---
model: Qwen/Qwen2.5-1.5B-Instruct-GGUF
library_name: llama.cpp / wllama
license: apache-2.0
---

https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct-GGUF — instruction-tuned
1.5B-parameter Qwen2.5 language model distributed as a GGUF quantization.
Distributed file: `qwen2.5-1.5b-instruct-q4_k_m.gguf` (Q4_K_M), redistributed
unchanged. Pinned revision:
`91cad51170dc346986eccefdc2dd33a9da36ead9`.

The weights run through `@wllama/wllama` (llama.cpp compiled to WebAssembly),
which is MIT licensed — Copyright (c) 2024 Xuan Son NGUYEN,
https://github.com/ngxson/wllama. The WASM runtime is staged at build time and
served unchanged from `/runtime/wllama.wasm`.

This model powers on-device scene selection, subtitle cleanup, description and
timestamp generation. It is fetched only when an inference-capable service is
prepared; no source media or user text leaves the device.
