make it easier to see output scene

[kfarr]

As a user making a text-to-street request via the python console running get_streetmix_json function inside texttostreet.py, when a response is provided I should be able to see the scene output after a few seconds.

ideas:

1. have an index.html with copy/paste destination clearly marked with comment. copy/paste the json output from the python function console output.
2. create 3dstreet json that can be downloaded, copy / pasted into a new file, or a new file is created in local filesystem that includes the new street json from open ai response as well as the original prompt in the title field.

[kfarr]

ticket for @Algorush but I can't seem to assign this

[Algorush]

@kfarr hi, what if I create simple flask server with form.html to send user queries to text-to-street.py, that will opened new tab with 3dstreet scene? Is this ok for first version?

[Algorush]

But I also created index-bot.html for option 1

[kfarr]

here is the ideal case for demo purposes:

1. a request is sent from python command line (done)
2. the response from chatgpt is shown streaming in console (done, see Algorush@e4303d2)
3. upon json being received, it opens up a new browser tab with the url of the generated scene -- needs help

I could use help with #3

You mentioned that you already made an index.html that can copy / paste?

Maybe we can use that, and the python command-line script generates a url with the generated JSON passed directly in the URL as URL encoded data such as localhost:7001/index-bot/#streetmix-json:[URI ENCODED STREETMIX JSON HERE]

[Algorush]

You mentioned that you already made an index.html that can copy / paste?

Yes, I slightly changed the code of the streetmix-loader component so that I could generate a scene from the ready-made json with streetmix format

[kfarr]

@Algorush great could you share that or use it do to the above idea of passing the json directly as URI data?

[Algorush]

Yes, now I’m doing just this to transmit json as URI data.
I just made a version yesterday with a simple flask server that returns generated json upon request. But it still needs to be checked first, validated before using it to generate a street. Because GPT sometimes returns invalid data in json (for example, the type of buildings that is not in the enum array).
I came up with this option - if the generated segment type or variant is not in the json schema, then use the default value from the schema

[kfarr]

Yes that makes sense. Can you check out the streaming demo version? It’s here https://github.com/3DStreet/text-to-street/blob/e4303d2af5edc90fb2a2fc31ce2ff0e687e4bc43/functions/texttostreetdemo.py#L53 I couldn’t get functions working with streaming so instead it parses the streamed output to look for a json snippet

…

On Mon, Sep 25, 2023 at 11:35 AM Alexander Goryushkin < ***@***.***> wrote: Yes, now I’m doing just this to transmit json as URI data. I just made a version yesterday with a simple flask server that returns generated json upon request. But it still needs to be checked first, validated before using it to generate a street. Because GPT sometimes returns invalid data in json (for example, the type of buildings that is not in the enum array). I came up with this option - if the generated segment type or variant is not in the json schema, then use the default value from the schema — Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AADS3TLNGU25MDYTNCF4KITX4HFGLANCNFSM6AAAAAA46ZNKM4> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[Algorush]

I found this (how to run function_call with streaming): https://community.openai.com/t/functions-calling-with-streaming/305742

[Algorush]

Judging by what I read about function_call, it is still better to use them when generating JSON, with them the structure of the model response is more accurate (in accordance with schema)