AlexBuz/llama-zip

Issues

• is it possible to use groq (llama 3.1 70b) api to compress text instead of running your own model?

  #20 opened 3 months ago by sprappcom
  3

• how to make n_gpu_layers work with cuda?

  #19 opened 4 months ago by sprappcom
  1

• ever tried tinyllama or smaller llamas?

  #16 opened 4 months ago by sprappcom
  1

• possible to provide benchmark for these?

  #18 opened 4 months ago by sprappcom
  1

• generate binary (for better compression) instead of text

  #17 opened 4 months ago by sprappcom
  2

• Does not work on all files because of utf-8 error

  #13 opened 5 months ago by secemp9
  4

• Using different models like Phi-3

  #5 opened 5 months ago by CyberTimon
  7

• ollama version?

  #14 opened 5 months ago by sokoow
  2

• Post compression ratios

  #3 opened 5 months ago by lee-b
  15

• Request: Add newest cmix compressor to comparison table.

  #12 opened 5 months ago by dillfrescott
  1

• compare with brotli

  #11 opened 5 months ago by jyrkialakuijala
  1

• Interesting side effect of decompression - original training data extraction

  #10 opened 5 months ago by bigattichouse
  5

• Using the colab in the repository, we can notice it does not use the GPU

  #8 opened 5 months ago by secemp9
  1

• Interactive mode generate text instead of compressing when using specific text

  #9 opened 5 months ago by secemp9
  2

• Question

  #7 opened 5 months ago by dillfrescott
  2

• Perform compression in batches for texts exceeding the 8192 token limit of llama3.

  #1 opened 5 months ago by dillfrescott
  6

• Gibberish produced on 1 word spaces

  #2 opened 5 months ago by P3GLEG
  2