mirror of https://github.com/hwchase17/langchain
Create ChatEverlyAI (#11357)
- Description: Adds the ChatEverlyAI class with llama-2 7b on [EverlyAI Hosted Endpoints](https://everlyai.xyz/) - It inherits from ChatOpenAI and requires openai (probably unnecessary but it made for a quick and easy implementation) --------- Co-authored-by: everly-studio <127131037+everly-studio@users.noreply.github.com>pull/11779/head
parent
7fb25b4154
commit
9d200e6cbe
@ -0,0 +1,214 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "642fd21c-600a-47a1-be96-6e1438b421a9",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"# EverlyAI\n",
|
||||
"\n",
|
||||
">[EverlyAI](https://everlyai.xyz) allows you to run your ML models at scale in the cloud. It also provides API access to [several LLM models](https://everlyai.xyz).\n",
|
||||
"\n",
|
||||
"This notebook demonstrates the use of `langchain.chat_models.ChatEverlyAI` for [EverlyAI Hosted Endpoints](https://everlyai.xyz/).\n",
|
||||
"\n",
|
||||
"* Set `EVERLYAI_API_KEY` environment variable\n",
|
||||
"* or use the `everlyai_api_key` keyword argument"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "d00d850917865298",
|
||||
"metadata": {
|
||||
"collapsed": false
|
||||
},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# !pip install openai"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 1,
|
||||
"id": "72340871-ae2f-415f-b399-0777d32dc379",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"import os\n",
|
||||
"from getpass import getpass\n",
|
||||
"\n",
|
||||
"os.environ[\"EVERLYAI_API_KEY\"] = getpass()"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "5d7fc704-3ea0-4c35-96e7-89fcae6c73fa",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"# Let's try out LLAMA model offered on EverlyAI Hosted Endpoints"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 2,
|
||||
"id": "0dc9428d-4217-47d2-97de-f784b1764186",
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
" Hello! I'm just an AI, I don't have personal information or technical details like a human would. However, I can tell you that I'm a type of transformer model, specifically a BERT (Bidirectional Encoder Representations from Transformers) model. B\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"from langchain.chat_models import ChatEverlyAI\n",
|
||||
"from langchain.schema import SystemMessage, HumanMessage\n",
|
||||
"\n",
|
||||
"messages = [\n",
|
||||
" SystemMessage(\n",
|
||||
" content=\"You are a helpful AI that shares everything you know.\"\n",
|
||||
" ),\n",
|
||||
" HumanMessage(\n",
|
||||
" content=\"Tell me technical facts about yourself. Are you a transformer model? How many billions of parameters do you have?\"\n",
|
||||
" ),\n",
|
||||
"]\n",
|
||||
"\n",
|
||||
"chat = ChatEverlyAI(model_name=\"meta-llama/Llama-2-7b-chat-hf\", temperature=0.3, max_tokens=64)\n",
|
||||
"print(chat(messages).content)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "7c4f124a-eaf7-4d78-a2c0-b0aa23fb25c4",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"# EverlyAI also supports streaming responses"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 3,
|
||||
"id": "1f94f5d2-569e-4a2c-965e-de53c2845fbb",
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
" Ah, a joke, you say? *adjusts glasses* Well, I've got a doozy for you! *winks*\n",
|
||||
" *pauses for dramatic effect*\n",
|
||||
"Why did the AI go to therapy?\n",
|
||||
"*drumroll*\n",
|
||||
"Because"
|
||||
]
|
||||
},
|
||||
{
|
||||
"data": {
|
||||
"text/plain": [
|
||||
"AIMessageChunk(content=\" Ah, a joke, you say? *adjusts glasses* Well, I've got a doozy for you! *winks*\\n *pauses for dramatic effect*\\nWhy did the AI go to therapy?\\n*drumroll*\\nBecause\")"
|
||||
]
|
||||
},
|
||||
"execution_count": 3,
|
||||
"metadata": {},
|
||||
"output_type": "execute_result"
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"from langchain.chat_models import ChatEverlyAI\n",
|
||||
"from langchain.schema import SystemMessage, HumanMessage\n",
|
||||
"from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler\n",
|
||||
"\n",
|
||||
"messages = [\n",
|
||||
" SystemMessage(\n",
|
||||
" content=\"You are a humorous AI that delights people.\"\n",
|
||||
" ),\n",
|
||||
" HumanMessage(\n",
