Hướng dẫn làm việc với các phiên bản GPT-3.5-Turbo và GPT-4

{“role”: “system”, “content”: “Provide some context and/or instructions to the model”},

{“role”: “user”, “content”: “The users messages goes here”}

OpenAI Python 0.28.1

OpenAI Python 1.x

import os

import openai

openai.api_type = “azure”

openai.api_version = “2023-05-15” 

openai.api_base = os.getenv(“AZURE_OPENAI_ENDPOINT”)  # Your Azure OpenAI resource’s endpoint value.

openai.api_key = os.getenv(“AZURE_OPENAI_KEY”)

conversation=[{“role”: “system”, “content”: “You are a helpful assistant.”}]

while True:

    user_input = input()      

    conversation.append({“role”: “user”, “content”: user_input})

    response = openai.ChatCompletion.create(

        engine=”gpt-35-turbo”, # The deployment name you chose when you deployed the GPT-35-turbo or GPT-4 model.

        messages=conversation

    )

    conversation.append({“role”: “assistant”, “content”: response[“choices”][0][“message”][“content”]})

    print(“\n” + response[‘choices’][0][‘message’][‘content’] + “\n”)

import os

from openai import AzureOpenAI

client = AzureOpenAI(

  api_key = os.getenv(“AZURE_OPENAI_KEY”),  

  api_version = “2023-05-15”,

  azure_endpoint = os.getenv(“AZURE_OPENAI_ENDPOINT”)  # Your Azure OpenAI resource’s endpoint value.

)

conversation=[{“role”: “system”, “content”: “You are a helpful assistant.”}]

while True:

    user_input = input(“Q:”)      

    conversation.append({“role”: “user”, “content”: user_input})

    response = client.chat.completions.create(

        model=”gpt-35-turbo”, # model = “deployment_name”.

        messages=conversation

    )

    conversation.append({“role”: “assistant”, “content”: response.choices[0].message.content})

    print(“\n” + response.choices[0].message.content + “\n”)

OpenAI Python 0.28.1

OpenAI Python 1.x

import tiktoken

import openai

import os

openai.api_type = “azure”

openai.api_version = “2023-05-15” 

openai.api_base = os.getenv(“AZURE_OPENAI_ENDPOINT”)  # Your Azure OpenAI resource’s endpoint value.

openai.api_key = os.getenv(“AZURE_OPENAI_KEY”)

system_message = {“role”: “system”, “content”: “You are a helpful assistant.”}

max_response_tokens = 250

token_limit = 4096

conversation = []

conversation.append(system_message)

def num_tokens_from_messages(messages, model=”gpt-3.5-turbo-0613″):

    “””Return the number of tokens used by a list of messages.”””

    try:

        encoding = tiktoken.encoding_for_model(model)

    except KeyError:

        print(“Warning: model not found. Using cl100k_base encoding.”)

        encoding = tiktoken.get_encoding(“cl100k_base”)

    if model in {

        “gpt-3.5-turbo-0613”,

        “gpt-3.5-turbo-16k-0613”,

        “gpt-4-0314”,

        “gpt-4-32k-0314”,

        “gpt-4-0613”,

        “gpt-4-32k-0613”,

        }:

        tokens_per_message = 3

        tokens_per_name = 1

    elif model == “gpt-3.5-turbo-0301”:

        tokens_per_message = 4  # every message follows <|start|>{role/name}\n{content}<|end|>\n

        tokens_per_name = -1  # if there’s a name, the role is omitted

    elif “gpt-3.5-turbo” in model:

        print(“Warning: gpt-3.5-turbo may update over time. Returning num tokens assuming gpt-3.5-turbo-0613.”)

        return num_tokens_from_messages(messages, model=”gpt-3.5-turbo-0613″)

    elif “gpt-4” in model:

        print(“Warning: gpt-4 may update over time. Returning num tokens assuming gpt-4-0613.”)

        return num_tokens_from_messages(messages, model=”gpt-4-0613″)

    else:

        raise NotImplementedError(

            f”””num_tokens_from_messages() is not implemented for model {model}. See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.”””

