You can deploy this model and start earning money today!

SAINITHIN/
Exaplanation_Distillation

Views

No views yet

Explanation distillation — VL-LLaMA 2 7B (finetuned)

Hugging Face–style checkpoint repository storing finetuned VideoLLaMA 2 7B Adapter weights for explanation-distillation work.

Links

Resource	URL
Model / project page (HF Hub)	SAINITHIN/Exaplanation_Distillation
Source code (DriveXplain)	github.com/Sainithin-bit/DriveXplain
Paper (WACV 2026, CVF open access)	Distilling What and Why: Enhancing Driver Intention Prediction with MLLMs (WACV 2026)
Project site (IIIT CVIT)	DriveXplain project page
Conference poster	WACV virtual poster #692

These checkpoints accompany DriveXplain — Distilling What and Why — for distilling maneuver + explanation pairs into a compact MLLM.

File	Description
`VL_LLaMA_2_7B_Finetuned.pth`	PyTorch checkpoint (`.pth`), tracked with Git LFS

The on-disk file you see in a fresh clone may be an LFS pointer only. The actual object is:

LFS OID: sha256:3cec0e2979ed7656e08ecc5b185c2229a3c577b4b7a4721a94bd461ba0447c6e
Approx. size: ~253 MB (265559201 bytes)

Clone the real weights

Install Git LFS and pull before using the checkpoint:

bash

git lfs install
git clone <this-repo-url>
cd Exaplanation_Distillation
git lfs pull

Confirm the .pth is a binary checkpoint (not a short version https://git-lfs.github.com/spec/v1 text file).

Usage notes

Load with the same code path / config you used when producing this checkpoint (architecture must match the saved state_dict keys).
This repo does not ship training or inference scripts—only the weight artifact. Pair it with your VL-LLaMA codebase and dependencies.

Large files

Binary and archive extensions in this repo are configured for LFS in .gitattributes (including *.pth, *.pt, *.safetensors, etc.), consistent with typical Hugging Face model repositories.

SAINITHIN/Exaplanation_Distillation