tu2022

关于pretraining时denoising的问题

3

我看bart的论文在pretraining的时候会有五种denoising的方法，在bart_dataset.py中我看insert_ratio和rotate_ratio是设为0，似乎不能将其设为大于0的数，是否意味着不能进行text infilling和rotation？

Compare the inference speed of quantized model and unquantized model

4

I have tested the inference speed of quantized model and unquantized model, which is first finetuned by my own dataset. I used **AutoAWQForCausalLM.from_quantized(quant_path, fuse_layers=True, max_seq_len=4096, device_map={"":"cuda:0"}, use_exllama_v2=True)** to load quantized...

I want to train a llama2 7b model, which is finedtuned by my own dataset. I don't know which dataset should i use while training the medusa heads? The dataset...