the Facebook LLaMA 2 model
æåŸã®2ã€ã¯ãMetaïŒFacebookïŒã®LlamaãæããŠããŸãããã€æåŸã®ãã®ã¯llama.cppã§æ±ãã圢åŒã«å€æããããã®ã玹ä»ããŠããŸãã
llama.cppã§ã¢ãã«ã䜿ãããã«ã¯ãGGUFãšãã圢åŒã«å€æããå¿
èŠããããŸããå€æçšã®ã¹ã¯ãªããã¯ãllama.cppãªããžããªãŒå
ã«
å«ãŸããŠããŸãã
äžã®ãªã¹ãã®Metaã®Llamaãllama.cppã§æ±ãã圢åŒã«å€æãããã®ããšããã®ã¯ããã®GGUFã®ããšã§ãã
ã€ãŸããæåã®åäœç¢ºèªã§äœ¿ã£ã以äžã®ã¢ãã«ã¯ããããããã®ã§ãã
- æå¿ãMetaãé åžããŠããLlamaã¢ãã«ãååŸããŠ
- GGUF圢åŒã«å€æããŠ
- Hugging FaceããããŠã³ããŒãã§ããããã«ãããã®
- ãã®ãšã³ããªãŒã®åäœç¢ºèªã§ã¯ã7B Chatã¢ãã«ã䜿çš
TheBloke/Llama-2-7B-Chat-GGUF · Hugging Face
åãã¢ãã«ã§ãã£ãŠããå質ã«éãããã£ããããã®ã§èŠãŠã¿ããšããã§ãããã
TheBloke / Llama-2-7B-Chat-GGUFProvided files
åäœç¢ºèªã§ã¯Q4_K_MïŒmedium, balanced quality - recommendedïŒã®ãã®ã䜿ã£ãŠããŸãã
13Bã70Bã®ã¢ãã«ããããŸãã
ããã§ããããåäœç¢ºèªã§è¡ã£ãæ å ±ãèªããããã«ãªã£ãæ°ãããŸãã
å®ã¯ãæåã®æ¹ã¯ãã®ã¢ãã«ã®éžæãééã£ãŠå šç¶å€ããŠããªããªã£ããããŠããŸããâŠã
llama-cpp-python
åæ²ã§ãããllama-cpp-pythonã¯llama.cppãPythonããæ±ããããã«ãããã€ã³ãã£ã³ã°ã§ãã
ãªãã±ïŒæ¥æ¬èªã¢ãã«ã䜿ããã
Metaã®Llama 2ã«æ¥æ¬èªã§è³ªåããŠãããããŠãè±èªã§è¿ã£ãŠããŸããè¿äºã«æ¥æ¬èªãå
¥ãããšããã£ãŠãã¡ãã£ãšé©ããŸããã
æ¥æ¬èªã§æžãã質åèªäœã¯ãç解ããŠããããã«ã¯æããŸãã
æ¥æ¬èªã®ã¢ãã«ã¯ãªãããªïŒãšèª¿ã¹ãŠã¿ããšãããLlama 2ãããŒã¹ã«ããæ¥æ¬èªã¢ãã«ããããŸããã
- Metaの「Llama 2」をベースとした商用利用可能な日本語LLM「ELYZA-japanese-Llama-2-7b」を公開しました|ELYZA, Inc.
