$ ata --config=.config/ata.toml                                                                                                                                                                                         ~
Ask the Terminal Anything

model: text-davinci-003
max_tokens: 500
temperature: 0

Prompt: 
Hi, who are you

Response: 


Prompt: 
Hi

Response: 


Prompt: 
test

Response: 


Prompt: 

This is going to be a bit of a strange pull request and I do apologize for that.

There are many very good reasons you may not want this massive patch.

Starting with the fact that I developed it entirely on my Android phone as the final project of a personal challenge to see how complicated of a patch I could actually write this way:

I consider the test over now because there's really nothing to do beyond this; all it does is just take longer but I feel like I've proven you can do everything. Also wrote several websites and patched ffmpeg.

Nevertheless, a lot of time did go into it. And I think my version is better because it has many features that yours or not:

• You can provide any API key/value pair as of 2023-03-02
• You can have multiple named configurations
• It does not do the very unsafe thing of recommending that you write configuration to the current directory
• You can provide no configuration at all and it will figure out where it could go and even offer to write you a skeleton.

Either way have fun 😊

Hi,

I downloaded ata-aarch64-apple-darwin and tried to run it in the Terminal and got -bash: /Users/plumps/Downloads/ata-aarch64-apple-darwin: Bad CPU type in executable as response.

Checking the file type with file, showed that:

file ~/Downloads/ata-aarch64-apple-darwin 
/Users/plumps/Downloads/ata-aarch64-apple-darwin: Mach-O 64-bit executable **x86_64**

message: "invalid type: floating point 0.5, expected i64"

It appears that setting a floating point temperature, such as 0.5 or 0.8, is not possible.

Thanks to @marioseixas. I'm a bit hesitant to set it to Inf because that could mean that the tool could in theory produce in an infinitely long response which would cause an infinitely high bill. Very unlikely, but who knows. Having 4096 tokens printed to the terminal is already pretty long.

• Closes #25.

Hello

congratz on this awesome tool

Please remove the max_tokens parameter, so it could default to (4096 - prompt tokens)

You should not set the max_token, according to the official openAI API, the default value is Inf, and the model can process up to 4096 tokens of information.

Support the newly released chat API (released 1st of March 2023): gpt-3.5-turbo. This model is better than the text models according to Greg Brockman. Also, gpt-3.5-turbo is 10 times cheaper than text-davinci-003 per token. Due to the increased complexity in supporting text models and chat models and me having only so much time in the day, this PR drops support for text models such as text-davinci-003. To support those again, tests should be added to the repository which mock both the OpenAI API and stdout (see https://github.com/kkawakam/rustyline/issues/652#issuecomment-1421314307 for a stdout mock example).

This PR is partially based on #21.

• Closes #19

Currently:

Ask the Terminal Anything

model: gpt-3.5-turbo-0301
max_tokens: 500
temperature: 0.8

Prompt:
hi there

Error:
{
  "error": {
    "message": "This is a chat model and not supported in the v1/completions endpoint. Did you mean to use v1/chat/completions?",
    "type": "invalid_request_error",
    "param": "model",
    "code": null
  }
}

This currently fails immediately:

Error:
{
  "error": {
    "message": "The server had an error while processing your request. Sorry about that!",
    "type": "server_error",
    "param": null,
    "code": null
  }
}

Given that this happens quite frequently in my experience and that it is solved by doing another request, ata should probably try again after a second

Hi

I think you did a mistake in your last commit.

Setting the max_tokens to 4096 will output always an error like this:

I said for you to remove every mention on the code about max_tokens or tokens parameters so that it defaults to Inf (4096 - prompt tokens). Inf is not infinite, Inf equals to 4096 - prompt tokens, so that no one needs to keep changing the ata.toml to reduce or to increase the max_tokens limit. It will automatically make the calculation:

Who explained that to me was the awesome dev Jack Wong on a issue in his incredible app: https://github.com/jw-12138/davinci-web/issues/8

This should be very much doable. To implement it, add an extra handler for CTRL + L and make that clear the message history. The handler was already added in https://github.com/rikhuijzer/ata/pull/23.

This would make it easier to add the binary to the bin folder, place the config in some default path, and then run

$ ata

without extra flags.

For now, this can be achieved by adding the following script to the bin folder (for example, ~/.local/bin):

#!/usr/bin/env bash

/path/to/ata/binary --config=/path/to/config/toml "$@"

Thanks to "$@", this will accept arbitrary arguments to ata.

When canceling a prompt with CTRL + C, it is very tricky to ensure that the stream doesn't end at the same time. If that happens, ata will shut down.

To solve this, there should be a timeout of a few seconds after the stream has finished in which CTRL + C is ignored.