|
||||
" content=\"Tell me a joke?\"\n",
|
||||
" ),\n",
|
||||
"]\n",
|
||||
"\n",
|
||||
"chat = ChatEverlyAI(model_name=\"meta-llama/Llama-2-7b-chat-hf\", temperature=0.3, max_tokens=64, streaming=True, callbacks=[StreamingStdOutCallbackHandler()])\n",
|
||||
"chat(messages)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "7de56d98",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"# Let's try a different language model on EverlyAI"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 4,
|
||||
"id": "d8a44114",
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
" OH HO HO! *adjusts monocle* Well, well, well! Look who's here! *winks*\n",
|
||||
"\n",
|
||||
"You want a joke, huh? *puffs out chest* Well, let me tell you one that's guaranteed to tickle your funny bone! *clears throat*\n",
|
||||
"\n",
|
||||
"Why couldn't the bicycle stand up by itself? *pauses for dramatic effect* Because it was two-tired! *winks*\n",
|
||||
"\n",
|
||||
"Hope that one put a spring in your step, my dear! *"
|
||||
]
|
||||
},
|
||||
{
|
||||
"data": {
|
||||
"text/plain": [
|
||||
"AIMessageChunk(content=\" OH HO HO! *adjusts monocle* Well, well, well! Look who's here! *winks*\\n\\nYou want a joke, huh? *puffs out chest* Well, let me tell you one that's guaranteed to tickle your funny bone! *clears throat*\\n\\nWhy couldn't the bicycle stand up by itself? *pauses for dramatic effect* Because it was two-tired! *winks*\\n\\nHope that one put a spring in your step, my dear! *\")"
|
||||
]
|
||||
},
|
||||
"execution_count": 4,
|
||||
"metadata": {},
|
||||
"output_type": "execute_result"
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"from langchain.chat_models import ChatEverlyAI\n",
|
||||
"from langchain.schema import SystemMessage, HumanMessage\n",
|
||||
"from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler\n",
|
||||
"\n",
|
||||
"messages = [\n",
|
||||
" SystemMessage(\n",
|
||||
" content=\"You are a humorous AI that delights people.\"\n",
|
||||
" ),\n",
|
||||
" HumanMessage(\n",
|
||||
" content=\"Tell me a joke?\"\n",
|
||||
" ),\n",
|
||||
"]\n",
|
||||
"\n",
|
||||
"chat = ChatEverlyAI(model_name=\"meta-llama/Llama-2-13b-chat-hf-quantized\", temperature=0.3, max_tokens=128, streaming=True, callbacks=[StreamingStdOutCallbackHandler()])\n",
|
||||
"chat(messages)"
|
||||
]
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3 (ipykernel)",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
"version": 3
|
||||
},
|
||||
"file_extension": ".py",
|
||||
"mimetype": "text/x-python",
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.10.1"
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 5
|
||||
}
|
@ -0,0 +1,154 @@
|
||||
"""EverlyAI Endpoints chat wrapper. Relies heavily on ChatOpenAI."""
|
||||
from __future__ import annotations
|
||||
|
||||
import logging
|
||||
import sys
|
||||
from typing import TYPE_CHECKING, Dict, Optional, Set
|
||||
|
||||
from langchain.adapters.openai import convert_message_to_dict
|
||||
from langchain.chat_models.openai import (
|
||||
ChatOpenAI,
|
||||
_import_tiktoken,
|
||||
)
|
||||
from langchain.pydantic_v1 import Field, root_validator
|
||||
from langchain.schema.messages import BaseMessage
|
||||
from langchain.utils import get_from_dict_or_env
|
||||
|
||||
if TYPE_CHECKING:
|
||||
import tiktoken
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
DEFAULT_API_BASE = "https://everlyai.xyz/hosted"
|
||||
DEFAULT_MODEL = "meta-llama/Llama-2-7b-chat-hf"
|
||||
|
||||
|
||||
class ChatEverlyAI(ChatOpenAI):
|
||||
"""`EverlyAI` Chat large language models.
|
||||
|
||||
To use, you should have the ``openai`` python package installed, and the
|
||||
environment variable ``EVERLYAI_API_KEY`` set with your API key.
|
||||
Alternatively, you can use the everlyai_api_key keyword argument.
|
||||
|
||||
Any parameters that are valid to be passed to the `openai.create` call can be passed
|
||||
in, even if not explicitly saved on this class.
|
||||
|
||||
Example:
|
||||
.. code-block:: python
|
||||
|
||||
from langchain.chat_models import ChatEverlyAI
|
||||
chat = ChatEverlyAI(model_name="meta-llama/Llama-2-7b-chat-hf")
|
||||
"""
|
||||
|
||||
@property
|
||||
def _llm_type(self) -> str:
|
||||
"""Return type of chat model."""