- 「Code Llama」をベースとした商用利用可能な日本語LLM「ELYZA-japanese-CodeLlama-7b」を公開しました Hugging Faceã§ã¯ä»¥äžã§ããã
- elyza/ELYZA-japanese-Llama-2-7b · Hugging Face
- elyza/ELYZA-japanese-Llama-2-7b-instruct · Hugging Face
- elyza/ELYZA-japanese-Llama-2-7b-fast · Hugging Face
- elyza/ELYZA-japanese-Llama-2-7b-fast-instruct · Hugging Face
- elyza/ELYZA-japanese-CodeLlama-7b · Hugging Face
- elyza/ELYZA-japanese-CodeLlama-7b-instruct · Hugging Face
ãã©ã¡ãŒã¿ãŒæ°ã¯7Bã®ããã§ãã
ããã«ããã®ã¢ãã«ãGGUF圢åŒã«å€æããŠãããã®ããããŸããã
- mmnga/ELYZA-japanese-Llama-2-7b-gguf · Hugging Face
- mmnga/ELYZA-japanese-Llama-2-7b-instruct-gguf · Hugging Face
- mmnga/ELYZA-japanese-Llama-2-7b-fast-gguf · Hugging Face
- mmnga/ELYZA-japanese-Llama-2-7b-fast-instruct-gguf · Hugging Face
- mmnga/ELYZA-japanese-CodeLlama-7b-gguf · Hugging Face
- mmnga/ELYZA-japanese-CodeLlama-7b-instruct-gguf · Hugging Face
ãšãªããšãllama.cppã§äœ¿ããããšã«ãªãã®ã§ä»¥äžã®ã¢ãã«ã§è©ŠããŠã¿ãŸããã
$ curl -LO https://huggingface.co/mmnga/ELYZA-japanese-Llama-2-7b-fast-gguf/resolve/main/ELYZA-japanese-Llama-2-7b-fast-q4_K_M.gguf
llama-cpp-pythonã§èµ·åã
$ python3 -m llama_cpp.server --model ELYZA-japanese-Llama-2-7b-fast-q4_K_M.gguf
質åããŠã¿ãŸãã
$ curl -s -XPOST -H 'Content-Type: application/json' localhost:8000/v1/chat/completions -d \ '{"messages": [{"role": "user", "content": "èªå·±çŽ¹ä»ããŠãã ãã"}]}' | jq
ã¬ã¹ãã³ã¹ã
{ "id": "chatcmpl-8f0e9a33-92ac-444c-9e9a-329a0c32bebd", "object": "chat.completion", "created": 1700935609, "model": "ELYZA-japanese-Llama-2-7b-fast-q4_K_M.gguf", "choices": [ { "index": 0, "message": { "content": " ã¯ãããŸããŠãç§ã¯AIïŒäººå·¥ç¥èœïŒã§ãã\n人éã®èšèªãç解ããŠã人éã«ä»£ãã£ãŠæ§ã ãªããšãå®è¡ããããã°ã©ã ã§ãã\nãã®ããŒãžãã芧ã«ãªã£ãŠãã ãã£ãŠãããšããã ãšã¯ãããªããç§ã®äœè ã§ããç§ã«ã€ããŠç¥ãããã£ãã®ãããããŸãããïŒ ããã«ã¡ã¯ïŒ\nå æ¥ã3æ25æ¥(å)ã«ãåŒç€ŸããçãŸããã第äºåãã®ç·ã®åãèªçããŸããâªãååã¯ããå°æŸç¿ããïŒã ããããïŒãã§ãã\nç£å£°ãèãããŠãããŸããããŒ(^^) ååã¯ãæ°çå