fix: split docs and start conceptual page (#1836)

This PR improves the guidance docs and adds a section that explains how
grammars are applied on a technical level
This commit is contained in:
drbh 2024-05-01 03:03:25 -04:00 committed by GitHub
parent dccab72549
commit 6073ece4fc
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 482 additions and 400 deletions

View File

@ -25,6 +25,8 @@
title: Non-core Model Serving
- local: basic_tutorials/safety
title: Safety
- local: basic_tutorials/using_guidance
title: Using Guidance, JSON, tools
- local: basic_tutorials/visual_language_models
title: Visual Language Models
title: Tutorials
@ -44,6 +46,6 @@
- local: conceptual/speculation
title: Speculation (Medusa, ngram)
- local: conceptual/guidance
title: Guidance, JSON, tools (using outlines)
title: How Guidance Works (via outlines)
title: Conceptual Guides

View File

@ -0,0 +1,419 @@
# Guidance
Text Generation Inference (TGI) now supports [JSON and regex grammars](#grammar-and-constraints) and [tools and functions](#tools-and-functions) to help developers guide LLM responses to fit their needs.
These feature are available starting from version `1.4.3`. They are accessible via the [text_generation](https://pypi.org/project/text-generation/) library. The tool support is compatible with OpenAI's client libraries. The following guide will walk you through the new features and how to use them!
_note: guidance is supported as grammar in the `/generate` endpoint and as tools in the `/chat/completions` endpoint._
## How it works
TGI leverages the [outlines](https://github.com/outlines-dev/outlines) library to efficiently parse and compile the grammatical structures and tools specified by users. This integration transforms the defined grammars into an intermediate representation that acts as a framework to guide and constrain content generation, ensuring that outputs adhere to the specified grammatical rules.
If you are interested in the technical details on how outlines is used in TGI, you can check out the [conceptual guidance documentation](../conceptual/guidance).
## Table of Contents 📚
### Grammar and Constraints
- [The Grammar Parameter](#the-grammar-parameter): Shape your AI's responses with precision.
- [Constrain with Pydantic](#constrain-with-pydantic): Define a grammar using Pydantic models.
- [JSON Schema Integration](#json-schema-integration): Fine-grained control over your requests via JSON schema.
- [Using the client](#using-the-client): Use TGI's client libraries to shape the AI's responses.
### Tools and Functions
- [The Tools Parameter](#the-tools-parameter): Enhance the AI's capabilities with predefined functions.
- [Via the client](#text-generation-inference-client): Use TGI's client libraries to interact with the Messages API and Tool functions.
- [OpenAI integration](#openai-integration): Use OpenAI's client libraries to interact with TGI's Messages API and Tool functions.
## Grammar and Constraints 🛣️
### The Grammar Parameter
In TGI `1.4.3`, we've introduced the grammar parameter, which allows you to specify the format of the response you want from the LLM.
Using curl, you can make a request to TGI's Messages API with the grammar parameter. This is the most primitive way to interact with the API and using [Pydantic](#constrain-with-pydantic) is recommended for ease of use and readability.
```json
curl localhost:3000/generate \
-X POST \
-H 'Content-Type: application/json' \
-d '{
"inputs": "I saw a puppy a cat and a raccoon during my bike ride in the park",
"parameters": {
"repetition_penalty": 1.3,
"grammar": {
"type": "json",
"value": {
"properties": {
"location": {
"type": "string"
},
"activity": {
"type": "string"
},
"animals_seen": {
"type": "integer",
"minimum": 1,
"maximum": 5
},
"animals": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": ["location", "activity", "animals_seen", "animals"]
}
}
}
}'
// {"generated_text":"{ \n\n\"activity\": \"biking\",\n\"animals\": [\"puppy\",\"cat\",\"raccoon\"],\n\"animals_seen\": 3,\n\"location\": \"park\"\n}"}
```
A grammar can be defined using Pydantic models, JSON schemas, or regular expressions. The LLM will then generate a response that conforms to the specified grammar.
> Note: A grammar must compile to an intermediate representation to constrain the output. Grammar compilation is a computationally expensive and may take a few seconds to complete on the first request. Subsequent requests will use the cached grammar and will be much faster.
### Constrain with Pydantic
Using Pydantic models we can define a similar grammar as the previous example in a shorter and more readable way.
