pytorch

A re-implementation of Stable-Diffusion using better code pratices with faster and lower-memory usage.

Implementation of GATO style Generalist Multimodal model capable of image, text, RL and Robotics tasks

Implementation of Z-BERT-A: a zero-shot pipeline for unknown intent detection.

reference pytorch code for intent classification

First open-source TurboQuant KV cache compression for LLM inference. Drop-in for HuggingFace. pip install turboquant.

Develop DL models using Pytorch and Hugging Face

In this project, several approaches for training/finetuning an audio gender recognition is provided. The code can simply be used for any other audio classification task by simply changing the number of classes and the input dataset.

Baseline model for multimodal classification based on images and text. Text representation obtained from pretrained BERT base model and image representation obtained from VGG16 pretrained model.

An end-to-end image captioning project using a CNN encoder (ResNet-50) and LSTM decoder in PyTorch. Includes vocabulary building, preprocessing, training with BLEU evaluation, and inference. Generates natural language captions for images with saved metrics, model checkpoints, and visualization outputs.

ONNX-compatible Fast SeamlessM4T—Massively Multilingual & Multimodal Machine Translation

RATransformers 🐭- Make your transformer (like BERT, RoBERTa, GPT-2 and T5) Relation Aware!

ONNX implementation of the BGE-M3 multilingual embedding model and tokenizer with native C#, Java, and Python implementations. Generates all three embedding types: dense, sparse, and ColBERT vectors.

End-to-end pipeline for seeing how LLMs actually process your prompts. Capture attention across every layer, render heatmaps and cooking curves, compare variants with evidence — not vibes.

PyTorch implementation of the ICASSP-24 paper: "Improving Audio Captioning Models with Fine-grained Audio Features, Text Embedding Supervision, and LLM Mix-up Augmentation"

Implementation of a modular, high-performance, and simplistic mamba for high-speed applications

A tiny language model that talks like a house cat named Miso.

Native Windows vLLM 0.27.1 wheels: Python 3.13, PyTorch 2.13 + CUDA 13.0, SM 7.5-12.0 for RTX 20/30/40/50, OpenAI-compatible serving, FlashAttention/Rust, 10 KV formats, Multi-TurboQuant, and experimental CPU/RAM/NVMe prompt-KV offload - no WSL or Docker.

🚀 Engram-PEFT: An unofficial implementation of DeepSeek Engram. Inject high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs / DeepSeek Engram 架构的非官方实现。通过参数高效微调 (PEFT) 为大语言模型注入超大规模条件记忆，支持稀疏更新且不增加推理开销。

Official resource of the paper "Knowledge Enhanced Masked Language Model for Stance Detection", NAACL 2021

A clean, single-file PyTorch implementation of Attention Residuals (Kimi Team, MoonshotAI, 2026), integrated with Grouped Query Attention (GQA), SwiGLU feed-forward networks, and Rotary Position Embeddings (RoPE).