diff --git a/docs/develop/rust/wasinn/llm_inference.md b/docs/develop/rust/wasinn/llm_inference.md index a7db3db1..885a3d3a 100644 --- a/docs/develop/rust/wasinn/llm_inference.md +++ b/docs/develop/rust/wasinn/llm_inference.md @@ -2,44 +2,11 @@ sidebar_position: 1 --- -# Llama 2 inference - -WasmEdge now supports running open source models in Rust. We will use [this example project](https://github.com/second-state/LlamaEdge/tree/main/chat) to show how to make AI inferences with the llama2 model in WasmEdge and Rust. - -WasmEdge now supports the following models: - -1. Llama-2-7B-Chat -1. Llama-2-13B-Chat -1. CodeLlama-13B-Instruct -1. Mistral-7B-Instruct-v0.1 -1. Mistral-7B-Instruct-v0.2 -1. MistralLite-7B -1. OpenChat-3.5-0106 -1. OpenChat-3.5-1210 -1. OpenChat-3.5 -1. Wizard-Vicuna-13B-Uncensored-GGUF -1. TinyLlama-1.1B-Chat-v1.0 -1. Baichuan2-13B-Chat -1. OpenHermes-2.5-Mistral-7B -1. Dolphin-2.2-Yi-34B -1. Dolphin-2.6-Mistral-7B -1. Samantha-1.2-Mistral-7B -1. Samantha-1.11-CodeLlama-34B -1. WizardCoder-Python-7B-V1.0 -1. Zephyr-7B-Alpha -1. WizardLM-13B-V1.0-Uncensored -1. Orca-2-13B -1. Neural-Chat-7B-v3-1 -1. Yi-34B-Chat -1. Starling-LM-7B-alpha -1. DeepSeek-Coder-6.7B -1. DeepSeek-LLM-7B-Chat -1. SOLAR-10.7B-Instruct-v1.0 -1. Mixtral-8x7B-Instruct-v0.1 -1. Nous-Hermes-2-Mixtral-8x7B-DPO -1. Nous-Hermes-2-Mixtral-8x7B-SFT - -And more, please check [the supported models](https://github.com/second-state/LlamaEdge/blob/main/models.md) for details. +# LLM inference + +WasmEdge now supports running open-source Large Language Models (LLMs) in Rust. We will use [this example project](https://github.com/second-state/LlamaEdge/tree/main/chat) to show how to make AI inferences with the llama-3.1-8B model in WasmEdge and Rust. + +Basically, WasmEdge can support any open-source LLMs. Please check [the supported models](https://github.com/second-state/LlamaEdge/blob/main/models.md) for details. ## Prerequisite @@ -55,23 +22,23 @@ First, get the latest llama-chat wasm application curl -LO https://github.com/LlamaEdge/LlamaEdge/releases/latest/download/llama-chat.wasm ``` -Next, let's get the model. In this example, we are going to use the llama2 7b chat model in GGUF format. You can also use other kinds of llama2 models, check out [here](https://github.com/second-state/llamaedge/blob/main/chat/README.md#get-model). +Next, let's get the model. In this example, we are going to use the llama-3.1-8B model in GGUF format. You can also use other kinds of LLMs, check out [here](https://github.com/second-state/llamaedge/blob/main/chat/README.md#get-model). ```bash -curl -LO https://huggingface.co/wasmedge/llama2/resolve/main/llama-2-7b-chat-q5_k_m.gguf +curl -LO https://huggingface.co/second-state/Meta-Llama-3.1-8B-Instruct-GGUF/resolve/main/Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf ``` Run the inference application in WasmEdge. ```bash -wasmedge --dir .:. --nn-preload default:GGML:AUTO:llama-2-7b-chat-q5_k_m.gguf llama-chat.wasm +wasmedge --dir .:. --nn-preload default:GGML:AUTO:Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf llama-chat.wasm -p llama-a-chat ``` After executing the command, you may need to wait a moment for the input prompt to appear. You can enter your question once you see the `[USER]:` prompt: ```bash [USER]: -I have two apples, each costing 5 dollars. What is the total cost of these apple +I have two apples, each costing 5 dollars. What is the total cost of these apples? [ASSISTANT]: The total cost of the two apples is 10 dollars. [USER]: @@ -95,19 +62,26 @@ Second, use `cargo` to build the example project. cargo build --target wasm32-wasi --release ``` -The output WASM file is `target/wasm32-wasi/release/llama-chat.wasm`. Next, use WasmEdge to load the llama-2-7b model and then ask the model to questions. +The output WASM file is `target/wasm32-wasi/release/llama-chat.wasm`. Next, use WasmEdge to load the llama-3.1-8b model and then ask the model questions. ```bash -wasmedge --dir .:. --nn-preload default:GGML:AUTO:llama-2-7b-chat-q5_k_m.gguf llama-chat.wasm +wasmedge --dir .:. --nn-preload default:GGML:AUTO:Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf llama-chat.wasm -p llama-3-chat ``` -After executing the command, you may need to wait a moment for the input prompt to appear. You can enter your question once you see the `[USER]:` prompt: +After executing the command, you may need to wait a moment for the input prompt to appear. You can enter your question once you see the `[You]:` prompt: ```bash -[USER]: -Who is Robert Oppenheimer? -[ASSISTANT]: -Robert Oppenheimer was an American theoretical physicist and director of the Manhattan Project, which developed the atomic bomb during World War II. He is widely regarded as one of the most important physicists of the 20th century and is known for his contributions to the development of quantum mechanics and the theory of the atomic nucleus. Oppenheimer was also a prominent figure in the post-war nuclear weapons debate and was a strong advocate for international cooperation on nuclear weapons control. +[You]: +Which one is greater? 