```python
import requests
from pydantic import BaseModel, conint
from typing import List
class Animals(BaseModel):
location: str
activity: str
animals_seen: conint(ge=1, le=5) # Constrained integer type
animals: List[str]
prompt = "convert to JSON: I saw a puppy a cat and a raccoon during my bike ride in the park"
data = {
"inputs": prompt,
"parameters": {
"repetition_penalty": 1.3,
"grammar": {
"type": "json",
"value": Animals.schema()
}
}
}
headers = {
"Content-Type": "application/json",
}
response = requests.post(
'http://127.0.0.1:3000/generate',
headers=headers,
json=data
)
print(response.json())
# {'generated_text': '{ "activity": "bike riding", "animals": ["puppy","cat","raccoon"],"animals_seen": 3, "location":"park" }'}
```
### JSON Schema Integration
If Pydantic's not your style, go raw with direct JSON Schema integration. This is similar to the first example but with programmatic control.
```python
import requests
json_schema = {
"properties": {
"location": {
"type": "string"
},
"activity": {
"type": "string"
},
"animals_seen": {
"type": "integer",
"minimum": 1,
"maximum": 5
},
"animals": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": ["location", "activity", "animals_seen", "animals"]
}
data = {
"inputs": "convert to JSON: I saw a puppy a cat and a raccoon during my bike ride in the park",
"parameters": {
"max_new_tokens": 200,
"repetition_penalty": 1.3,
"grammar": {
"type": "json",
"value": json_schema
}
}
}
headers = {
"Content-Type": "application/json",
}
response = requests.post(
'http://127.0.0.1:3000/generate',
headers=headers,
json=data
)
print(response.json())
# {'generated_text': '{\n"activity": "biking",\n"animals": ["puppy","cat","raccoon"]\n , "animals_seen": 3,\n "location":"park"}'}
```
### Using the client
TGI provides a client library to that make it easy to send requests with all of the parameters we've discussed above. Here's an example of how to use the client to send a request with a grammar parameter.
```python
from text_generation import AsyncClient
from text_generation.types import GrammarType
# NOTE: tools defined above and removed for brevity
# Define an async function to encapsulate the async operation
async def main():
client = AsyncClient(base_url="http://localhost:3000")
# Use 'await' to wait for the async method 'chat' to complete
response = await client.generate(
"Whats Googles DNS",
max_new_tokens=10,
decoder_input_details=True,
seed=1,
grammar={
"type": GrammarType.Regex,
"value": "((25[0-5]|2[0-4]\\d|[01]?\\d\\d?)\\.){3}(25[0-5]|2[0-4]\\d|[01]?\\d\\d?)",
},
)
# Once the response is received, you can process it
print(response.generated_text)
# Ensure the main async function is run in the event loop
if __name__ == "__main__":
import asyncio
asyncio.run(main())
# 118.8.0.84
```
## Tools and Functions 🛠️
### The Tools Parameter
In addition to the grammar parameter, we've also introduced a set of tools and functions to help you get the most out of the Messages API.
Tools are a set of user defined functions that can be used in tandem with the chat functionality to enhance the LLM's capabilities. Functions, similar to grammar are defined as JSON schema and can be passed as part of the parameters to the Messages API.
Functions, similar to grammar are defined as JSON schema and can be passed as part of the parameters to the Messages API.
```json
curl localhost:3000/v1/chat/completions \
-X POST \
-H 'Content-Type: application/json' \
-d '{
"model": "tgi",
"messages": [
{
"role": "user",
"content": "What is the weather like in New York?"
}
],
"tools": [
{
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get the current weather",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA"
},
"format": {
"type": "string",
"enum": ["celsius", "fahrenheit"],
"description": "The temperature unit to use. Infer this from the users location."
}
},
"required": ["location", "format"]
}
}
}
],
"tool_choice": "get_current_weather"
}'
// {"id":"","object":"text_completion","created":1709051640,"model":"HuggingFaceH4/zephyr-7b-beta","system_fingerprint":"1.4.3-native","choices":[{"index":0,"message":{"role":"assistant","tool_calls":{"id":0,"type":"function","function":{"description":null,"name":"tools","parameters":{"format":"celsius","location":"New York"}}}},"logprobs":null,"finish_reason":"eos_token"}],"usage":{"prompt_tokens":157,"completion_tokens":19,"total_tokens":176}}
```
### Text Generation Inference Client
TGI provides a client library to interact with the Messages API and Tool functions. The client library is available in both synchronous and asynchronous versions.
