llama icon indicating copy to clipboard operation
llama copied to clipboard

params.json: FAILED

Open aaronjolson opened this issue 1 year ago • 1 comments

Before submitting a bug, please make sure the issue hasn't been already addressed by searching through the FAQs and existing/past issues

Describe the bug

After downloading all of the model parts for 70b-Instruct and 70b-Python I am getting the following error

consolidated.01.pth: OK consolidated.02.pth: OK consolidated.03.pth: OK consolidated.04.pth: OK consolidated.05.pth: OK consolidated.06.pth: OK consolidated.07.pth: OK params.json: FAILED tokenizer.model: OK md5sum: WARNING: 1 line is improperly formatted md5sum: WARNING: 1 computed checksum did NOT match

Minimal reproducible example

This is the contents of params.json

{
    "dim": 8192,
    "n_heads": 64,
    "n_kv_heads": 8,
    "n_layers": 80,
    "multiple_of": 4096,
    "ffn_dim_multiplier": 1.3,
    "norm_eps": 1e-5,
    "rope_theta": 10000
}

Output

consolidated.01.pth: OK consolidated.02.pth: OK consolidated.03.pth: OK consolidated.04.pth: OK consolidated.05.pth: OK consolidated.06.pth: OK consolidated.07.pth: OK params.json: FAILED tokenizer.model: OK md5sum: WARNING: 1 line is improperly formatted md5sum: WARNING: 1 computed checksum did NOT match

Runtime Environment

  • Model: [CodeLlama-70b-Instruct]
  • Using via huggingface?: [no]
  • OS: [Windows]
  • GPU VRAM: 24gb
  • Number of GPUs: 1
  • GPU Make: [Nvidia]

Additional context How does params.json fail? It exists.

aaronjolson avatar Jan 30 '24 01:01 aaronjolson

hey @aaronjolson, we had an issue with the checklist.chk we uploaded for the Python and Instruct models. This should be fixed now, can you try and download the models again?

Romainsauvestre avatar Jan 31 '24 17:01 Romainsauvestre

Please re-open if you need help.

samuelselvan avatar Feb 07 '24 01:02 samuelselvan