ã®ãšãã«åçãæ®ã£ãã®ã§ãã⊠ã»31æ¥ååã®ã¢ãžã¢äž»èŠåœã®å€åœçºæ¿åžå Žã§åçžå Žã¯äžæããŠããã\næ±äº¬åžå Žã§ã¯1ãã«=97 å60éè¿èŸºãã97å42éä»è¿ãŸã§äžãå¹ ãåºããã\næ¬§ç±³æ ªå®ãåŒãç¶ãéè·ãšãªãäžæ¹ã察ãŠãŒãã®é«å€åã§æšç§»ããŠããããäœãªã¹ã¯é貚ããšãããåãè²·ãæ»ãåããåºãŠããããã ã ãç§ã¯ã ã®ãŸãŸæ»ãã§ããŸãã®ã§ã¯ãªããâŠïŒããããªæãããäºãèããªãããããã®å€ããã€ãã®æ§ã«ç ãã«ã€ãã\nãããšäžæè°ãªäºã«ãç¿æèµ·ããæã«ã¯å·Šè ã®å·ãçããŠããïŒ\nããã§ãã£ãšåœŒå¥³ã¯å® å¿ããŠä»äºã«æã¡èŸŒãäºãåºæ¥ãã ããã ãããã§ããããšé ·ããåŸãããŸããã§ãåã¯æåãããããªæãã®äžç芳ãã€ã¡ãŒãžãããããããªãã§ããã©ãã\nãŸãããã£ã©ã¯ã¿ãŒãšãã·ãã¥ãšãŒã· ã§ã³ãèããŠâŠâŠã§ããã£ã±ãåãæãããã£ãã®ã¯ãçŸå°å¥³æŠå£«ã»ãŒã©ãŒã ãŒã³ããšããäœåã ã£ããšããããšã§ããã\nããã¡ããããã»ãŒã©ãŒã ãŒã³ãã¯ããäœåãªãã§ããã©ïŒç¬ïŒããã®åãåäœã®ã»ããä¹ ãã¶ãã«èªãã§ã¿ããããªãã ãä»èªã¿è¿ããŠã¿ããšç©è¶³ããªããªã£ãŠæã£ããã§ããã\nãã£ã©ã¯ã¿ãŒãå šç¶åããªããããªãã§ããïŒ\næŠéã·ãŒã³ãã»ãšãã©ãªããŠããããŸãåãããªããšãããâŠâŠã 2018幎1ææ«ã«ãªãŒãã³ããã°ããã®æ°åºã\nåºã ãšããåºå ã¯ãããŒãªã³ã°ä»æ§ã§ãæ€ åãšããŒãã«ãäžæã«ç§»åããããšçŽ60åãŸã§å容å¯èœã\nã«ã©ãªã±æ©çš®ã¯ãJOYSOUNDãã2å°èšçœ® ãããBGMãå å®ããŠããã 1945幎8æ6æ¥æªæãåºå³¶çååžã®å枯西æ¹ã®æµ·äžã«ãŠç±³è»ã®ååç匟ã«ãã被çãçºçãããšãããŠããã\nãã®æ¥ã®åºå³¶åžå ãžã®æäžã¯æªã 確èªãããŠããããåºå³¶åžã«èœäžãããã©ãããäžæã§ããããšãããåºå³¶çæããšããè¡šçŸãçšããããããšãããïŒé¢é£ïŒåºå³¶çååžã®ãååç匟æäžäºä»¶ãïŒã ãŸãããã®çç±ãšããŠããã®æ°åã«éããŸããããç§èªèº«ã¯ãèª åã«ã¯é¢ä¿ã®ãªãããšã ããšç¡<e9><96><a2>å¿ã«ãªã£ãŠããããšãå€ãã®ã§ã¯ãªãããšæããŸãã\nããããæ¥æ¬äººã®å€ãããããªã®ã§ã¯ãªãããšæã£ãŠããŸãïŒããã¯ç§ã®æã蟌ã¿ã®ç¯å²ã§ããïŒ\näŸãã°ããæ°å¹¹ç·ã®éè³ãå®ããããããšããåé¡ã«å¯Ÿããæ°å¹¹ç·ã䜿ã£ãŠç§»åããããšèªäœãèŽ æ²¢ã§ã¯ãªãã®ãïŒããªã©ãšãã£ãæèŠããä»ã®æ¥æ¬ã®ç€ŸäŒã¯ãèªå·±è²¬ä»»ã§è§£æ±ºãã¹ãã§ã¯ãªããïŒ\nãã®ããã«äœããã®çšéãæå ¥ããŠãç¡é§ãªããšã§ã¯ãããŸããïŒããªã©ã®ãããªæèŠããããšæããŸãããããããªã«äžäŸ¿ãªæ¥æ¬ç€ŸäŒã®ã·ã¹ãã ãæ§ç¯ããŠããã®ã¯èª°ã ïŒããšããæ¹å€ã»éé£ããŸã£ãããªãããšã¯ãããŸããã ãã®ããŒãžã§ã¯ã岩æçã§è©±é¡ã®å³æ¥èè³ãã£ãã·ã³ã°æ¥è ã玹ä»ããŠããŸãã\nãã¡ãã審æ»ããã®å Žã§ééããå¯èœæ§ãé«ããªããŸãã®ã§æ¯éåèã«ããŠäžããã\nãŸã岩æçã§ã®å£ã³ã æ å ±ãªã©ããŸãšããŠããã®ã§åãããŠã芧ãã ããã