```python
from text_generation import AsyncClient
# NOTE: tools defined above and removed for brevity
# Define an async function to encapsulate the async operation
async def main():
client = AsyncClient(base_url="http://localhost:3000")
# Use 'await' to wait for the async method 'chat' to complete
response = await client.chat(
max_tokens=100,
seed=1,
tools=tools,
presence_penalty=-1.1,
messages=[
{
"role": "system",
"content": "You're a helpful assistant! Answer the users question best you can.",
},
{
"role": "user",
"content": "What is the weather like in Brooklyn, New York?",
},
],
)
# Once the response is received, you can process it
print(response.choices[0].message.tool_calls)
# Ensure the main async function is run in the event loop
if __name__ == "__main__":
import asyncio
asyncio.run(main())
# {"id":"","object":"text_completion","created":1709051942,"model":"HuggingFaceH4/zephyr-7b-beta","system_fingerprint":"1.4.3-native","choices":[{"index":0,"message":{"role":"assistant","tool_calls":{"id":0,"type":"function","function":{"description":null,"name":"tools","parameters":{"format":"celsius","location":"New York"}}}},"logprobs":null,"finish_reason":"eos_token"}],"usage":{"prompt_tokens":157,"completion_tokens":20,"total_tokens":177}}
```
<details>
<summary>Tools used in example above</summary>
```python
tools = [
{
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get the current weather",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA",
},
"format": {
"type": "string",
"enum": ["celsius", "fahrenheit"],
"description": "The temperature unit to use. Infer this from the users location.",
},
},
"required": ["location", "format"],
},
},
},
{
"type": "function",
"function": {
"name": "get_n_day_weather_forecast",
"description": "Get an N-day weather forecast",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA",
},
"format": {
"type": "string",
"enum": ["celsius", "fahrenheit"],
"description": "The temperature unit to use. Infer this from the users location.",
},
"num_days": {
"type": "integer",
"description": "The number of days to forecast",
},
},
"required": ["location", "format", "num_days"],
},
},
}
]
```
</details>
### OpenAI integration
TGI exposes an OpenAI-compatible API, which means you can use OpenAI's client libraries to interact with TGI's Messages API and Tool functions.
However there are some minor differences in the API, for example `tool_choice="auto"` will ALWAYS choose the tool for you. This is different from OpenAI's API where `tool_choice="auto"` will choose a tool if the model thinks it's necessary.
```python
from openai import OpenAI
# Initialize the client, pointing it to one of the available models
client = OpenAI(
base_url="http://localhost:3000/v1",
api_key="_",
)
# NOTE: tools defined above and removed for brevity
chat_completion = client.chat.completions.create(
model="tgi",
messages=[
{
"role": "system",
"content": "Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous.",
},
{
"role": "user",
"content": "What's the weather like the next 3 days in San Francisco, CA?",
},
],
tools=tools,
tool_choice="auto", # tool selected by model
max_tokens=500,
)
called = chat_completion.choices[0].message.tool_calls
print(called)
# {
# "id": 0,
# "type": "function",
# "function": {
# "description": None,
# "name": "tools",
# "parameters": {
# "format": "celsius",
# "location": "San Francisco, CA",
# "num_days": 3,
# },
# },
# }
```

View File

@ -1,425 +1,86 @@
# Guidance
Text Generation Inference (TGI) now supports [JSON and regex grammars](#grammar-and-constraints) and [tools and functions](#tools-and-functions) to help developers guide LLM responses to fit their needs.
## What is Guidance?
These feature are available starting from version `1.4.3`. They are accessible via the [text_generation](https://pypi.org/project/text-generation/) library. The tool support is compatible with OpenAI's client libraries. The following guide will walk you through the new features and how to use them!
Guidance is a feature that allows users to constrain the generation of a large language model with a specified grammar. This feature is particularly useful when you want to generate text that follows a specific structure or uses a specific set of words or produce output in a specific format.
> The Grammar guidance support is currently only available in the TGI API due to lack of support in Open AI API.
## How is it used?
## Quick Start
Guidance can be in many ways and the community is always finding new ways to use it. Here are some examples of how you can use guidance:
Before we jump into the deep end, ensure your system is using TGI version `1.4.3` or later to access all the features we're about to explore in this guide.
Technically, guidance can be used to generate:
If you're not up to date, grab the latest version and let's get started!