|
||||
return "everlyai-chat"
|
||||
|
||||
@property
|
||||
def lc_secrets(self) -> Dict[str, str]:
|
||||
return {"everlyai_api_key": "EVERLYAI_API_KEY"}
|
||||
|
||||
everlyai_api_key: Optional[str] = None
|
||||
"""EverlyAI Endpoints API keys."""
|
||||
model_name: str = Field(default=DEFAULT_MODEL, alias="model")
|
||||
"""Model name to use."""
|
||||
everlyai_api_base: str = DEFAULT_API_BASE
|
||||
"""Base URL path for API requests."""
|
||||
available_models: Optional[Set[str]] = None
|
||||
"""Available models from EverlyAI API."""
|
||||
|
||||
@staticmethod
|
||||
def get_available_models() -> Set[str]:
|
||||
"""Get available models from EverlyAI API."""
|
||||
# EverlyAI doesn't yet support dynamically query for available models.
|
||||
return set(
|
||||
[
|
||||
"meta-llama/Llama-2-7b-chat-hf",
|
||||
"meta-llama/Llama-2-13b-chat-hf-quantized",
|
||||
]
|
||||
)
|
||||
|
||||
@root_validator(pre=True)
|
||||
def validate_environment_override(cls, values: dict) -> dict:
|
||||
"""Validate that api key and python package exists in environment."""
|
||||
values["openai_api_key"] = get_from_dict_or_env(
|
||||
values,
|
||||
"everlyai_api_key",
|
||||
"EVERLYAI_API_KEY",
|
||||
)
|
||||
values["openai_api_base"] = DEFAULT_API_BASE
|
||||
|
||||
try:
|
||||
import openai
|
||||
|
||||
except ImportError as e:
|
||||
raise ValueError(
|
||||
"Could not import openai python package. "
|
||||
"Please install it with `pip install openai`.",
|
||||
) from e
|
||||
try:
|
||||
values["client"] = openai.ChatCompletion
|
||||
except AttributeError as exc:
|
||||
raise ValueError(
|
||||
"`openai` has no `ChatCompletion` attribute, this is likely "
|
||||
"due to an old version of the openai package. Try upgrading it "
|
||||
"with `pip install --upgrade openai`.",
|
||||
) from exc
|
||||
|
||||
if "model_name" not in values.keys():
|
||||
values["model_name"] = DEFAULT_MODEL
|
||||
|
||||
model_name = values["model_name"]
|
||||
|
||||
available_models = cls.get_available_models()
|
||||
|
||||
if model_name not in available_models:
|
||||
raise ValueError(
|
||||
f"Model name {model_name} not found in available models: "
|
||||
f"{available_models}.",
|
||||
)
|
||||
|
||||
values["available_models"] = available_models
|
||||
|
||||
return values
|
||||
|
||||
def _get_encoding_model(self) -> tuple[str, tiktoken.Encoding]:
|
||||
tiktoken_ = _import_tiktoken()
|
||||
if self.tiktoken_model_name is not None:
|
||||
model = self.tiktoken_model_name
|
||||
else:
|
||||
model = self.model_name
|
||||
# Returns the number of tokens used by a list of messages.
|
||||
try:
|
||||
encoding = tiktoken_.encoding_for_model("gpt-3.5-turbo-0301")
|
||||
except KeyError:
|
||||
logger.warning("Warning: model not found. Using cl100k_base encoding.")
|
||||
model = "cl100k_base"
|
||||
encoding = tiktoken_.get_encoding(model)
|
||||
return model, encoding
|
||||
|
||||
def get_num_tokens_from_messages(self, messages: list[BaseMessage]) -> int:
|
||||
"""Calculate num tokens with tiktoken package.
|
||||
|
||||
Official documentation: https://github.com/openai/openai-cookbook/blob/
|
||||
main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb"""
|
||||
if sys.version_info[1] <= 7:
|
||||
return super().get_num_tokens_from_messages(messages)
|
||||
model, encoding = self._get_encoding_model()
|
||||
tokens_per_message = 3
|
||||
tokens_per_name = 1
|
||||
num_tokens = 0
|
||||
messages_dict = [convert_message_to_dict(m) for m in messages]
|
||||
for message in messages_dict:
|
||||
num_tokens += tokens_per_message
|
||||
for key, value in message.items():
|
||||
# Cast str(value) in case the message value is not a string
|
||||
# This occurs with function messages
|
||||
num_tokens += len(encoding.encode(str(value)))
|
||||
if key == "name":
|
||||
num_tokens += tokens_per_name
|
||||
# every reply is primed with <im_start>assistant
|
||||
num_tokens += 3
|
||||
return num_tokens
|
Loading…
Reference in New Issue