Skip to content
Snippets Groups Projects
Unverified Commit bfbf10bf authored by Simonas's avatar Simonas Committed by GitHub
Browse files

Merge pull request #33 from aurelio-labs/luca/fix-on-embeddings-check 

Fix for embeddings
parents 45ce5991 80118447
No related branches found
Tags v0.0.10
No related merge requests found
Hide whitespace changes
Inline Side-by-side
.gitignore
+ 1
− 0
View file @ bfbf10bf
...@@ -17,3 +17,4 @@ mac.env ...@@ -17,3 +17,4 @@ mac.env
.coverage .coverage
.coverage.* .coverage.*
.pytest_cache .pytest_cache
test.py
\ No newline at end of file
docs/examples/function_calling.ipynb
+ 66
− 72
View file @ bfbf10bf
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
## Define LLMs ## Define LLMs
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
%reload_ext dotenv
%dotenv
```
%% Cell type:code id: tags:
``` python
# OpenAI # OpenAI
import os
import openai import openai
from semantic_router.utils.logger import logger from semantic_router.utils.logger import logger
# Docs # https://platform.openai.com/docs/guides/function-calling # Docs # https://platform.openai.com/docs/guides/function-calling
def llm_openai(prompt: str, model: str = "gpt-4") -> str: def llm_openai(prompt: str, model: str = "gpt-4") -> str:
try: try:
logger.info(f"Calling {model} model") logger.info(f"Calling {model} model")
response = openai.chat.completions.create( response = openai.chat.completions.create(
model=model, model=model,
messages=[ messages=[
{"role": "system", "content": f"{prompt}"}, {"role": "system", "content": f"{prompt}"},
], ],
) )
ai_message = response.choices[0].message.content ai_message = response.choices[0].message.content
if not ai_message: if not ai_message:
raise Exception("AI message is empty", ai_message) raise Exception("AI message is empty", ai_message)
logger.info(f"AI message: {ai_message}") logger.info(f"AI message: {ai_message}")
return ai_message return ai_message
except Exception as e: except Exception as e:
raise Exception("Failed to call OpenAI API", e) raise Exception("Failed to call OpenAI API", e)
``` ```
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
# Mistral # Mistral
import os import os
import requests import requests
# Docs https://huggingface.co/docs/transformers/main_classes/text_generation # Docs https://huggingface.co/docs/transformers/main_classes/text_generation
HF_API_TOKEN = os.environ["HF_API_TOKEN"] HF_API_TOKEN = os.getenv("HF_API_TOKEN")
def llm_mistral(prompt: str) -> str: def llm_mistral(prompt: str) -> str:
api_url = "https://z5t4cuhg21uxfmc3.us-east-1.aws.endpoints.huggingface.cloud/" api_url = "https://z5t4cuhg21uxfmc3.us-east-1.aws.endpoints.huggingface.cloud/"
headers = { headers = {
"Authorization": f"Bearer {HF_API_TOKEN}", "Authorization": f"Bearer {HF_API_TOKEN}",
"Content-Type": "application/json", "Content-Type": "application/json",
} }
logger.info("Calling Mistral model") logger.info("Calling Mistral model")
response = requests.post( response = requests.post(
api_url, api_url,
headers=headers, headers=headers,
json={ json={
"inputs": f"You are a helpful assistant, user query: {prompt}", "inputs": f"You are a helpful assistant, user query: {prompt}",
"parameters": { "parameters": {
"max_new_tokens": 200, "max_new_tokens": 200,
"temperature": 0.1, "temperature": 0.1,
}, },
}, },
) )
if response.status_code != 200: if response.status_code != 200:
raise Exception("Failed to call HuggingFace API", response.text) raise Exception("Failed to call HuggingFace API", response.text)
ai_message = response.json()[0]["generated_text"] ai_message = response.json()[0]["generated_text"]
if not ai_message: if not ai_message:
raise Exception("AI message is empty", ai_message) raise Exception("AI message is empty", ai_message)
logger.info(f"AI message: {ai_message}") logger.info(f"AI message: {ai_message}")
return ai_message return ai_message
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Now we need to generate config from function schema using LLM ### Now we need to generate config from function schema using LLM
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
import inspect import inspect
from typing import Any from typing import Any
def get_function_schema(function) -> dict[str, Any]: def get_function_schema(function) -> dict[str, Any]:
schema = { schema = {
"name": function.__name__, "name": function.__name__,
"description": str(inspect.getdoc(function)), "description": str(inspect.getdoc(function)),
"signature": str(inspect.signature(function)), "signature": str(inspect.signature(function)),
"output": str( "output": str(
inspect.signature(function).return_annotation, inspect.signature(function).return_annotation,
), ),
} }
return schema return schema
``` ```
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
import json import json
from semantic_router.utils.logger import logger from semantic_router.utils.logger import logger
def generate_route(function) -> dict: def generate_route(function) -> dict:
logger.info("Generating config...") logger.info("Generating config...")
