overview for kristaller486

Yi-23B-Llama: Distil version of Yi-34B-Llama in c/localllama@poweruser.forum

[–] kristaller486@alien.top 1 points 11 months ago (2 children)

Is there a code for distillation?

Good and fast model around ~1B to run on web? in c/localllama@poweruser.forum

[–] kristaller486@alien.top 1 points 1 year ago

You can try MLC-LLM (https://llm.mlc.ai/), it has tools for inference of quantized models on the web

We are Higgsfield AI. We have a large GPU cluster and want to finetune your dataset. in c/localllama@poweruser.forum

[–] kristaller486@alien.top 1 points 1 year ago

This is awesome! What training parameters are you using?

Finetuning LLMs: Does it add new knowledge to model or not? in c/localllama@poweruser.forum

[–] kristaller486@alien.top 1 points 1 year ago (1 children)

Full-weight fine-tuning should add new knowledge. Not LoRa.

PHIND V7: Red Flags in c/localllama@poweruser.forum

[–] kristaller486@alien.top 1 points 1 year ago (1 children)

They trained their model using synthetic GPT-3.5-turbo data + a mix of their data. It is normal that V7 says "I am gpt-3.5", but it is not normal that Phind uses synthetic OpenAI GPT data because it violates OpenAI terms.