-
Stencil in CUDA
Exploring different optimization methods in the implementation of 3D stencils in CUDA.
-
Convolutions in CUDA
Exploring different optimization methods in the implementation of Convolutions in CUDA.
-
Histogram in CUDA
Exploring different optimization methods in the implementation of Histograms in CUDA.
-
FakeLLaVA — Building a Lightweight LLaVa style VLM with Qwen2.5 and CLIP
Training a lightweight LLaVA-style VLM by connecting a CLIP vision encoder to Qwen2.5-0.5B through a two-layer projection MLP.
-
CLIP Models and Image Captioning using CLIP embeddings.
Exploring contrastive learning and Info-NCE loss and training a CLIP vision encoder, then building a GPT-2 image captioning model (ClipCap) trained on chest X-ray CLIP embeddings.