A fine-tuned, Q4_K_M-quantised version of unsloth/Qwen2.5-Coder-0.5B-Instruct trained to act as a React + TypeScript Architect capable of producing full, multi-file component trees with correct typings and CSS.

Intended Use

Platform	Runtime
Android (Termux)	llama.cpp
Desktop	llama.cpp / Ollama / LM Studio

Prompt Format

### System: You are a React TS Architect. ...
### Instruction: <task>
### Input: <optional context>
### Response:

Training Details

Parameter	Value
Base model	`unsloth/Qwen2.5-Coder-0.5B-Instruct`
LoRA rank	32
Max sequence length	2048
Optimiser	AdamW 8-bit
Steps	150
Quantisation	Q4_K_M (GGUF)
Framework	Unsloth + TRL

Datasets

mhhmm/typescript-instruct-20k — TypeScript instruction pairs
iamdyeus/ui-instruct-4k — UI / component generation instructions
Agent-Ark/Toucan-1.5M (SFT split) — Agentic tool-calling trajectories

Mist26/react-ts-architect-0.5b-gguf

react-ts-architect-0.5b-gguf

Intended Use

Prompt Format

Training Details

Datasets

Mist26/
react-ts-architect-0.5b-gguf