æ±äº¬ã¯å€äžã®2æãã4æããããŸã§èµ·ããŠãŸããïŒ\nãããŠæ5æã®é»è»ã«ä¹ã£ãŠåºå€ã§ãã(T_T)ããããã«ç²ããŸããããä¹ ãã¶ãã®æ¥æ¬ã§é£ã¹ ãã飯ãçŸå³ãããŠããã£ãã§ããâãã§ããåç£ãè²·ãæéããªãã»ã»ã»åž°åœçŽåã®å€é ããŸã§è²·ãç©ããŠãŸããã ã»ããªãªã³ãã¹ãOLYMPUS PEN mini E-PM1ãã³ã³ãã¯ãããžã«ã¡å£²ãçã©ã³ãã³ã°ã®9äœã«ã©ã³ã¯ã€ã³ïŒ\n軜éã§å°åãªãã©ãŒã¬ã¹äžçŒã«ã¡ã©ã¯ãæ è¡ãå€åºæã«äœ¿ããããã®ã§å¥³æ§ã«äººæ°ãé«ãéžã°ããåŸåããããŸãã ããã°ãã¯ïœæšå€ã¯ãŸãé ããªããŸããŠãã¿ãŸããã\nä»æ¥ãè¯ã倩æ°ã§ããïœãæ£æ©æ¥åã§ãïŒ\nã§ããïœç§ãšäºäººã®ç掻ã«ãªã£ãŠãããäœã ãå€ãªæããããã®ããïŒ\n(ç¬)æè¿ããããäžå¹ç«ã¡ãããæ¥ãŠãããã°ããã®ã«ã»ã»ã£ãŠæã£ããããŠãŸãâªãã®ãŸãŸ ã§ã¯ãç¬ã ãã«å¯ããæãããããŠããŸãããã§ã ãããã§çµããããªããŠæžãã€ããã¯ãããŸãããã\nã§ãä»ãŸã§ã®çµç·¯ãæžããšãããããããªããªâŠãšèªçããŠãããŸãã\nãããã°ããã¯ããã§ã®æŽæ°ã¯æ¢ãããããªããšæã£ãŠãŸãã®ã§ãèªè ããŸã«ã¯ã容赊ããã ãããã§ãã 12æ3æ¥(æ°Ž)ã倧éªã®æ±æŽå€§åŠã«ãŠéå¬ãããã第4åæ¥æ¬å¿çåŠäŒå€§äŒãã«æ¬åŠé¢ç8åãåå ããçºè¡šãè¡ããŸããïŒ\nãæè²çã¢ã»ã¹ã¡ã³ãããšããç 究åéã«ãããŠã¯ãäž»äœæ§ãå調æ§ãè²ãããã®æè²å 容ã»æ¹æ³ãªã©ã®æ€èšã«ã€ããŠç 究ãé²ããŠããŸããã\nä»åããã®ææãšããŠãæ±æŽå€§åŠã®å±±ç°å çãšãšãã«3åãååã§çºè¡šãè¡ããŸããïŒ ãŸãã12æã«å ¥ã£ãŠããã®é±æ«ã«ãããŠã¯ãç±³åœæ ªåžå Žã§ã®æ¥éš°ã«ããäžæçã«ãªã¹ã¯è³ç£ãžã®è³éæµå ¥ã掻çºåããŠããããšãèæ¯ã«ãå šè¬çã«ã¯åå®ã»ãã«é«æ¹åãžåããã\nãã ãã®åŸã«ã¯ãç±³é·æéå©ã®äžæãåããŠæ¥ç±³ã®æ ªåŒçžå Žãäžèœããããšã§ã察ãã«ã§ãå€ãæ¶ãå±éãšãªã£ãã ãã®äžã§ãäžå¹Žçã¯äžåŠåéšãèããæãã芪ãšãšãã«èããªããã°ãªã ãªãããšããšèšãã®ããããŸããã\n確ãã«ç§éã®äžä»£ãŸã§ã¯ããã»ã©çå£ã«èããŠããªãã£ãã®ã§ãããä»ã®äžä»£ã¯å¡Ÿã«è¡ãããããšèªäœãåœããåã§ããããé£ã³è¶ããŠäžåŠåéšããåã©ããå€ãã®ã ããã§ãã 1975幎çãŸãã\nåæµ·é倧åŠèŸ²åŠéšåæ¥åŸãæ ªåŒäŒç€Ÿãšã€ãã»ã¢ã€ã»ãšã¹ïŒHISïŒã«å°±è·ããåœå å€ã®ããã«äºçŽæ¥åã«åŸäºããåŸã2006幎ã«ç¬ç«ã\næ è¡äŒç€ŸãšããŠãæµ·å€æ è¡ããçµå©åŒãã軞ãšããäŒç»ã»å¶æ¥æŽ»åãè¡ãªã£ãŠããã ä»æ¥ã¯å€æ¹ããéšãéãåºããŠæ¥ãŠãããäžæã¯åŒ·é¢šã§å€§ããªæšã ã¯æºãåããŠããã\nå€ã«ãªã£ãŠãæ¢ãæ°é ããªãã®ã§ãã®ãŸãŸææ¥ãç¶ãã®ããšæ㣠ããæŒéãã«ã¯éšã¯äžãã£ãŠæ空ãèŠãããšãåºæ¥ãã\næ霢ã調ã¹ããšæºæã®ïŒïŒæ¥ååŸã«ãªããä»æ¥ã¯ïŒïŒïŒãšè¥å¹²å°ããã\næšæ¥ã¯ãææ¹ããéšã§æãèŠããªãäžã§æ®åœ±ããã®ã§äžæã ãã¢ããããã ã10幎åã«ãã£ãããšãšåãããšããã£ãŠããã