text-generation-inference
Improve the defaults for the launcher
#1727
Merged

Commits
  • Easier defaults for models stemmed from configs.
    Narsil committed 2 years ago
  • Better defaults (and LOG_COLORIZE).
    Narsil committed 2 years ago
  • Update default doc.
    Narsil committed 2 years ago
  • No unwrap.
    Narsil committed 2 years ago
  • Making things work most of the time.
    Narsil committed 2 years ago
  • Change things around when we don't have a tokenizer.
    Narsil committed 2 years ago
  • Remove the override ?
    Narsil committed 2 years ago
  • Adding some wiggle room.
    Narsil committed 2 years ago
  • Fixing default for BNB + cuda graphs (they don't work together).
    Narsil committed 2 years ago
  • Max_seq_len (old mpt config.)
    Narsil committed 2 years ago
  • "Fixing t5" just use more RAM for this test.
    Narsil committed 2 years ago
  • Smaller default for max_input_length.
    Narsil committed 2 years ago
  • Forgot the doc again.
    Narsil committed 2 years ago
  • Update launcher/src/main.rs
    Narsil committed 2 years ago
  • Update the doc.
    Narsil committed 2 years ago
Loading