example_schema = { example_schema = {
"name": "get_weather", "name": "get_weather",
"description": "Useful to get the weather in a specific location", "description": "Useful to get the weather in a specific location",
"signature": "(location: str) -> str", "signature": "(location: str) -> str",
"output": "<class 'str'>", "output": "<class 'str'>",
} }
example_config = { example_config = {
"name": "get_weather", "name": "get_weather",
"utterances": [ "utterances": [
"What is the weather like in SF?", "What is the weather like in SF?",
"What is the weather in Cyprus?", "What is the weather in Cyprus?",
"weather in London?", "weather in London?",
"Tell me the weather in New York", "Tell me the weather in New York",
"what is the current weather in Paris?", "what is the current weather in Paris?",
], ],
} }
function_schema = get_function_schema(function) function_schema = get_function_schema(function)
prompt = f""" prompt = f"""
You are a helpful assistant designed to output JSON. You are a helpful assistant designed to output JSON.
Given the following function schema Given the following function schema
{function_schema} {function_schema}
generate a routing config with the format: generate a routing config with the format:
{example_config} {example_config}
For example: For example:
Input: {example_schema} Input: {example_schema}
Output: {example_config} Output: {example_config}
Input: {function_schema} Input: {function_schema}
Output: Output:
""" """
ai_message = llm_openai(prompt) ai_message = llm_openai(prompt)
ai_message = ai_message.replace("CONFIG:", "").replace("'", '"').strip().rstrip(",") ai_message = ai_message.replace("CONFIG:", "").replace("'", '"').strip().rstrip(",")
try: try:
route_config = json.loads(ai_message) route_config = json.loads(ai_message)
logger.info(f"Generated config: {route_config}") logger.info(f"Generated config: {route_config}")
return route_config return route_config
except json.JSONDecodeError as json_error: except json.JSONDecodeError as json_error:
logger.error(f"JSON parsing error {json_error}") logger.error(f"JSON parsing error {json_error}")
print(f"AI message: {ai_message}") print(f"AI message: {ai_message}")
return {"error": "Failed to generate config"} return {"error": "Failed to generate config"}
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
Extract function parameters using `Mistral` open-source model Extract function parameters using `Mistral` open-source model
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
def extract_parameters(query: str, function) -> dict: def extract_parameters(query: str, function) -> dict:
logger.info("Extracting parameters...") logger.info("Extracting parameters...")
example_query = "How is the weather in Hawaii right now in International units?" example_query = "How is the weather in Hawaii right now in International units?"
example_schema = { example_schema = {
"name": "get_weather", "name": "get_weather",
"description": "Useful to get the weather in a specific location", "description": "Useful to get the weather in a specific location",
"signature": "(location: str, degree: str) -> str", "signature": "(location: str, degree: str) -> str",
"output": "<class 'str'>", "output": "<class 'str'>",
} }
example_parameters = { example_parameters = {
"location": "London", "location": "London",
"degree": "Celsius", "degree": "Celsius",
} }
prompt = f""" prompt = f"""
You are a helpful assistant designed to output JSON. You are a helpful assistant designed to output JSON.
Given the following function schema Given the following function schema
{get_function_schema(function)} {get_function_schema(function)}
and query and query
{query} {query}
extract the parameters values from the query, in a valid JSON format. extract the parameters values from the query, in a valid JSON format.