        )

    num_tokens = 0

    for message in messages:

        num_tokens += tokens_per_message

        for key, value in message.items():

            num_tokens += len(encoding.encode(value))

            if key == “name”:

                num_tokens += tokens_per_name

    num_tokens += 3  # every reply is primed with <|start|>assistant<|message|>

    return num_tokens

while True:

    user_input = input(“”)     

    conversation.append({“role”: “user”, “content”: user_input})

    conv_history_tokens = num_tokens_from_messages(conversation)

    while conv_history_tokens + max_response_tokens >= token_limit:

        del conversation[1] 

        conv_history_tokens = num_tokens_from_messages(conversation)

    response = openai.ChatCompletion.create(

        engine=”gpt-35-turbo”, # The deployment name you chose when you deployed the GPT-35-Turbo or GPT-4 model.

        messages=conversation,

        temperature=0.7,

        max_tokens=max_response_tokens,

    )

    conversation.append({“role”: “assistant”, “content”: response[‘choices’][0][‘message’][‘content’]})

    print(“\n” + response[‘choices’][0][‘message’][‘content’] + “\n”)

import tiktoken

import os

from openai import AzureOpenAI

client = AzureOpenAI(

  api_key = os.getenv(“AZURE_OPENAI_KEY”),  

  api_version = “2023-05-15”,

  azure_endpoint = os.getenv(“AZURE_OPENAI_ENDPOINT”)  # Your Azure OpenAI resource’s endpoint value.

)

system_message = {“role”: “system”, “content”: “You are a helpful assistant.”}

max_response_tokens = 250

token_limit = 4096

conversation = []

conversation.append(system_message)

def num_tokens_from_messages(messages, model=”gpt-3.5-turbo-0613″):

    “””Return the number of tokens used by a list of messages.”””

    try:

        encoding = tiktoken.encoding_for_model(model)

    except KeyError:

        print(“Warning: model not found. Using cl100k_base encoding.”)

        encoding = tiktoken.get_encoding(“cl100k_base”)

    if model in {

        “gpt-3.5-turbo-0613”,

        “gpt-3.5-turbo-16k-0613”,

        “gpt-4-0314”,

        “gpt-4-32k-0314”,

        “gpt-4-0613”,

        “gpt-4-32k-0613”,

        }:

        tokens_per_message = 3

        tokens_per_name = 1

    elif model == “gpt-3.5-turbo-0301”:

        tokens_per_message = 4  # every message follows <|start|>{role/name}\n{content}<|end|>\n

        tokens_per_name = -1  # if there’s a name, the role is omitted

    elif “gpt-3.5-turbo” in model:

        print(“Warning: gpt-3.5-turbo may update over time. Returning num tokens assuming gpt-3.5-turbo-0613.”)

        return num_tokens_from_messages(messages, model=”gpt-3.5-turbo-0613″)

    elif “gpt-4” in model:

        print(“Warning: gpt-4 may update over time. Returning num tokens assuming gpt-4-0613.”)

        return num_tokens_from_messages(messages, model=”gpt-4-0613″)

    else:

        raise NotImplementedError(

            f”””num_tokens_from_messages() is not implemented for model {model}. See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.”””

        )

    num_tokens = 0

    for message in messages:

        num_tokens += tokens_per_message

        for key, value in message.items():

            num_tokens += len(encoding.encode(value))

            if key == “name”:

                num_tokens += tokens_per_name

    num_tokens += 3  # every reply is primed with <|start|>assistant<|message|>

    return num_tokens

while True:

    user_input = input(“Q:”)      

    conversation.append({“role”: “user”, “content”: user_input})

    conv_history_tokens = num_tokens_from_messages(conversation)

    while conv_history_tokens + max_response_tokens >= token_limit:

        del conversation[1] 

        conv_history_tokens = num_tokens_from_messages(conversation)

    response = client.chat.completions.create(

        model=”gpt-35-turbo”, # model = “deployment_name”.

        messages=conversation,

        temperature=0.7,

        max_tokens=max_response_tokens

    )

    conversation.append({“role”: “assistant”, “content”: response.choices[0].message.content})

    print(“\n” + response.choices[0].message.content + “\n”)

<|im_start|>system 

Provide some context and/or instructions to the model.

<|im_end|> 

<|im_start|>user 

The user’s message goes here

<|im_end|> 

<|im_start|>assistant

import tiktoken 

cl100k_base = tiktoken.get_encoding(“cl100k_base”) 

enc = tiktoken.Encoding( 

    name=”gpt-35-turbo”,  

    pat_str=cl100k_base._pat_str, 

    mergeable_ranks=cl100k_base._mergeable_ranks, 

    special_tokens={ 

        **cl100k_base._special_tokens, 

        “<|im_start|>”: 100264, 

        “<|im_end|>”: 100265

    } 

) 

tokens = enc.encode( 

    “<|im_start|>user\nHello<|im_end|><|im_start|>assistant”,  

    allowed_special={“<|im_start|>”, “<|im_end|>”} 

) 

assert len(tokens) == 7 

assert tokens == [100264, 882, 198, 9906, 100265, 100264, 78191]