CoolFace
Modelpublic

Hanish/quill-models

sourceHugging Faceapache-2.0updated 8d agoView on Hugging Face
0likes72downloads
Model Card

Quill on-device chat models (GGUF mirror)

Byte-identical mirrors of upstream GGUF files used by the Quill app (offline chat + AI keyboard, Android). Nothing here is modified or fine-tuned; the mirror exists so the app's downloads do not depend on upstream repo names. Quill's own keyboard model lives in Hanish/quill-fix-v1.

FileModelQuantSizeUpstream GGUFSoft RAM floor in Quill
Qwen3.5-0.8B-Q4_K_M.ggufQwen3.5-0.8B (instruct)Q4KM533 MBunsloth/Qwen3.5-0.8B-GGUF3 GB ("Chat Lite")
Qwen3.5-2B-Q4_K_M.ggufQwen3.5-2B (instruct)Q4KM1.28 GBunsloth/Qwen3.5-2B-GGUF5 GB ("Chat")
gemma-4-E2B-it-Q4_0.ggufGemma 4 E2B (instruct)Q4_02.84 GBggml-org/gemma-4-E2B-it-GGUF7 GB ("Chat Pro")

sha256 (also pinned in the app):

bd258782e35f7f458f8aced1adc053e6e92e89bc735ba3be89d38a06121dc517  Qwen3.5-0.8B-Q4_K_M.gguf
aaf42c8b7c3cab2bf3d69c355048d4a0ee9973d48f16c731c0520ee914699223  Qwen3.5-2B-Q4_K_M.gguf
8e30dff3ac4c8434c49a7036fa15564bdbb6044e42bf04550bf1a096ad7e6a52  gemma-4-E2B-it-Q4_0.gguf

Licenses and credit

  • —Qwen3.5 models: © Alibaba Cloud, Apache-2.0. GGUF conversions by Unsloth.
  • —Gemma 4 E2B: © Google DeepMind, Apache-2.0. GGUF conversion by the ggml-org team.
  • —Runs with llama.cpp (MIT). Quill uses build b6b003d or newer (qwen35 and gemma4 architectures).

Prompt formats used by Quill

  • —Qwen3.5: ChatML with an empty <think>\n\n</think> block after <|im_start|>assistant (thinking off).
  • —Gemma 4: <bos><start_of_turn>user\n{system}\n\n{user}<end_of_turn>\n<start_of_turn>model\n.