Skip to content

Conversation

@amitbcp
Copy link
Contributor

@amitbcp amitbcp commented Aug 12, 2024

Idefics 3 follows same pattern as Idefics2.

Building HF from the Source Code and huggingface/transformers#32473 enables Idefics3

The model has been tested with the transformers library

@HugoLaurencon
Copy link
Contributor

Sounds good to me, it's possible that there's nothing to change indeed.

Because there are small discrepancies between generating with our internal codebase and Transformers integration, please ping me if the scores differ significantly from the officially reported ones

@amitbcp
Copy link
Contributor Author

amitbcp commented Aug 20, 2024

@HugoLaurencon : Yes the changes are only to load the new model version of Idefics3 via config and re-use same style of inference as in Idefics2, other aspects remains constant.

@kennymckormick kennymckormick merged commit 5d1e0f9 into open-compass:main Aug 24, 2024
shan23chen pushed a commit to shan23chen/VLMEvalKit that referenced this pull request Oct 3, 2024
* VILA added

* Update README.md

* resolve config merge conflict

* Fix error on Idefics for longer prompt

* Fix naming convention to make consistent with Idefics2 and better readability

* update config for idefics

* Make LLava consistent as well

* Add VILA 1.5 3B

* Add VILA 1.5 3B

* fix naming convention to be similar to the HF models

* Multi-Turn added for Phi3-Vision and tested with MMDU

* Add multi turn for Intern VL

* fix formatting

* Add Idefics3 Config

* Warning message to build from source

---------

Co-authored-by: aamita <aamita@sdg-slurm-bm-gpu-b4-8-ad3-009.compute.sdgdevvcn.oraclevcn.com>
Co-authored-by: Junming Yang <60545459+junming-yang@users.noreply.github.com>
Co-authored-by: Haodong Duan <dhd@pku.edu.cn>
Koii2k3 pushed a commit to wjnwjn59/VLMEvalKit that referenced this pull request Nov 13, 2025
* VILA added

* Update README.md

* resolve config merge conflict

* Fix error on Idefics for longer prompt

* Fix naming convention to make consistent with Idefics2 and better readability

* update config for idefics

* Make LLava consistent as well

* Add VILA 1.5 3B

* Add VILA 1.5 3B

* fix naming convention to be similar to the HF models

* Multi-Turn added for Phi3-Vision and tested with MMDU

* Add multi turn for Intern VL

* fix formatting

* Add Idefics3 Config

* Warning message to build from source

---------

Co-authored-by: aamita <aamita@sdg-slurm-bm-gpu-b4-8-ad3-009.compute.sdgdevvcn.oraclevcn.com>
Co-authored-by: Junming Yang <60545459+junming-yang@users.noreply.github.com>
Co-authored-by: Haodong Duan <dhd@pku.edu.cn>
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.

4 participants