Skip to content

Conversation

VinitP1102
Copy link

Summary

This PR introduces the integration of the Mixtral-8x22B model into the codebase. Specifically, the following changes have been made:

  • Added the Mixtral model in mixtral.py to support the 8x22B architecture.
  • Introduced a new engine for the Mixtral model in mixtral_engine.py
  • Updated __init__.py files in the models and engines directories to register the new Mixtral model.
  • Modified generation_config.yaml to include parameters for the Mixtral model's generation tasks.
  • Updated finetuning_config.yaml to configure Mixtral model-specific parameters for finetuning.
  • Updated documentation, including README.md and supported_models.md, to reflect the addition of the Mixtral-8x22B model, with its identifier key as "mixtral".

Checklist

  • Tested the integration of the Mixtral model with both generation and finetuning tasks.
  • Updated documentation files to reflect the changes.

Additional Information

  • A new example script, mixtral.py, has been added in the examples directory. This script demonstrates how to use the Mixtral model with xTuring and provides instructions for testing the model.
  • These changes enable the Mixtral-8x22B model's functionality for both generation and finetuning tasks within the xTuring project. The model's parameters have been incorporated into the respective configuration files.

StochasticRomanAgeev and others added 30 commits April 6, 2023 23:36
Added information checks and typo fixes for README.md
Added information about VRAM to main README.md
Added licence information to main README.md
Fixed licence line in README.md
Added important installation step
Added fixes for information, removed docker dependancy
Added batch size control to docker
Changed comment to be more informative
Changed README.md both for main and INT4 example
…nges

docs: refined readme changes for INT4
Run Docker container INT4 from the shell
…_int4

feat: run Docker container INT4 from the shell
changed README.md
…ription

docs: add new feature highlighted section
removed new feature in README.md
removed new feature in README.md
…ription

docs: change casing on README.md
Explained why we're getting the numbers we're getting and what the optimal configuratio could be
Adding int4 finetuning pipeline
fix: added explanatory paragraph and more benchmarks
MarcosRiveraMartinez and others added 28 commits September 6, 2023 20:23
Fixed docs links and naming
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Integrate ITREX to support int8 model on the CPU-only devices
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Modified README.md with better formatting and replaced link in the getting started example
- load via MambaForCausalLM
- upgrade Transformers
- add mamba to yamls
- Added Mixtral engine to support the model.
- Added example of Mixtral model.
- Edited `README.md` to display the latest added model.
- Edited config files for Mixtral model.
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.