- a specific JSON object
- a function signature
- typed output like a list of integers
## How it works
However these use cases can span a wide range of applications, such as:
TGI leverages the [outlines](https://github.com/outlines-dev/outlines) library to efficiently parse and compile the grammatical structures and tools specified by users. This integration transforms the defined grammars into an intermediate representation that acts as a framework to guide and constrain content generation, ensuring that outputs adhere to the specified grammatical rules.
- extracting structured data from unstructured text
- summarizing text into a specific format
- limit output to specific classes of words (act as a LLM powered classifier)
- generate the input to specific APIs or services
- provide reliable and consistent output for downstream tasks
- extract data from multimodal inputs
## Table of Contents 📚
## How it works?
### Grammar and Constraints
Diving into the details, guidance is enabled by including a grammar with a generation request that is compiled, and used to modify the chosen tokens.
- [The Grammar Parameter](#the-grammar-parameter): Shape your AI's responses with precision.
- [Constrain with Pydantic](#constrain-with-pydantic): Define a grammar using Pydantic models.
- [JSON Schema Integration](#json-schema-integration): Fine-grained control over your requests via JSON schema.
- [Using the client](#using-the-client): Use TGI's client libraries to shape the AI's responses.
This process can be broken down into the following steps:
### Tools and Functions
1. A request is sent to the backend, it is processed and placed in batch. Processing includes compiling the grammar into a finite state machine and a grammar state.
- [The Tools Parameter](#the-tools-parameter): Enhance the AI's capabilities with predefined functions.
- [Via the client](#text-generation-inference-client): Use TGI's client libraries to interact with the Messages API and Tool functions.
- [OpenAI integration](#openai-integration): Use OpenAI's client libraries to interact with TGI's Messages API and Tool functions.
<div class="flex justify-center">
<img
class="block dark:hidden"
src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/tgi/request-to-batch.gif"
/>
<img
class="hidden dark:block"
src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/tgi/request-to-batch-dark.gif"
/>
</div>
## Grammar and Constraints 🛣️
2. The model does a forward pass over the batch. This returns probabilities for each token in the vocabulary for each request in the batch.
### The Grammar Parameter
3. The process of choosing one of those tokens is called `sampling`. The model samples from the distribution of probabilities to choose the next token. In TGI all of the steps before sampling are called `processor`. Grammars are applied as a processor that masks out tokens that are not allowed by the grammar.
In TGI `1.4.3`, we've introduced the grammar parameter, which allows you to specify the format of the response you want from the AI. This is a game-changer for those who need precise control over the AI's output.
<div class="flex justify-center">
<img
class="block dark:hidden"
src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/tgi/logit-grammar-mask.gif"
/>
<img
class="hidden dark:block"
src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/tgi/logit-grammar-mask-dark.gif"
/>
</div>
Using curl, you can make a request to TGI's Messages API with the grammar parameter. This is the most primitive way to interact with the API and using [Pydantic](#constrain-with-pydantic) is recommended for ease of use and readability.
4. The grammar mask is applied and the model samples from the remaining tokens. Once a token is chosen, we update the grammar state with the new token, to prepare it for the next pass.
```json
curl localhost:3000/generate \
-X POST \
-H 'Content-Type: application/json' \
-d '{
"inputs": "I saw a puppy a cat and a raccoon during my bike ride in the park",
"parameters": {
"repetition_penalty": 1.3,
"grammar": {
"type": "json",
"value": {
"properties": {
"location": {
"type": "string"
},
"activity": {
"type": "string"
},
"animals_seen": {
"type": "integer",
"minimum": 1,
"maximum": 5
},
"animals": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": ["location", "activity", "animals_seen", "animals"]
}
}
}
}'
// {"generated_text":"{ \n\n\"activity\": \"biking\",\n\"animals\": [\"puppy\",\"cat\",\"raccoon\"],\n\"animals_seen\": 3,\n\"location\": \"park\"\n}"}
<div class="flex justify-center">
<img
class="block dark:hidden"
src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/tgi/sample-logits.gif"
/>
<img
class="hidden dark:block"
src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/tgi/sample-logits-dark.gif"
/>
</div>
```
## How to use Guidance?
A grammar can be defined using Pydantic models, JSON schemas, or regular expressions. The AI will then generate a response that conforms to the specified grammar.
There are two main ways to use guidance; you can either use the `/generate` endpoint with a grammar or use the `/chat/completion` endpoint with tools.
