Would you ever consider adding fields to control the parameters ( like temp, top_p, etc ) for the model for hosted versions like OpenRouter and DeepInfra as well as local hosted solutions? Looking through the docs, this appears to be part of the base OpenAI API specs.
{
"model":"shuyuej/Mistral-Nemo-Instruct-2407-GPTQ",
"temperature": "some temp",
"top_p": "etc",
"min_p": "etc",
"messages":[ ... ],
"max_tokens":1024,"stream":true,"stop":["\n"] }
https://platform.openai.com/docs/api-reference/completions/create
https://openrouter.ai/docs/api-reference/overview
https://lmstudio.ai/docs/app/api/endpoints/openai
( Deep Infra API docs require going to deployment to view otherwise I would list them here too)
I was able to replicate the POST requests your game makes with some added fields and it didn't immediately fail though the output was interesting.
I'm not sure what effects this would have on Oobabooga but I would hope it wouldn't break anything if its following the API specs
Asking since DeepInfra and OpenRouter AI only seem to support setting these through the API and some models might benefit from fine tuning these parameters.