gemma-3-4B-T1-it-mlx-fp16/README.md

license, language, base_model, library_name, tags, datasets, metrics, model-index

license	language	base_model	library_name	tags	datasets	metrics	model-index
gemma	en zh	twinkle-ai/gemma-3-4B-T1-it	transformers	Taiwan R.O.C zhtw SLM Gemma-3 gemma3 mlx mlx-my-repo	lianghsun/tw-reasoning-instruct lianghsun/tw-contract-review-chat minyichen/tw-instruct-R1-200k minyichen/tw_mm_R1 minyichen/LongPaper_multitask_zh_tw_R1 nvidia/Nemotron-Instruction-Following-Chat-v1	accuracy	name results gemma-3-4B-T1-it task dataset metrics type name question-answering Single Choice Question name type config split revision tmmlu+ ikala/tmmluplus all test c0e8ae955997300d5dbf0e382bf0ba5115f85e8c type value name accuracy 47.44 single choice task dataset metrics type name question-answering Single Choice Question name type config split revision mmlu cais/mmlu all test c30699e type value name accuracy 59.13 single choice task dataset metrics type name question-answering Single Choice Question name type config split revision tw-legal-benchmark-v1 lianghsun/tw-legal-benchmark-v1 all test 66c3a5f type value name accuracy 44.18 single choice

thliang01/gemma-3-4B-T1-it-mlx-fp16

The Model thliang01/gemma-3-4B-T1-it-mlx-fp16 was converted to MLX format from twinkle-ai/gemma-3-4B-T1-it using mlx-lm version 0.29.1.

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("thliang01/gemma-3-4B-T1-it-mlx-fp16")

prompt="hello"

if hasattr(tokenizer, "apply_chat_template") and tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )

response = generate(model, tokenizer, prompt=prompt, verbose=True)