ã£ãŠæãã¡ããããšãã声ãèãããå®ã¯ãã®æ²ã¯åäœã®ãã¢ãŒãTHE FIRST TOUR -TOMORROW-ãã§ãæŒå¥ããã楜æ²ã§ããã\nåœæããã¡ã³ããŒã®æèŠã¯äžèŽããŠãããã絶察ã«ãããããšæã£ãŠããããšã®ããšã§åã³ã¹ããŒãžã§é³Žããããšã«æåããã®ã ãšããã ããã®äžã¯ãçè«ãšå®è·µãå¥ã ã«ãªã£ãŠããããã«èŠããŸãããå®ã¯äžã€ãªãã§ã ã\nããšãã°ãéå ·ã補äœãããšããããã§ãã\nãã®ã¥ããã«ã¯ããŸãäœãå¿ èŠããèããŸãããïŒ\nãããŠå¿ èŠãªãã®ããããã°ã次ã«ã©ããã£ãããããæã«å ¥ããããšãã§ããã®ãèããããã§ãã ÑевÑÐ°Ð»Ñ 2018 â 31æ¥ã¯ã¯ãªã¹ãã¹ãçµãã£ãŠæ£æåã®æºåããããã ã幎çã®ãå°é£ããããã£ããããŠæ¥œããã ãããŸãã\nãããã1æã¯å¯ããå³ããå¬ã®ææã§ãç¹ã«ãã®æ¥ãæäœæ°æž©ã¯ãã€ãã¹9床ãŸã§äžãããŸãã A.ãã¯ãã\nã客ããŸã«ã¯å€§å€ãè¿·æããããããŸããããã¹ã¿ããäžäžžãšãªã£ãŠå¯Ÿå¿ãããŠããã ããŸããã\nä»ãŸã§ã®çµéšã掻ãããå šåã§ã客æ§ã®ç«å Žã«ç«ã¡ããå®å¿ã»å®å šããªãµãŒãã¹ããæäŸã§ããããå¿ãããŠããŸãã®ã§ããããããã©ããå®ãããé¡ãèŽããŸãã ãèªåã®åã©ããäžç»æ ¡ã«ãªããããããªãããããã¯åŠæ ¡ã«è¡ããªããªã£ããã©ããããããš ãã芪埡ãããå€ãããã£ããããŸãã\nãã®å Žåã«ã¯ãŸãçžè«çªå£ã調ã¹ãŠã¿ãããšããå§ãããŸãã\nãŸãã¯å°é家ãçµéšã®è±å¯ãªã«ãŠã³ã»ã©ãŒãšè©±ãããŠã¿ãŠ", "role": "assistant" }, "finish_reason": "length" } ], "usage": { "prompt_tokens": 16, "completion_tokens": 2032, "total_tokens": 2048 } }
埮åŠãªè¿äºãâŠã
llama_print_timings: load time = 2185.05 ms llama_print_timings: sample time = 1641.06 ms / 2032 runs ( 0.81 ms per token, 1238.22 tokens per second) llama_print_timings: prompt eval time = 2184.98 ms / 16 tokens ( 136.56 ms per token, 7.32 tokens per second) llama_print_timings: eval time = 666306.88 ms / 2031 runs ( 328.07 ms per token, 3.05 tokens per second) llama_print_timings: total time = 683503.91 ms
ããšãæ»ã£ãŠãããŸã§ã«10åãããããããŸããã
çãåçã«ãªãããã«ããŠã¯ã©ãã ãããšè©ŠããŠã¿ãã®ã§ãã
$ curl -s -XPOST -H 'Content-Type: application/json' localhost:8000/v1/chat/completions -d \ '{"messages": [{"role": "user", "content": "100æå以å ã§æšæ¶ããŠãã ãã"}]}' | jq
ããŸãå€ãããã