9.11 or 9.8? + +[Bot]: +9.11 is greater. + +[You]: +why + +[Bot]: +11 is greater than 8. ``` ## Options @@ -118,13 +92,13 @@ You can configure the chat inference application through CLI options. -m, --model-alias Model alias [default: default] -c, --ctx-size - Size of the prompt context [default: 4096] + Size of the prompt context [default: 512] -n, --n-predict Number of tokens to predict [default: 1024] -g, --n-gpu-layers Number of layers to run on the GPU [default: 100] -b, --batch-size - Batch size for prompt processing [default: 4096] + Batch size for prompt processing [default: 512] -r, --reverse-prompt Halt generation at PROMPT, return control. -s, --system-prompt @@ -148,8 +122,8 @@ The `--prompt-template` option is perhaps the most interesting. It allows the ap Furthermore, the following command tells WasmEdge to print out logs and statistics of the model at runtime. ```bash -wasmedge --dir .:. --nn-preload default:GGML:AUTO:llama-2-7b-chat-q5_k_m.gguf \ - llama-chat.wasm --prompt-template llama-2-chat --log-stat +wasmedge --dir .:. --nn-preload default:GGML:AUTO:Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf \ + llama-chat.wasm --prompt-template llama-3-chat --log-stat .................................................................................................. llama_new_context_with_model: n_ctx = 512 llama_new_context_with_model: freq_base = 10000.0 @@ -173,7 +147,7 @@ You can make the inference program run faster by AOT compiling the wasm file fir ```bash wasmedge compile llama-chat.wasm llama-chat.wasm -wasmedge --dir .:. --nn-preload default:GGML:AUTO:llama-2-7b-chat-q5_k_m.gguf llama-chat.wasm +wasmedge --dir .:. --nn-preload default:GGML:AUTO:Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf llama-chat.wasm ``` ## Understand the code @@ -185,7 +159,7 @@ The [main.rs](https://github.com/second-state/llamaedge/blob/main/chat/src/main. curl -LO https://github.com/second-state/llamaedge/releases/latest/download/llama-simple.wasm # Give it a prompt and ask it to use the model to complete it. -wasmedge --dir .:. --nn-preload default:GGML:AUTO:llama-2-7b-chat-q5_k_m.gguf llama-simple.wasm \ +wasmedge --dir .:. --nn-preload default:GGML:AUTO:Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf llama-simple.wasm \ --prompt 'Robert Oppenheimer most important achievement is ' --ctx-size 512 output: in 1942, when he led the team that developed the first atomic bomb, which was dropped on Hiroshima, Japan in 1945. @@ -275,7 +249,7 @@ Next, execute the model inference. context.compute().expect("Failed to complete inference"); ``` -After the inference is finished, extract the result from the computation context and lose invalid UTF8 sequences handled by converting the output to a string using `String::from_utf8_lossy`. +After the inference is finished, extract the result from the computation context and losing invalid UTF8 sequences handled by converting the output to a string using `String::from_utf8_lossy`. ```rust let mut output_buffer = vec![0u8; *CTX_SIZE.get().unwrap()]; @@ -296,5 +270,5 @@ println!("\noutput: {}", output); ## Resources * If you're looking for multi-turn conversations with llama 2 models, please check out the above mentioned chat example source code [here](https://github.com/second-state/llamaedge/tree/main/chat). -* If you want to construct OpenAI-compatible APIs specifically for any open-source LLMs, please check out the source code [for the API server](https://github.com/second-state/llamaedge/tree/main/api-server). +* If you want to construct OpenAI-compatible APIs specifically for your llama2 model, or the Llama2 model itself, please check out the source code [for the API server](https://github.com/second-state/llamaedge/tree/main/api-server). * To learn more, please check out [this article](https://medium.com/stackademic/fast-and-portable-llama2-inference-on-the-heterogeneous-edge-a62508e82359).