Projects

48 8.6B Parameters

Instruction-tuned PLaMo-3-8B model optimized for Japanese language tasks. Featured in awesome-japanese-llm (⭐1.3k).

Text Generation Japanese LLM Instruction Tuning

44

Vision-language model fine-tuned for disaster scene recognition from aerial imagery using QLoRA 4-bit quantization.

Image-Text-to-Text Disaster Recognition QLoRA

51

Japanese vision-language model fine-tuned for natural photo conversations using QLoRA efficient training.

Image-Text-to-Text Japanese VLM QLoRA

Flow Matching-based generative model using U-Net architecture for 64x64 celebrity face generation.

Generative Model Flow Matching Image Generation

113 100+ downloads

High-quality cleaned dataset for training Japanese vision-language models on photo conversation tasks.

Japanese Vision-Language Conversation

15

Multilingual (English, Chinese, Japanese) dataset for disaster recognition from remote sensing imagery.

Remote Sensing Disaster Recognition Multilingual

1

LangGraph ReAct Agent supporting local/remote vLLM (Qwen 2.5) and cloud APIs, integrated with DeepSeek OCR as a remote MCP server.

LangGraph vLLM MCP DeepSeek OCR

DeepSort-IMM

9

Improved Kalman Filter implementation for DeepSort using Interactive Multiple Model (IMM) algorithm for enhanced multi-object tracking performance.

Multi-Object Tracking Kalman Filter IMM DeepSort

Models