Wasper pins its bundled local LLM to a specific, byte-for-byte reproducible file. Mirroring the Q4_K_M GGUF under our own namespace lets us pin a commit hash that cannot change under us and guarantees the first-run download experience stays reliable regardless of upstream availability. The file here is byte-identical to what Unsloth published upstream; we only re-host it for pinning stability.

This mirror is not updated out of band. It exists solely to serve the Wasper macOS app's local AI onboarding flow.

Verification

The Wasper app verifies the downloaded file's SHA-256 against a pinned value before installing. If you are curious:

shasum -a 256 qwen3-4b-instruct-2507-q4km.gguf

Expected: 3605803b982cb64aead44f6c1b2ae36e3acdb41d8e46c8a94c6533bc4c67e597

Attribution

Model weights and training: Qwen team at Alibaba Cloud — Qwen/Qwen3-4B-Instruct-2507
GGUF conversion: Unsloth — unsloth/Qwen3-4B-Instruct-2507-GGUF
Mirrored by: osa911 for the Wasper macOS app

Apache 2.0 license text from the upstream Qwen repo applies unchanged. See NOTICE at huggingface.co/Qwen/Qwen3-4B-Instruct-2507.

osa911/qwen3-4b-instruct-2507-q4km-gguf

Qwen 3 4B Instruct 2507 — Q4_K_M (Wasper mirror)

Why this mirror exists

Verification

Attribution

osa911/
qwen3-4b-instruct-2507-q4km-gguf