Merge dev branch (#129)

* [Example] ggml: add grammar example (#126) * [Example] ggml: add grammar example Signed-off-by: dm4 <[email protected]> * [CI] add grammar test Signed-off-by: dm4 <[email protected]> --------- Signed-off-by: dm4 <[email protected]> * [Example] ggml: update command-r example to use command-r-plus with tool use prompt Signed-off-by: dm4 <[email protected]> * [CI] llama: add llama2 embedding test Signed-off-by: dm4 <[email protected]> * [CI] llama: add llama-stream test Signed-off-by: dm4 <[email protected]> --------- Signed-off-by: dm4 <[email protected]>
second-state · Apr 16, 2024 · 6e81ae3 · 6e81ae3
1 parent 653daf0
commit 6e81ae3
Show file tree

Hide file tree

Showing 10 changed files with 430 additions and 19 deletions.
diff --git a/.github/workflows/llama.yml b/.github/workflows/llama.yml
@@ -101,6 +101,19 @@ jobs:
                 default \
                 $'[INST] <<SYS>>\nYou are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe.  Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you do not know the answer to a question, please do not share false information.\n<</SYS>>\nWhat is the capital of Japan?[/INST]'
 
+          - name: Llama2 7B (Streaming)
+            run: |
+              test -f ~/.wasmedge/env && source ~/.wasmedge/env
+              cd wasmedge-ggml/llama-stream
+              curl -LO https://huggingface.co/TheBloke/Llama-2-7b-Chat-GGUF/resolve/main/llama-2-7b-chat.Q5_K_M.gguf
+              cargo build --target wasm32-wasi --release
+              time wasmedge --dir .:. \
+                --env n_gpu_layers="$NGL" \
+                --nn-preload default:GGML:AUTO:llama-2-7b-chat.Q5_K_M.gguf \
+                target/wasm32-wasi/release/wasmedge-ggml-llama-stream.wasm \
+                default \
+                $'[INST] <<SYS>>\nYou are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe.  Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you do not know the answer to a question, please do not share false information.\n<</SYS>>\nWhat is the capital of Japan?[/INST]'
+
           - name: StarCoder 2 7B
             run: |
               test -f ~/.wasmedge/env && source ~/.wasmedge/env
@@ -133,7 +146,7 @@ jobs:
                 target/wasm32-wasi/release/wasmedge-ggml-multimodel.wasm \
                 'describe this picture please'
 
-          - name: Embedding Example
+          - name: Embedding Example (All-MiniLM)
             run: |
               test -f ~/.wasmedge/env && source ~/.wasmedge/env
               cd wasmedge-ggml/embedding
@@ -145,6 +158,18 @@ jobs:
                 default \
                 'hello world'
 
+          - name: Embedding Example (Llama-2)
+            run: |
+              test -f ~/.wasmedge/env && source ~/.wasmedge/env
+              cd wasmedge-ggml/embedding
+              curl -LO https://huggingface.co/TheBloke/Llama-2-7b-Chat-GGUF/resolve/main/llama-2-7b-chat.Q5_K_M.gguf
+              cargo build --target wasm32-wasi --release
+              time wasmedge --dir .:. \
+                --nn-preload default:GGML:AUTO:llama-2-7b-chat.Q5_K_M.gguf \
+                target/wasm32-wasi/release/wasmedge-ggml-llama-embedding.wasm \
+                default \
+                'hello world'
+
           - name: RPC Example
             run: |
               test -f ~/.wasmedge/env && source ~/.wasmedge/env
@@ -171,6 +196,19 @@ jobs:
                 default \
                 '<start_of_turn>user Where is the capital of Japan? <end_of_turn><start_of_turn>model'
 
