Recently, the hot word auto-gpt, which is popular all over the Internet, appears frequently on the entire website:
"ChatGPT is outdated, Auto-GPT is the future"
"Its ability is mainly an "autonomous", the kind that does not require human intervention at all!"
What exactly is auto-gpt?
1. What is the difference between Auto-GPT and ChatGPT
Both Torantulino/Auto-GPT and ChatGPT are neural network models based on the GPT model, but there are some essential differences between them:
- Different goals: The goal of Auto-GPT is to use an automated search algorithm to find the optimal pre-trained neural network structure to improve the performance of language modeling tasks. The goal of ChatGPT is to fine-tune chat scenarios to generate more natural and smooth conversations.
- The data sets are different: Auto-GPT uses a large general text corpus for pre-training, while ChatGPT uses a corpus collected specifically for chat conversations for fine-tuning training.
- Different hyperparameter settings: Auto-GPT uses a larger model size and a more complex search algorithm, requiring more computing resources and time for training. The ChatGPT model is relatively small and can be trained on a general computer.
- Different developers: Torantulino/Auto-GPT is a Python package developed by the community, while ChatGPT is a neural network model developed by OpenAI.
Therefore, although Torantulino/Auto-GPT and ChatGPT are variants based on the GPT model, their application scenarios, data sets, training methods, and hyperparameters are very different, and it is necessary to select the appropriate model according to specific needs.
In general
That is, chatgpt is an AI in the form of chat, and auto-gpt is an AI in an autonomous form (more automatic reasoning, planning and execution ). You can give a role and a goal
After it is given, it will make a plan
Then call the google query
Analyze the results
For example, a netizen asked AutoGPT to develop a website. As a result, in less than 3 minutes, AI used React and Tailwind CSS to "swish, swipe and swipe" to get it done.
2. github related projects
According to news on April 13, recently, the code hosting platform GitHub launched a new GPT-4-based open source application project AutoGPT, which exploded in the developer circle with more than 51k Stars.
Project source code address:
3. Build
Let's experience the construction of auto-gpt
Require:
- Python 3.8 or later
- OpenAI API secret
installation steps
To install Auto-GPT, follow these steps:
-
Make sure you have all the above requirements, if not, install them.
-
Clone the repository: For this step, you'll need Git installed, but you can download the zip file by clicking the button at the top of the page ☝️
git clone github.com/Torantulino…
-
Navigate to the project directory: (Type your target in the CMD window to navigate to the repository you just downloaded)
cd Auto-GPT
-
Install the required dependencies:
pip install -r requirements.txt
-
Rename .env.template to .env and fill in OPENAI_API_KEY. If you plan to use voice mode, please also fill in ELEVEN_LABS_API_KEY.
-
Get the OpenAI API key from: https://platform.openai.com/account/api-keys.
#.env.template .env is a hidden file, ll ls can’t see it, but it exists cp .env.template .env
Then fill in your openai key and save it.
-
To start, run the Python script in the main.py terminal:
python scripts/main.py
At this point, the installation is complete and you can experience it:
At the beginning, you will be asked to enter the external name and role, just whatever you want, and then Input: Let you enter
After entering the question, Thinking...
The above is its answer.
If after installation, start reporting this error:
Low priority: ARGUMENTS = 'dict' object has no attribute 'replace' :) #1085
You can add the following code on line 372 of the code main.py (the number of lines of code varies in different versions), and pay attention to the ** indentation of the python code. **Save and restart.
try:
json.loads(assistant_reply)
except ValueError as e:
if assistant_reply.count('{') == assistant_reply.count('}') == 0:
# remove " and '
assistant_reply = assistant_reply.replace('"', '').replace("'", '')
assistant_reply = '{' \
'"thoughts": {' \
'"text": "' + assistant_reply + '",' \
'"reasoning": "",' \
'"plan": "",' \
'"criticism": "",' \
'"speak": ""' \
'},' \
'"command": {' \
'"name": "do_nothing", "args": {}' \
'}' \
'}'
elif assistant_reply.count('{') == assistant_reply.count('}'):
# remove everything before the first { and after the last }
assistant_reply = assistant_reply[assistant_reply.find('{'):assistant_reply.rfind('}') + 1]
else:
while assistant_reply.count('{') != assistant_reply.count('}'):
if assistant_reply.count('{') > assistant_reply.count('}'):
# add a } to the end
assistant_reply = assistant_reply + '}'
else:
# add a { to the beginning
assistant_reply = '{' + assistant_reply
复制代码If the number of lines is not accurate, you can search for keywords
Just add "# Print Assistant thoughts" to it, and the final effect is similar to the code in the blue frame. That is to add above this line of code
4. Other issues
A lot has been said above, and I have already experienced auto-gpt. The gpt-4 that auto-gpt said,
Are we using gpt-3 or gpt-4?
It can be found that the openai key we use is still gpt-3, so how to apply for gpt-4? There are two ways online, one is to become a gpt plus member, and the other is to apply for a gpt-4 api waitlist.
ps Some people on the Internet say that you must become a plus before you can apply for a waitlist. However, after checking some information, it seems that there is no such requirement. It just says that if you want to use the gpt4 api, you must also apply for a waitlist. Anyway, try to apply first. . . .
at last
If my article is helpful to you, please like + follow! ! !