> Note: A grammar must compile to an intermediate representation to constrain the output. Grammar compilation is a computationally expensive and may take a few seconds to complete on the first request. Subsequent requests will use the cached grammar and will be much faster.
Under the hood tools are a special case of grammars that allows the model to choose one or none of the provided tools.
### Constrain with Pydantic
Please refer to [using guidance](../basic_tutorial/using_guidance) for more examples and details on how to use guidance in Python, JavaScript, and cURL.
Pydantic is a powerful library for data validation and settings management. It's the perfect tool for crafting the a specific response format.
### Getting the most out of guidance
Using Pydantic models we can define a similar grammar as the previous example in a shorter and more readable way.
Depending on how you are using guidance, you may want to make use of different features. Here are some tips to get the most out of guidance:
```python
import requests
from pydantic import BaseModel, conint
from typing import List
class Animals(BaseModel):
location: str
activity: str
animals_seen: conint(ge=1, le=5) # Constrained integer type
animals: List[str]
prompt = "convert to JSON: I saw a puppy a cat and a raccoon during my bike ride in the park"
data = {
"inputs": prompt,
"parameters": {
"repetition_penalty": 1.3,
"grammar": {
"type": "json",
"value": Animals.schema()
}
}
}
headers = {
"Content-Type": "application/json",
}
response = requests.post(
'http://127.0.0.1:3000/generate',
headers=headers,
json=data
)
print(response.json())
# {'generated_text': '{ "activity": "bike riding", "animals": ["puppy","cat","raccoon"],"animals_seen": 3, "location":"park" }'}
```
### JSON Schema Integration
If Pydantic's not your style, go raw with direct JSON Schema integration. It's like having a conversation with the AI in its own language. This is simliar to the first example but with programmatic control.
```python
import requests
json_schema = {
"properties": {
"location": {
"type": "string"
},
"activity": {
"type": "string"
},
"animals_seen": {
"type": "integer",
"minimum": 1,
"maximum": 5
},
"animals": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": ["location", "activity", "animals_seen", "animals"]
}
data = {
"inputs": "convert to JSON: I saw a puppy a cat and a raccoon during my bike ride in the park",
"parameters": {
"max_new_tokens": 200,
"repetition_penalty": 1.3,
"grammar": {
"type": "json",
"value": json_schema
}
}
}
headers = {
"Content-Type": "application/json",
}
response = requests.post(
'http://127.0.0.1:3000/generate',
headers=headers,
json=data
)
print(response.json())
# {'generated_text': '{\n"activity": "biking",\n"animals": ["puppy","cat","raccoon"]\n , "animals_seen": 3,\n "location":"park"}'}
```
### Using the client
TGI provides a client library to that make it easy to send requests with all of the parameters we've discussed above. Here's an example of how to use the client to send a request with a grammar parameter.
```python
from text_generation import AsyncClient
from text_generation.types import GrammarType
# NOTE: tools defined above and removed for brevity
# Define an async function to encapsulate the async operation
async def main():
client = AsyncClient(base_url="http://localhost:3000")
# Use 'await' to wait for the async method 'chat' to complete
response = await client.generate(
"Whats Googles DNS",
max_new_tokens=10,
decoder_input_details=True,
seed=1,
grammar={
"type": GrammarType.Regex,
"value": "((25[0-5]|2[0-4]\\d|[01]?\\d\\d?)\\.){3}(25[0-5]|2[0-4]\\d|[01]?\\d\\d?)",
},
)
# Once the response is received, you can process it
print(response.generated_text)
# Ensure the main async function is run in the event loop
if __name__ == "__main__":
import asyncio
asyncio.run(main())
# 118.8.0.84
```
## Tools and Functions 🛠️
### The Tools Parameter
In addition to the grammar parameter, we've also introduced a set of tools and functions to help you get the most out of the Messages API.
Tools are a set of user defined functions that can be used in tandem with the chat functionality to enhance the AI's capabilities. You can use these tools to perform a variety of tasks, such as data manipulation, formatting, and more.
Functions, similar to grammar are defined as JSON schema and can be passed as part of the parameters to the Messages API.
```json
curl localhost:3000/v1/chat/completions \
-X POST \
-H 'Content-Type: application/json' \
-d '{
"model": "tgi",
"messages": [
{
"role": "user",
"content": "What is the weather like in New York?"
}
],
"tools": [
{
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get the current weather",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA"
},
"format": {
"type": "string",
"enum": ["celsius", "fahrenheit"],
"description": "The temperature unit to use. Infer this from the users location."