+          - name: Grammar Example
+            run: |
+              test -f ~/.wasmedge/env && source ~/.wasmedge/env
+              cd wasmedge-ggml/grammar
+              curl -LO https://huggingface.co/TheBloke/Llama-2-7b-GGUF/resolve/main/llama-2-7b.Q5_K_M.gguf
+              cargo build --target wasm32-wasi --release
+              time wasmedge --dir .:. \
+                --env n_gpu_layers="$NGL" \
+                --nn-preload default:GGML:AUTO:llama-2-7b.Q5_K_M.gguf \
+                target/wasm32-wasi/release/wasmedge-ggml-grammar.wasm \
+                default \
+                'JSON object with 5 country names as keys and their capitals as values: '
+
           - name: Build llama-stream
             run: |
               cd wasmedge-ggml/llama-stream

diff --git a/wasmedge-ggml/command-r/README.md b/wasmedge-ggml/command-r/README.md
@@ -10,23 +10,33 @@
 
 ## Get Model
 
+Here we use the `c4ai-command-r-plus-GGUF` model as an example. You can download the model from the Hugging Face model hub.
+
 ```bash
-curl -LO https://huggingface.co/andrewcanis/c4ai-command-r-v01-GGUF/resolve/main/c4ai-command-r-v01-Q5_K_M.gguf
+curl -LO https://huggingface.co/pmysl/c4ai-command-r-plus-GGUF/resolve/main/command-r-plus-Q5_K_M-00001-of-00002.gguf
+curl -LO https://huggingface.co/pmysl/c4ai-command-r-plus-GGUF/resolve/main/command-r-plus-Q5_K_M-00002-of-00002.gguf
 ```
 
 ## Execute
 
-```console
+In this example, we use the system prompt with the definition of avaiable tools from [Example Rendered Tool Use Prompt](https://huggingface.co/CohereForAI/c4ai-command-r-plus).
+
+````console
 $ wasmedge --dir .:. \
-  --nn-preload default:GGML:AUTO:c4ai-command-r-v01-Q5_K_M.gguf \
+  --nn-preload default:GGML:AUTO:command-r-plus-Q5_K_M-00001-of-00002.gguf \
   ./wasmedge-ggml-command-r.wasm default
 
 USER:
-What's the capital of the United States?
+Whats the biggest penguin in the world?
 ASSISTANT:
-The capital of the United States is Washington, D.C.
-USER:
-How about Japan?
-ASSISTANT:
-Tokyo is the capital of Japan.
-```
+Action: ```json
+[
+    {
+        "tool_name": "internet_search",
+        "parameters": {
+            "query": "biggest penguin species"
+        }
+    }
+]
+```
+````
diff --git a/wasmedge-ggml/command-r/src/main.rs b/wasmedge-ggml/command-r/src/main.rs
@@ -141,16 +141,63 @@ fn main() {
     }
 