Example: Example:
Input: Input:
query: {example_query} query: {example_query}
schema: {example_schema} schema: {example_schema}
Output: Output:
parameters: {example_parameters} parameters: {example_parameters}
Input: Input:
query: {query} query: {query}
schema: {get_function_schema(function)} schema: {get_function_schema(function)}
Output: Output:
parameters: parameters:
""" """
ai_message = llm_mistral(prompt) ai_message = llm_mistral(prompt)
ai_message = ai_message.replace("CONFIG:", "").replace("'", '"').strip().rstrip(",") ai_message = ai_message.replace("CONFIG:", "").replace("'", '"').strip().rstrip(",")
try: try:
parameters = json.loads(ai_message) parameters = json.loads(ai_message)
logger.info(f"Extracted parameters: {parameters}") logger.info(f"Extracted parameters: {parameters}")
return parameters return parameters
except json.JSONDecodeError as json_error: except json.JSONDecodeError as json_error:
logger.error(f"JSON parsing error {json_error}") logger.error(f"JSON parsing error {json_error}")
return {"error": "Failed to extract parameters"} return {"error": "Failed to extract parameters"}
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
Set up the routing layer Set up the routing layer
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
from semantic_router.schema import Route from semantic_router.schema import Route
from semantic_router.encoders import CohereEncoder, OpenAIEncoder
from semantic_router.layer import RouteLayer
from semantic_router.utils.logger import logger
def create_router(routes: list[dict]) -> RouteLayer:
logger.info("Creating route layer...")
encoder = OpenAIEncoder
```
%% Cell type:code id: tags:
``` python
from semantic_router.schema import Route
from semantic_router.encoders import CohereEncoder from semantic_router.encoders import CohereEncoder
from semantic_router.layer import RouteLayer from semantic_router.layer import RouteLayer
from semantic_router.utils.logger import logger from semantic_router.utils.logger import logger
def create_router(routes: list[dict]) -> RouteLayer: def create_router(routes: list[dict]) -> RouteLayer:
logger.info("Creating route layer...") logger.info("Creating route layer...")
encoder = CohereEncoder() encoder = OpenAIEncoder()
route_list: list[Route] = [] route_list: list[Route] = []
for route in routes: for route in routes:
if "name" in route and "utterances" in route: if "name" in route and "utterances" in route:
print(f"Route: {route}") print(f"Route: {route}")
route_list.append(Route(name=route["name"], utterances=route["utterances"])) route_list.append(Route(name=route["name"], utterances=route["utterances"]))
else: else:
logger.warning(f"Misconfigured route: {route}") logger.warning(f"Misconfigured route: {route}")
return RouteLayer(encoder=encoder, routes=route_list) return RouteLayer(encoder=encoder, routes=route_list)
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
Set up calling functions Set up calling functions
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
from typing import Callable from typing import Callable
def call_function(function: Callable, parameters: dict[str, str]): def call_function(function: Callable, parameters: dict[str, str]):
try: try:
return function(**parameters) return function(**parameters)
except TypeError as e: except TypeError as e:
logger.error(f"Error calling function: {e}") logger.error(f"Error calling function: {e}")
def call_llm(query: str): def call_llm(query: str):
return llm_mistral(query) return llm_mistral(query)
def call(query: str, functions: list[Callable], router: RouteLayer): def call(query: str, functions: list[Callable], router: RouteLayer):
function_name = router(query) function_name = router(query)
if not function_name: if not function_name:
logger.warning("No function found") logger.warning("No function found")
return call_llm(query) return call_llm(query)
for function in functions: for function in functions:
if function.__name__ == function_name: if function.__name__ == function_name:
parameters = extract_parameters(query, function) parameters = extract_parameters(query, function)
print(f"parameters: {parameters}") print(f"parameters: {parameters}")
return call_function(function, parameters) return call_function(function, parameters)
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Workflow ### Workflow
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
def get_time(location: str) -> str: def get_time(location: str) -> str:
"""Useful to get the time in a specific location""" """Useful to get the time in a specific location"""
print(f"Calling `get_time` function with location: {location}") print(f"Calling `get_time` function with location: {location}")
return "get_time" return "get_time"
def get_news(category: str, country: str) -> str: def get_news(category: str, country: str) -> str:
