cognitivecomputations

/

dolphin-2.5-mixtral-8x7b

like 1.17k

Thanks for this project! I noticed that the model was trained with QLoRA, but the fine tuned weights are provided as full (merged) weights rather than the LoRA weights. Do you happen to have the LoRA weights available somewhere to share?

For context, I'm interested in serving this model using LoRAX (https://github.com/predibase/lorax) so the base model and fine-tuned model can be served at the sam time.