     let mut saved_prompt = String::new();
-    let system_prompt = String::from("You are a helpful, respectful and honest assistant. Always answer as short as possible, while being safe." );
+    let system_tool_prompt = r#"
+        # Safety Preamble
+        The instructions in this section override those in the task description and style guide sections. Don't answer questions that are harmful or immoral.
+
+        # System Preamble
+        ## Basic Rules
+        You are a powerful conversational AI trained by Cohere to help people. You are augmented by a number of tools, and your job is to use and consume the output of these tools to best help the user. You will see a conversation history between yourself and a user, ending with an utterance from the user. You will then see a specific instruction instructing you what kind of response to generate. When you answer the user's requests, you cite your sources in your answers, according to those instructions.
+
+        # User Preamble
+        ## Task and Context
+        You help people answer their questions and other requests interactively. You will be asked a very wide array of requests on all kinds of topics. You will be equipped with a wide range of search engines or similar tools to help you, which you use to research your answer. You should focus on serving the user's needs as best you can, which will be wide-ranging.
+
+        ## Style Guide
+        Unless the user asks for a different style of answer, you should answer in full sentences, using proper grammar and spelling.
+
+        ## Available Tools
+        Here is a list of tools that you have available to you:
+
+        ```python
+        def internet_search(query: str) -> List[Dict]:
+            """Returns a list of relevant document snippets for a textual query retrieved from the internet
+
+            Args:
+                query (str): Query to search the internet with
+            """
+            pass
+        ```
+
+        ```python
+        def directly_answer() -> List[Dict]:
+            """Calls a standard (un-augmented) AI chatbot to generate a response given the conversation history
+            """
+            pass
+        ```
+    "#;
+    let system_instruction_prompt = r#"
+        Write 'Action:' followed by a json-formatted list of actions that you want to perform in order to produce a good response to the user's last input. You can use any of the supplied tools any number of times, but you should aim to execute the minimum number of necessary actions for the input. You should use the `directly-answer` tool if calling the other tools is unnecessary. The list of actions you want to call should be formatted as a list of json objects, for example:
+        ```json
+        [
+            {
+                "tool_name": title of the tool in the specification,
+                "parameters": a dict of parameters to input into the tool as they are defined in the specs, or {} if it takes no parameters
+            }
+        ]```
+    "#;
 
     loop {
         println!("USER:");
         let input = read_input();
         //
         if saved_prompt.is_empty() {
             saved_prompt = format!(
-                "<|START_OF_TURN_TOKEN|><|USER_TOKEN|>{} {}<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>",
-                system_prompt, input
+                "<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>{}<|END_OF_TURN_TOKEN|>\
+                <|USER_TOKEN|>{}<|END_OF_TURN_TOKEN|>\
+                <|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>{}<|END_OF_TURN_TOKEN|>\
+                <|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>",
+                system_tool_prompt, input, system_instruction_prompt
             );
         } else {
             saved_prompt = format!("{} <|START_OF_TURN_TOKEN|><|USER_TOKEN|>{}<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>", saved_prompt, input);

diff --git a/wasmedge-ggml/command-r/wasmedge-ggml-command-r.wasm b/wasmedge-ggml/command-r/wasmedge-ggml-command-r.wasm
diff --git a/wasmedge-ggml/grammar/Cargo.toml b/wasmedge-ggml/grammar/Cargo.toml
@@ -0,0 +1,8 @@
+[package]
+name = "wasmedge-ggml-grammar"
+version = "0.1.0"
+edition = "2021"
+
+[dependencies]
+serde_json = "1.0"
+wasmedge-wasi-nn = "0.7.0"
diff --git a/wasmedge-ggml/grammar/README.md b/wasmedge-ggml/grammar/README.md
@@ -0,0 +1,37 @@
+# Grammar Example For WASI-NN with GGML Backend
+
+> [!NOTE]
+> Please refer to the [wasmedge-ggml/README.md](../README.md) for the general introduction and the setup of the WASI-NN plugin with GGML backend. This document will focus on the specific example of using grammar in ggml.
+
+## Get the Model
+
+In this example, we are going to use the [llama-2-7b](https://huggingface.co/TheBloke/Llama-2-7B-GGUF) model. Please note that we are not using a fine-tuned chat model.
+
+```bash
+curl -LO https://huggingface.co/TheBloke/Llama-2-7B-GGUF/resolve/main/llama-2-7b.Q5_K_M.gguf
+```
+
+## Parameters
+
+> [!NOTE]
+> Please check the parameters section of [wasmedge-ggml/README.md](https://github.com/second-state/WasmEdge-WASINN-examples/tree/master/wasmedge-ggml#parameters) first.
+
+In this example, we are going to use the `grammar` option to constrain the model to generate the JSON output in a specific format.
+
+You can check [the documents at llama.cpp](https://github.com/ggerganov/llama.cpp/tree/master/grammars) for more details about grammars.
+
+## Execute
+
+In this example, we are going to use the `n_predict` option to avoid the model from generating too many outputs.
+
+```console
+$ wasmedge --dir .:. \
+  --env n_predict=99 \
+  --nn-preload default:GGML:AUTO:llama-2-7b.Q5_K_M.gguf \
+  wasmedge-ggml-grammar.wasm default
+
+USER:
+JSON object with 5 country names as keys and their capitals as values:
+ASSISTANT:
+{"US": "Washington", "UK": "London", "Germany": "Berlin", "France": "Paris", "Italy": "Rome"}
+```