"""Useful to get the news in a specific country""" """Useful to get the news in a specific country"""
print( print(
f"Calling `get_news` function with category: {category} and country: {country}" f"Calling `get_news` function with category: {category} and country: {country}"
) )
return "get_news" return "get_news"
# Registering functions to the router # Registering functions to the router
route_get_time = generate_route(get_time) route_get_time = generate_route(get_time)
route_get_news = generate_route(get_news) route_get_news = generate_route(get_news)
routes = [route_get_time, route_get_news] routes = [route_get_time, route_get_news]
router = create_router(routes) router = create_router(routes)
# Tools # Tools
tools = [get_time, get_news] tools = [get_time, get_news]
``` ```
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
call(query="What is the time in Stockholm?", functions=tools, router=router) def get_time(location: str) -> str:
call(query="What is the tech news in the Lithuania?", functions=tools, router=router) """Useful to get the time in a specific location"""
call(query="Hi!", functions=tools, router=router) print(f"Calling `get_time` function with location: {location}")
``` return "get_time"
def get_news(category: str, country: str) -> str:
"""Useful to get the news in a specific country"""
print(
f"Calling `get_news` function with category: {category} and country: {country}"
)
return "get_news"
# Registering functions to the router
route_get_time = generate_route(get_time)
route_get_news = generate_route(get_news)
routes = [route_get_time, route_get_news]
router = create_router(routes)
%% Output # Tools
tools = [get_time, get_news]
```
2023-12-15 11:41:54 INFO semantic_router.utils.logger Extracting parameters... %% Cell type:markdown id: tags:
2023-12-15 11:41:54 INFO semantic_router.utils.logger Calling Mistral model
2023-12-15 11:41:55 INFO semantic_router.utils.logger AI message:
{
'location': 'Stockholm'
}
2023-12-15 11:41:55 INFO semantic_router.utils.logger Extracted parameters: {'location': 'Stockholm'}
parameters: {'location': 'Stockholm'}
Calling `get_time` function with location: Stockholm
2023-12-15 11:41:55 INFO semantic_router.utils.logger Extracting parameters...
2023-12-15 11:41:55 INFO semantic_router.utils.logger Calling Mistral model
2023-12-15 11:41:56 INFO semantic_router.utils.logger AI message:
{
'category': 'tech',
'country': 'Lithuania'
}
2023-12-15 11:41:56 INFO semantic_router.utils.logger Extracted parameters: {'category': 'tech', 'country': 'Lithuania'}
parameters: {'category': 'tech', 'country': 'Lithuania'}
Calling `get_news` function with category: tech and country: Lithuania
2023-12-15 11:41:57 WARNING semantic_router.utils.logger No function found
2023-12-15 11:41:57 INFO semantic_router.utils.logger Calling Mistral model
2023-12-15 11:41:57 INFO semantic_router.utils.logger AI message: How can I help you today?
' How can I help you today?' call(query="What is the time in Stockholm?", functions=tools, router=router)
call(query="What is the tech news in the Lithuania?", functions=tools, router=router)
call(query="Hi!", functions=tools, router=router)
......
pyproject.toml
+ 1
− 1
View file @ bfbf10bf
[tool.poetry] [tool.poetry]
name = "semantic-router" name = "semantic-router"
version = "0.0.9" version = "0.0.10"
description = "Super fast semantic router for AI decision making" description = "Super fast semantic router for AI decision making"
authors = [ authors = [
"James Briggs <james@aurelio.ai>", "James Briggs <james@aurelio.ai>",
......
semantic_router/encoders/openai.py
+ 1
− 1
View file @ bfbf10bf
...@@ -36,7 +36,7 @@ class OpenAIEncoder(BaseEncoder): ...@@ -36,7 +36,7 @@ class OpenAIEncoder(BaseEncoder):
try: try:
logger.info(f"Encoding {len(docs)} documents...") logger.info(f"Encoding {len(docs)} documents...")
embeds = self.client.embeddings.create(input=docs, model=self.name) embeds = self.client.embeddings.create(input=docs, model=self.name)
if isinstance(embeds, dict) and "data" in embeds: if "data" in embeds:
break break
except OpenAIError as e: except OpenAIError as e:
sleep(2**j) sleep(2**j)
......
tests/unit/encoders/test_openai.py
+ 1
− 0
View file @ bfbf10bf
...@@ -71,6 +71,7 @@ class TestOpenAIEncoder: ...@@ -71,6 +71,7 @@ class TestOpenAIEncoder:
) )
with pytest.raises(ValueError) as e: with pytest.raises(ValueError) as e:
openai_encoder(["test document"]) openai_encoder(["test document"])
assert "OpenAI API call failed. Error: Non-OpenAIError" in str(e.value) assert "OpenAI API call failed. Error: Non-OpenAIError" in str(e.value)
def test_openai_encoder_call_successful_retry(self, openai_encoder, mocker): def test_openai_encoder_call_successful_retry(self, openai_encoder, mocker):
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment

:feet::heart: Strive to be the person your dogs believe you are :heart::feet: