Modelpublic

sylvester-francis/rust-mentor-4b-GGUF

sourceHugging Facegemmaupdated 6mo agoView on Hugging Face

0likes81downloads

Model Card

RustMentor 4B — GGUF (Q4KM)

A Rust programming tutor fine-tuned from Gemma 3 4B-IT. Teaches Rust by drawing comparisons to Python, Go, and TypeScript.


Base model	google/gemma-3-4b-it
Fine-tuning	QLoRA (r=16, alpha=16) via Unsloth + TRL
Quantization	Q4KM (k-quant mixed precision via llama.cpp)
Size	~2.5 GB
Format	GGUF (compatible with llama.cpp, PocketPal AI, Maid)
License	Gemma

Offline Android coding tutor for developers learning Rust. The model:

—Explains Rust concepts using parallels to Python, Go, and TypeScript
—Covers ownership, borrowing, lifetimes, traits, error handling, async, and more
—Refuses out-of-domain questions (non-Rust topics)
—Resists prompt injection attempts

PocketPal AI (Android): Download from the Play Store, then load this model from the HF Hub.

llama.cpp (CLI):

bash

./llama-cli -m rust-mentor-4b-Q4_K_M.gguf -p "<start_of_turn>user
Explain ownership in Rust<end_of_turn>
<start_of_turn>model
"

Fine-tuned on 500 samples covering:

—16 Rust teaching conversations (ownership, borrowing, lifetimes, traits, enums, pattern matching, async, iterators, structs, strings, serde, cargo, error handling, testing, memory model)
—10 out-of-domain refusal examples (Java, C++, Ruby, cooking, math, etc.)
—9 prompt injection hardening examples (persona override, instruction extraction, jailbreak roleplay, encoded bypass, etc.)