{ "id": "chatcmpl-944575e4-6db7-4c6d-91bb-f08411ed0972", "object": "chat.completion", "created": 1700934635, "model": "ELYZA-japanese-Llama-2-7b-fast-q4_K_M.gguf", "choices": [ { "index": 0, "message": { "content": " ããã«ã¡ã¯ïŒ\nåããŸããŠãã¢ã·ã¹ã¿ã³ãã®Kã§ãã\nãããããé¡ãèŽããŸãã\nç§ã¯å€§åŠçã§ããããã瀟äŒäººã«ãªãããä»ã®ãã¡ã«å°ãã§ã瀟äŒäººã®çµéšãç©ã¿ãããšæã就掻 äžã§ãã ã»å¹³æ28幎床ããå ¥åæãããããŸãã\nïŒäžéšå¹Œçšåã«ãã£ãŠç°ãªããŸãïŒè©³çŽ°ã¯äžèšãªã³ã¯å ããåç §ãã ããã\nhttp://www.pref.mie.lg.jp/soshiki/koseki_kyoiku/17025243500.html ããŠãïŒæïŒïŒæ¥ã«å ¬éããããæ¥æ¬ãäžçã«èªãã¢ããã®ç¹éã¯ããæ¥æ¬ã®ããã®ã¥ããããæ¯ããé©åœçãªæè¡ãšãã®å é²çåã ããšããããŒãã§ãïŒïŒäžçŽã代衚ããçºæã»æ°è£œåã»å·¥æ³ãªã© ïŒïŒåãéžã³ãŸããã\nããã¯ïŒïŒïŒïŒå¹ŽïŒæå·ã«ãæ¥æ¬ãäžçã«èªãã¢ããã®ç¹éãšããŠåºçãããã®ã§ããã®æãïŒïŒïŒåã»ã©éžãã ã®ã§ãããä»åºŠã¯ãããããã«åã«ããŠå šéšã§ïŒïŒïŒçš®ãæããŠããŸãïŒâ»ïŒã æè¿ã«ãªã£ãŠããããããæ£æäŒã¿ã ãšããããšãããã£ãŠããã\nãã®éã¯æ¯æ¥ä»äºã ã£ãã®ã ã\nãæ¥æ¬äººãã®äŒæ¥æèŠãããããã£ãã\næ¥æ¬äººã®äŒæ¥ã¯å¹³çã«ïŒïŒæ¥ã«éçŽãããŠããããã«èŠããã\nãã®ãŸãŸãããšããïŒïŒæ¥ããšããæžããªãã®ã§ã¯ãªããïŒ 1945幎ãæ±äº¬çãŸãã\næ±å€§æé€åŠéšåã\nç±³åœå€§åŠé¢ãçµãŠãåœéåºç£æ倧åŠåæ¥åŸãæ¥æ¬IBMå ¥ç€Ÿã\nå¶æ¥è·ããžãŠã³ã³ãã¥ãŒã¿ç ä¿®éçºå®€é·ã«å°±ä»»ã\nãã®åŸã人æãããžã¡ã³ãã®éèŠæ§ã«çç®ãã1986幎ããç¬ç«ããŠã³ã³ãµã«ã¿ã³ããšããŠæŽ»åãéå§ã ä»æ¥ã®ååäžã¯ãä¹ ã ã®éšæš¡æ§ã§æ°æž©ãé«ããªãéãããã ã㣠ãã®ã§ãå ¬åæ£æ©ã楜ãã¿ãŸããã\nïŒã§ããåæ¥ãšã¯éã人混ã¿ã ã£ãããïŒ ä»æ¥ã¯çãããèªè»¢è»ã«ä¹ããã«æ©ããŠã¿ãŸããã\nå°ãåã«è²·ã£ããé·éŽãããªãå¿«é©ã ã£ãã®ã§ããã®ãŸãŸå¬ãŸã§å±¥ ããããªãïŒ å€§äººæ°ã®ãã¹ã«ã©ãšã¢ã€ã·ã£ããŠã¯æãããïŒ\n2019幎ç§ãæ°äœã¢ã€ã©ã€ããŒãç¶ã ãšçºå£²ãããŸãâä»åã玹ä»ããŸãã®ã¯ãã»ã¶ã³ãããçºå£²ããããã¢ãã«ç®ãã\nããŸã€æ¯ãé· ããªã£ãŠããããç®ãã«ã©ãããããšãã£ãããŸãã§èžèœäººã®ãããªâæ§ãã®ç®âãå¶ããŠãããã¢ã€ãã ãªãã§ãïŒ 1924幎10æ23æ¥ã«ããããåµæ¥ããããŸã§ããããã¯æååæã«æ ããŠããäžååžè¥¿ åºå±±æ å°åºã«ãã£ãè£œç³žå·¥å Žã®è·¡å°ãå©çšããŠå·¥å Žãšç€Ÿå® ã建ãŠãŸããã\nãã®å°ããŸãå€ãã®æŽå²ããããåœæã¯äŒè€åæã®ç§éžããã£ããªã©æ°ã ã®ãšããœãŒããæ®ããŠããŸãã\nåµæ¥ããä»æ¥ã«è³ ããŸã§ã®ãããã¯ããã®éœåºŠæ代ã®å€åã«åãããŠæè¡é©æ°ãè¡ããªããæ©ãã§ããŸããã\nç¹ã«2018幎ã«ã¯ç¬¬43åæ±äº¬ã¢ãŒã¿ãŒã·ã§ãŒã«ãããŠããVision Coupe ConceptïŒãŽã£ãžã§ã³ã³ãŠãã³ã³ã»ã ãïŒãã çºè¡šããŸããã ãã®ã¬ããŒãã§ã¯ãã€ã³ããã·ã¢ã®é»ååååŒåžå Žã«é¢ããåžå Žèª¿æ»ãå®æœããŸãã\nããªã³ã©ã€ã³ã»ãªãã©ã€ã³ããBtoBã»CtoCããªã©ã®åºåã«åºã¥ããŠãååžå Žã®ç¹åŸŽãååããŸãä» åŸã®æé·çã«ã€ããŠãèšåããŠããŸãã\nAlibaba Group ã¯ãã¢ãªããã°ã«ãŒãã®äž»èŠãªãã©ã³ãã§ããã\nå瀟ã¯ã€ã³ã¿ãŒãããäžã§ã®ã·ã§ããã³ã°ã¢ãŒã«ããã³ããžãã¹ãœãªã¥ãŒã·ã§ã³ã æäŸããã\n2014幎æç¹ã§ã幎é売äžã3,569å人æ°å ïŒçŽ6.0å åïŒã\nAlibaba Group ã¯ãã¢ãªããã°ã«ãŒãã®äž»èŠãªãã©ã³ãã§ããã\nå瀟ã¯ã€ã³ã¿ãŒãããäžã§ã®ã·ã§ããã³ã°ã¢ãŒã«ããã³ããžãã¹ãœãªã¥ãŒã·ã§ã³ãæäŸããã\n2014幎æç¹ã§ã幎é売äžã3,569å人æ°å ïŒçŽ6.0å åïŒ ããã«ã¡ã¯ïŒ\næè¿ã§ã¯ãä»äºããã©ã€ããŒãã§ã䜿ãããããšãå€ããªã£ãLINEã§ããã å®ã¯ãã¹ãã ã¡ãŒã«ããå±ãããšãããã®ã§ãã\nãããªããšãç¥ããªãæ¹ã¯ãæå€ãšå€ããšæããŸãã\nããã§ä»åã¯ããªããã®ãããªäºãçºçããã®ããªã©ã«ã€ããŠçŽ¹ä»ããŠãããŸãããã\nãŸã察çæ¹æ³ãªã©ãåãããŠãäŒãããŠãããŸãã®ã§ãã²æåŸãŸã§ã芧äžããïŒ\n[âŠ] äžåŠæ ¡æè«ã»æå°äž»äºïŒçæ°ç§ïŒãçµãŠçŸåšã¯çŠå²¡çå ã®å ¬ç«äžé«äžè²«æ ¡ã§æè²å®è·µã®ãµããŒããè¡ãåãã倧åŠã«ãŠåŠã³çŽãäžã§ãã©ãããŠãããªã£ãã®ããããªããããªçµæã«ãªã£ãã®ããã®çç±ãæ¢æ±ããæ¢ç©¶çãªåŠç¿æ³ãå®è·µçã«æå°ããæèåãåµé æ§ãšãã£ãèœåãè²ãŠãææ¥ãå±éããŠããŸãã å°ããæãããã£ãšéçãç¶ã㊠ããã®ã§ã ããã©ãããŠãæéŠã«è² æ ãããã£ãŠããŸãããŸãè©ãããé ·ãããŠãŸããã\nã§ããã®æ²»çé¢ã§æ²»çããŠããã£ãããã¹ãããªããçã¿ããªããªã£ãŠãä»ã®ãšããã¯ãµãã«ãŒã楜ãããŠã ãã®ã§ããšãŠããããããã§ãïŒ 2014幎8æã«ã¯å€§æã¬ã·ããµã€ããã¯ãã¯ããããããé£ã¹ãã°ããè²·åããããšã§è©±é¡ãéããã\nãã®åŸ9æã«ãå瀟ã¯è²·åé¡ã«åœåäºå®ããŠãã50ååãè¶ ããçŽ60ååã§ååŒ ãå®è¡ãããã®ããšã«ã€ããŠæ¹å€çãªæèŠãå€æ°å¯ããããã ãNQNéŠæž¯=é·å°Ÿä¹ è¯ã24æ¥ã®éŠæž¯æ ªåŒçžå Žã¯å°å¹ ãªåããšãªã£ãŠããã\nãã³ã»ã³ææ°ã¯åæ¥çµå€ãšæ¯ã¹ãŠãããã«å°å¹ ã«äž ããŠå§ãŸãã ãã¡ã§æšç§»ããŠããã\nåå°äœè£œé è£ çœ®ã®å€§æãè±ã¢ãŒã ã»ããŒã«ãã£ã³ã°ã¹ã®10%è¶ é«ãçžå Žã®éè·ãšãªããéŠæž¯åžå Žã§ã¯å©ç確å®ç®çã®å£²ããç¶ããŠããã\nãã£ãšãããã³ã»ã³ææ°ãæ§æãã50é æã®ãã¡28éæãäžæããŠããã»ããäžèœã¯6éæã«ãšã©ãŸããªã©å€åãã®è»œãéæãäžå¿ã«è²·ãããŠããããšããããçžå Žå šäœã®åºå ãã¯ä¿ãããŠããã\néŠæž¯ã¡ãŒã³ããŒã(æ±èšŒ1éšã«çžåœ)ã®å£²è²·ä»£éã¯çŽ235åéŠæž¯ãã«ãšäœèª¿ã§ãææ°ãæ§æããéæã§ã¯ãåå°äœåèšçç£äŒç€Ÿã®è¯è¯é»åãäžèœããŠæšç§»ããŠããã [ããªãŒçµµç»] ãŠã£ãªã¢ã