nikimori/gemma-4-e2b-it-4bit-slim

Gemma 4 E2B-it, 4-bit MLX (slim)

A slimmer re-pack of mlx-community/gemma-4-e2b-it-4bit, a conversion of google/gemma-4-E2B-it:

• The audio tower is removed. The vision tower stays, because Pretype's speculative-decoding load (with Google's MTP drafter) needs it.
• All other weights are unchanged.

The weights are 2.94 GB instead of 3.55 GB (−17%). In Pretype's evals the outputs match the original (details). The license is the upstream one (Apache-2.0).

Made for Pretype, on-device autocomplete for macOS. It is the model that fixes a selection (⌥Tab) and drafts replies on Pretype's Gemma E2B 4-bit tier.

On Pretype's held-out sets (2026-10-11, 17–19 languages), it restores 50% [45–54] of single-typo sentences exactly, proposes a change on 29% of clean lines, and drafts a usable reply for 96% [93–97] of on-screen conversations.

from mlx_lm import load, generate
model, tokenizer = load("nikimori/gemma-4-e2b-it-4bit-slim")
messages = [{"role": "user", "content": "Fix the typo: I recieved your letter."}]
prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True)
print(generate(model, tokenizer, prompt=prompt, max_tokens=40))