Llama-2/README.md

# Llama-2 Chat

*author: Jael*

<br />

## Description

A LLM operator generates answer given prompt in messages using a large language model or service.
This operator uses a pretrained [Llama-2](https://ai.meta.com/llama) to generate response.
By default, it will download the model file from [HuggingFace](https://huggingface.co/TheBloke) 
and then run the model with [Llama-cpp](https://github.com/ggerganov/llama.cpp).

This operator will automatically install and run model with llama-cpp.
If the automatic installation fails in your environment, please refer to [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) for instructions of manual installation.

<br />

## Code Example

Use the default model to continue the conversation from given messages.

*Write a pipeline with explicit inputs/outputs name specifications:*

```python
from towhee import pipe, ops

p = (
    pipe.input('question', 'docs', 'history')
        .map(('question', 'docs', 'history'), 'prompt', ops.prompt.question_answer())
        .map('prompt', 'answer', ops.LLM.Llama_2('llama-2-7b-chat'))
        .output('answer')
)

history=[('Who won the world series in 2020?', 'The Los Angeles Dodgers won the World Series in 2020.')]
question = 'Where was it played?'
answer = p(question, [], history).get()[0]
```

*Write a [retrieval-augmented generation pipeline](https://towhee.io/tasks/detail/pipeline/retrieval-augmented-generation) with explicit inputs/outputs name specifications:*

```python
from towhee import pipe, ops


temp = '''Use the following pieces of context to answer the question at the end.

{context}

Question: {question}
'''

system_msg = 'Your name is TowheeChat.'

q1 = 'Who are you?'
q2 = 'What is Towhee?'

p = (
    pipe.input('question', 'docs', 'history')
        .map(('question', 'docs', 'history'),
             'prompt',
             ops.prompt.template(temp, ['question', 'context'], sys_message))
        .map('prompt', 'answer',
             ops.LLM.Llama_2(max_tokens=200))
        .output('answer')
)

history = []
docs = []
ans1 = p(q1, docs, history).get()[0]
print(q1, ans1)

history.append((q1, ans1))
docs.append('Towhee is a cutting-edge framework designed to streamline the processing of unstructured data through the use of Large Language Model (LLM) based pipeline orchestration.')
ans2 = p(q2, docs, history)

print(q2, ans2)
```

<br />

## Factory Constructor

Create the operator via the following factory method:

***LLM.Llama_2(model_name_or_file: str)***

**Parameters:**

***model_name_or_file***: *str*

The model name or path to the model file in string, defaults to 'llama-2-7b-chat'.
If model name is in `supported_model_names`, it will download corresponding model file from HuggingFace models.
You can also use the local path of a model file, which can be ran by llama-cpp-python.

***\*\*kwargs***

Other model parameters such as temperature, max_tokens.

<br />

## Interface

The operator takes a piece of text in string as input.
It returns answer in json.

***\_\_call\_\_(txt)***

**Parameters:**

***messages***: *list*

​	A list of messages to set up chat.
Must be a list of dictionaries with key value from "system", "question", "answer". For example, [{"question": "a past question?", "answer": "a past answer."}, {"question": "current question?"}]

**Returns**:

*answer: str*

​	The answer generated.

<br />

***supported_model_names()***

**Returns**:

A dictionary of supported models with model name as key and huggingface hub id & model filename as value.

    {
        'llama-2-7b-chat': {
            'hf_id': 'TheBloke/Llama-2-7B-GGML',
            'filename': 'llama-2-7b.ggmlv3.q4_0.bin'
            },
        'llama-2-13-b-chat': {
            'hf_id': 'TheBloke/Llama-2-13B-GGML',
            'filename': 'llama-2-13b-chat.ggmlv3.q4_0.bin'
        }
    }