v1.13

Backend updates

• llama-cpp-python: bump to 0.2.85 (adds Llama 3.1 support).

UI updates

• Make compress_pos_emb float (#6276). Thanks @hocjordan.
• Make n_ctx, max_seq_len, and truncation_length numbers rather than sliders, to make it possible to type the context length manually.
• Improve the style of headings in chat messages.
• LaTeX rendering:
  • Add back single $ for inline equations.
  • Fix rendering for equations enclosed between \[ and \].
  • Fix rendering for multiline equations.

Bug fixes

• Fix saving characters through the UI.
• Fix instruct mode displaying "quotes" as ""double quotes"".
• Fix chat sometimes not scrolling down after sending a message.
• Fix the chat "stop" event.
• Make --idle-timeout work for API requests.

Other changes

• Model downloader: improve the progress bar by adding the filename, size, and download speed for each downloaded file.
• Better handle the Llama 3.1 Jinja2 template by not including its optional "tools" headers.