gguf-ggml-model-mirror

GGUF / GGML model mirror

Byte-for-byte copies of four quantised models: two GGUF language models for llama.cpp and two GGML speech recognition models for whisper.cpp.

Nothing here is original work and nothing has been modified. Each file is a copy of the file named in the table below, mirrored so that a fixed set of weights stays available at a stable address. Upstream repositories get reorganised, re-quantised and occasionally deleted; a mirror pinned to a known checksum does not.

If you are choosing models to use, prefer the upstream repositories linked below — they are maintained, this is not.

Files

Provenance and checksums

Copied 2026-08-30. Each file was verified against its SHA-256 after download and again after upload.

Credit for the quantisations and conversions belongs to their publishers: Qwen3 GGUF by unsloth, Llama 3.2 GGUF by bartowski, and the Whisper GGML conversions by Georgi Gerganov. The base weights are by Alibaba Cloud (Qwen3), Meta (Llama 3.2) and OpenAI (Whisper).

Licences

Three different licences apply, each only to its own files. See NOTICE for the attribution notices that must travel with these copies.

• Qwen3-1.7B-Q4_K_M.gguf — Apache License 2.0, LICENSE-apache-2.0.txt.

• Llama-3.2-3B-Instruct-Q4_K_M.gguf — Llama 3.2 Community License, LICENSE-llama-3.2.txt. Use is also governed by Meta's Acceptable Use Policy at https://www.llama.com/llama3_2/use-policy, which the licence incorporates by reference.

  Built with Llama. Llama 3.2 is licensed under the Llama 3.2 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved.

• ggml-base.bin, ggml-tiny.bin — MIT, LICENSE-mit.txt.