Pinned Loading

1. nf4-triton-kernel nf4-triton-kernel Public

   An optimized, high-performance NF4 (NormalFloat 4-bit) dequantization Triton GPU kernel achieving up to 1.41x speedup over bitsandbytes.

   Python 5

2. fussion fussion Public

   A flexible PyTorch library for cross-modal fusion, connecting any modality encoder to frozen LLMs via modular bridges or internal cross-attention.

   Python

3. H-JEPA-LM H-JEPA-LM Public

   A PyTorch implementation of a Hierarchical Joint-Embedding Predictive Language Model (H-JEPA-LM) featuring action conditioning and latent-space world model planning.

   Python

4. Hyper-transformer Hyper-transformer Public

   A PyTorch research evolution of language models from standard Euclidean transformers to adaptive hybrid-manifold architectures with spiking neural networks.

   Python

5. xmerge-engine xmerge-engine Public

   Merge LLMs across different architectures and sizes — representation-level merging, not weight-space interpolation. Merge GPT-2, OPT, DistilGPT-2, SmolLM2, and more

   Python

6. TTA-Torch TTA-Torch Public

   A PyTorch library for real-time, confidence-gated Test-Time Adaptation (TTA) in LLMs using dynamic LoRA updates.

   Python