torchscript

From-Scratch High Valyrian MiniGPT — Build and train a GPT-like transformer model fully from scratch, including custom tokenizer, dataloader, transformer, training, and inference! Learn the internals of LLMs layer-by-layer.

End-to-end moderation system for text & images. FastAPI backend with provider toggles (OpenAI & Google Perspective/Vision), PyTorch TorchScript prefilter microservice, policy.yaml rules, token-bucket rate limiting, in-memory caching, React/Vite frontend, training scripts, and Kubernetes manifests.

Continuously checks torch.onnx op support status

A robust, field-ready Plant Disease Classifier built with PyTorch, designed for mobile deployment and high-reliability diagnostics.

Unofficial fairseq-free PyTorch implementation of SCOREQ (2024), matching the original system.

PotatoNet - LPCVC 2021 Video Track Submission

CPU inference example for Kazakh, Russian, and mixed-speech TorchScript STT models hosted on Hugging Face.

Highway-env reinforcement learning workspace with DQN training and a versioned TorchScript policy service.

This repository contains scripts to export T5 model to torchscript or onnx.

Architectural plan elements detection based on Ultralytics YOLO and Faster R-CNN (TorchScript) with a convenient Gradio UI and FastAPI API

PyQt5 ECG AI assist monitor with TorchScript inference, offline replay, model contract, and one-click demo.

YOLOXのTorchscriptモデルで推論

Windows EXE GUI for dataset creation, YOLO training, self-training, and model export (ONNX/TorchScript) with Valorant-style UI.

High-performance PyTorch inference for edge devices — TorchScript operator fusion, custom CUDA kernels, FP16 mixed precision; 2.5x latency reduction

Deployed web application for alzheimers classification using ResNeXt-50 built from scratch in PyTorch

Raspberry Pi Torchscript Power/Runtime Evaluation Toolkit

Readable architecture diagrams for PyTorch models. The tracer supplies the facts, an agent supplies the abstraction, and a coverage check proves nothing was silently dropped.

Benchmark harness that turns model-serving configs into dollars per 1M requests. Serves a 4.9M-param intent classifier as fp32, dynamic int8, and TorchScript variants with and without micro-batching, measures p50/p95/p99, req/s, and peak RSS, and prices throughput against a dated AWS snapshot.

Self-trained PyTorch models for single-image super-resolution and low-light enhancement - a full pipeline from data to training, evaluation, TorchScript export, FastAPI backend and an interactive web demo. Research portfolio for AI/CS graduate school applications.