Making the model that runs GPT Pilot have no cost. about gpt-pilot

[KevinGeLe]

If the licence from the models, allows it to be integrated to gpt-pilot. Then we could simple have an option under .env, named "local model=" and have it setup automatically.

The thing is, your project is not the same as mine. If the model is too creative the temperature needs to be lowered and wiseversa. Also the models you talk about, have been trained to mimic openai's models, instead of making the dataset bigger. So alot of data that the average user doesnt need, simply isnt there. That means low quality code (if they can even do that.)

So we need bigger models? Then we also need to make sure the model can run on that specific hardware. Soooo, make a benchmark, get score and then download the appropriate model? Seems simple. Except that its hard to get a specific benchmark from the hardware, as hardware generation, cpu, gpu, ram, ai acceleration etc. Plays a huge role.

Great idea. But unnecessary when gpt-pilot itself needs different instruction per model.

Kevin

[takimdigital]

Try this

https://github.com/tgscan-dev/g4f-api