}
},
"required": ["location", "format"]
}
}
}
],
"tool_choice": "get_current_weather"
}'
// {"id":"","object":"text_completion","created":1709051640,"model":"HuggingFaceH4/zephyr-7b-beta","system_fingerprint":"1.4.3-native","choices":[{"index":0,"message":{"role":"assistant","tool_calls":{"id":0,"type":"function","function":{"description":null,"name":"tools","parameters":{"format":"celsius","location":"New York"}}}},"logprobs":null,"finish_reason":"eos_token"}],"usage":{"prompt_tokens":157,"completion_tokens":19,"total_tokens":176}}
```
<details>
<summary>Tools used in example below</summary>
```python
tools = [
{
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get the current weather",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA",
},
"format": {
"type": "string",
"enum": ["celsius", "fahrenheit"],
"description": "The temperature unit to use. Infer this from the users location.",
},
},
"required": ["location", "format"],
},
},
},
{
"type": "function",
"function": {
"name": "get_n_day_weather_forecast",
"description": "Get an N-day weather forecast",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA",
},
"format": {
"type": "string",
"enum": ["celsius", "fahrenheit"],
"description": "The temperature unit to use. Infer this from the users location.",
},
"num_days": {
"type": "integer",
"description": "The number of days to forecast",
},
},
"required": ["location", "format", "num_days"],
},
},
}
]
```
</details>
### Text Generation Inference Client
TGI provides a client library to interact with the Messages API and Tool functions. The client library is available in both synchronous and asynchronous versions.
```python
from text_generation import AsyncClient
# NOTE: tools defined above and removed for brevity
# Define an async function to encapsulate the async operation
async def main():
client = AsyncClient(base_url="http://localhost:3000")
# Use 'await' to wait for the async method 'chat' to complete
response = await client.chat(
max_tokens=100,
seed=1,
tools=tools,
presence_penalty=-1.1,
messages=[
{
"role": "system",
"content": "You're a helpful assistant! Answer the users question best you can.",
},
{
"role": "user",
"content": "What is the weather like in Brooklyn, New York?",
},
],
)
# Once the response is received, you can process it
print(response.choices[0].message.tool_calls)
# Ensure the main async function is run in the event loop
if __name__ == "__main__":
import asyncio
asyncio.run(main())
# {"id":"","object":"text_completion","created":1709051942,"model":"HuggingFaceH4/zephyr-7b-beta","system_fingerprint":"1.4.3-native","choices":[{"index":0,"message":{"role":"assistant","tool_calls":{"id":0,"type":"function","function":{"description":null,"name":"tools","parameters":{"format":"celsius","location":"New York"}}}},"logprobs":null,"finish_reason":"eos_token"}],"usage":{"prompt_tokens":157,"completion_tokens":20,"total_tokens":177}}
```
### OpenAI integration
TGI exposes an OpenAI-compatible API, which means you can use OpenAI's client libraries to interact with TGI's Messages API and Tool functions.
However there are some minor differences in the API, for example `tool_choice="auto"` will ALWAYS choose the tool for you. This is different from OpenAI's API where `tool_choice="auto"` will choose a tool if the model thinks it's necessary.
```python
from openai import OpenAI
# Initialize the client, pointing it to one of the available models
client = OpenAI(
base_url="http://localhost:3000/v1",
api_key="_",
)
# NOTE: tools defined above and removed for brevity
chat_completion = client.chat.completions.create(
model="tgi",
messages=[
{
"role": "system",
"content": "Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous.",
},
{
"role": "user",
"content": "What's the weather like the next 3 days in San Francisco, CA?",
},
],
tools=tools,
tool_choice="auto", # tool selected by model
max_tokens=500,
)
called = chat_completion.choices[0].message.tool_calls
print(called)
# {
# "id": 0,
# "type": "function",
# "function": {
# "description": None,
# "name": "tools",
# "parameters": {
# "format": "celsius",
# "location": "San Francisco, CA",
# "num_days": 3,
# },
# },
# }
```
- If you are using the `/generate` with a `grammar` it is recommended to include the grammar in the prompt prefixed by something like `Please use the following JSON schema to generate the output:`. This will help the model understand the context of the grammar and generate the output accordingly.
- If you are getting a response with many repeated tokens, please use the `frequency_penalty` or `repetition_penalty` to reduce the number of repeated tokens in the output.