ã»ã¢ã³ããªã¥ãŒã¹(1860~1947)ããŽã£ããªã¢ã® éã(1882-85é ).JPG[92330704_p0_s10_c9x2.jpg] ïœ ç¡æç»åçŽ æéTokyolens - ããªãŒç»åãããŠã³ããŒãããåçšãOKã§ã nobodyãããæžãããŠããããã«ããã®æ¬ã¯ãæ¥æ¬åœæ²æ³ã®æç«éçšãã«ã€ããŠã®åºæ¬è³æãšããŠèªãã®ãè¯ããšæããŸãã\nç¹ã«æ²æ³å¶å®ã®äž»å°æš©äºãã«ã€ããŠã¯èå³æ·±ãã§ãã\nç§ã¯ã1940幎代ã®è»éšã®å¢åã匷倧ãªã®ã¯åœç¶ã ãšæã£ãŠããŸããããããããŸã§æ°äž»çãªæ¹æ£ããªãããããšãç¥ã£ãŠé©ããŸããã (13æ50åãã³ãŒã8601)åå Žã¯åèœãããã®ã®åŸå Žã«äžãå¹ ãåºããã\näžæã¯åæ¥æ¯7å(2.2%)é«ã®329åãŸã§äžæããé£æ¥ã§å¹Žåæ¥é«å€ãæŽæ°ããã\nååŸã« å ¥ããæ¥çµå¹³å ãäžãæžã£ãããšããã£ããã«è²·ããå ¥ã£ãŠããããã ã\nåžå Žã§ã¯ãäžåœãªã©ã®æ°èåœçµæžã®å è¡ãäžéææãæ ¹åŒ·ããã®ã®ãåœå ã®æ¯æ°å埩æåŸ ã¯æ ¹åŒ·ãæ ªäŸ¡ãžã®è¿œã颚ã«ãªã£ãŠã ãã(岩äºã³ã¹ã¢ 蚌åž)ãšã®å£°ããã£ãã̶ 12æ3æ¥ã«äžå Žæ¥é«å€ãæŽæ°ããäžè±UFJ(8306)ãããããã®åŸå Žã«äžãå¹ ãåºãã第äžçåœ(8750)ãªã©äž»åæ ªãå 調ãªããšãè¿œã颚ã«ãªã£ãŠããããã ã\nãæ¥çµQUICKãã¥ãŒã¹ ã Hinweis: Die Verwendung von Cookies auf Ihrem Computer ist in Ihrem Browser-Browser deaktiviert. Es wird empfohlen, das Cookie-Javascript-Szenario anzusehen ÑÑÑÐœÑ 2014 ÑÐŸÐºÑ ÐŒÑÐœÑÑÑеÑÑÑвП кÑлÑÑÑÑО, ÑÑÑÐžÐ·ÐŒÑ Ñа ÑпПÑÑÑ Ð·Ð°ÑвеÑ", "role": "assistant" }, "finish_reason": "length" } ], "usage": { "prompt_tokens": 22, "completion_tokens": 2026, "total_tokens": 2048 } }
æéã®åŸåãå€ããããã§ããã
llama_print_timings: load time = 3746.86 ms llama_print_timings: sample time = 2042.77 ms / 2026 runs ( 1.01 ms per token, 991.79 tokens per second) llama_print_timings: prompt eval time = 3746.78 ms / 22 tokens ( 170.31 ms per token, 5.87 tokens per second) llama_print_timings: eval time = 781484.80 ms / 2025 runs ( 385.92 ms per token, 2.59 tokens per second) llama_print_timings: total time = 801202.56 ms INFO: 127.0.0.1:59150 - "POST /v1/chat/completions HTTP/1.1" 200 OK
ãšãããããåããåã«ã¯Metaã®Llama 2ãè¯ãããã ãªãšæããŸããã
ãããã«
llama-cpp-pythonã§ãOpenAI APIäºæã®ãµãŒããŒãåãããŠã¿ãŸããã
ãšãã£ãŠããçšèªã®ç¢ºèªãæŽçãäž»ã§ã¯ãã£ãã®ã§ããã
ãªããšãªãæ å ±ã¯æŽçã§ããŠããæ°ãããã®ã§ãå°ããã€ãã®ããããæãåºããŠããããã§ããã