SirSahOl/Phi-4-mini-instruct-chat-mlx-4bit

Phi-4-mini-instruct-mlx-4bit

4-bit MLX conversion of microsoft/Phi-4-mini-instruct for Apple Silicon.

Converted by: SirSahOl Source model: microsoft/Phi-4-mini-instruct Framework: MLX by Apple Quantization: 4-bit Format: safetensors License: mit

---

Quick Start

Installation

pip install mlx-lm

CLI Usage

# Chat interactively
mlx_lm.chat --model SirSahOl/Phi-4-mini-instruct-chat-mlx-4bit

# Generate text
mlx_lm.generate --model SirSahOl/Phi-4-mini-instruct-chat-mlx-4bit --prompt "Your prompt here"

Python Usage

from mlx_lm import load, generate

model, tokenizer = load("SirSahOl/Phi-4-mini-instruct-chat-mlx-4bit")
response = generate(model, tokenizer, prompt="Your prompt here", max_tokens=256)
print(response)

---

Performance Benchmarks

| Metric | 4-bit | |--------|--------|| Tokens/sec | 23.99 | | TTFT | 41.69 ms | | Peak Memory | 2306.2 MB |

Benchmarked on Apple M1 with 8GB unified memory. Average over 5 runs with 256 max tokens.

---

Who Should Use This?

Your Hardware	Recommended Quantization
M1/M2 (8GB)	4-bit — Best balance of quality and memory usage
M1/M2 Pro/Max (16-32GB)	8-bit — Higher quality with reasonable memory
M2/M3/M4 Ultra (64GB+)	16-bit — Full precision, no quality loss

General guidance:

• Use 4-bit if you want to run this model alongside other applications
• Use 8-bit if you have the memory and want better quality
• Use 16-bit for research, evaluation, or if memory isn't a concern

---

Other Quantization Variants

Variant	Link
4-bit	SirSahOl/Phi-4-mini-instruct-chat-mlx-4bit
8-bit	SirSahOl/Phi-4-mini-instruct-chat-mlx-8bit
16-bit	SirSahOl/Phi-4-mini-instruct-chat-mlx-16bit

---

Conversion Details

Property	Value
Source Model	microsoft/Phi-4-mini-instruct
Quantization	4-bit
mlx-lm Version	0.31.3
Conversion Time	3386.82s
Output Size	2.0 GB
Date	2026-09-12T02:36:32.858764+00:00

Reproduction

To reproduce this conversion:

pip install mlx-lm==0.31.3
python3 -m mlx_lm.convert --hf-path microsoft/Phi-4-mini-instruct --mlx-path output/Phi-4-mini-instruct-mlx-4bit -q --q-bits 4

---

Limitations & Known Issues

• Performance may degrade with very long contexts (>8K tokens) at lower quantization levels.
• This is a weight-only conversion; the model architecture and behavior are inherited from the source model.
• Quantization introduces a small quality loss compared to the original model. Lower bit counts = more loss.
• This model requires Apple Silicon (M1 or later) to run with MLX.

---

License

This model conversion inherits the license of the source model: mit.

See the original model card for full license details.

---

Changelog

Version	Date	Changes
v1.0	2026-09-12	Initial conversion

---

Converted with MLX Foundry — a professional pipeline for converting models to Apple MLX format.