Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

TGI 2.3.1 #103

Open
wants to merge 4 commits into
base: main
Choose a base branch
from
Open

TGI 2.3.1 #103

wants to merge 4 commits into from

Conversation

philschmid
Copy link
Contributor

@philschmid philschmid commented Oct 7, 2024

The TGI team worked really hard on adding M-Llama i.e. Vision Llama support so kudos to them! :robot_clap:
Some of the features of the new TGI release besides M-Llama are:

  • Renamed HUGGINGFACE_HUB_CACHE to use HF_HOME (to harmonize environment variables across HF ecosystem)
  • Add prefix caching by default. To help with long running queries TGI will use prefix caching a reuse pre-existing queries in the KV-cache in order to speed up Time-To-First-Token (TTFT).
  • Changed default kernels from paged_attention to flashinfer (and flashdecoding as a fallback for some specific models that aren't supported by flashinfer).
  • Fixed support for Gemma2 and variants such as DataGemma.
  • Included lots of performance improvements with Marlin and quantization.
  • Uses Python 3.11 instead of Python 3.10; and uses CUDA 12.4 instead of CUDA 12.1
  • Adds support for M-Llama i.e. latest Vision Llama 3.2
  • FP8 and MoE improvements

@philschmid philschmid requested a review from a team as a code owner October 7, 2024 10:55
@yusharon yusharon mentioned this pull request Oct 7, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant