ndgold/Qwen3-1.7B-EasyLanguage-4bit

Qwen3-1.7B-EasyLanguage (4-bit, MLX)

LoRA fine-tune of Qwen/Qwen3-1.7B that rewrites live speech transcripts into easy-to-read registers across 29 language locales — German (Leichte Sprache), French (FALC), English (Easy/Plain English), Spanish (Lectura Fácil), Easy-to-Read Arabic (Inclusion Europe / Information for All), Letlæst (Inclusion Europe ETR), and more — for the Live Linguist on-device captioner. Each locale follows its own national / European Easy-to-Read or plain-language standard. Quantized to 4-bit for Apple-silicon inference via MLX.

It splits run-ons into short sentences, drops disfluencies, keeps names/numbers, and stays in the input language. Trained on a lean prompt (no few-shots) so the register is internalized — shorter prompts, lower live-caption latency.

Evaluation (held-out test set; SARI = simplification quality)

lang	SARI ft	SARI stock	chrF ft	LID ft	compliance ft
de	51.15	41.92	46.62	1.0	1.0
fr	62.03	54.15	55.63	1.0	0.995
es	61.66	56.9	57.79	0.995	0.99
en	62.37	52.0	57.12	1.0	1.0
ar	52.14	51.05	50.53	1.0	0.985
da	55.56	49.17	52.76	0.995	0.985
et	48.41	45.59	50.68	0.995	0.99
fi	50.57	47.25	55.96	1.0	0.995
hi	54.54	49.22	47.64	0.975	0.845
it	56.22	49.53	51.89	1.0	0.8866
ja	8.35	8.36	45.49	1.0	0.985
ko	50.1	46.48	42.85	1.0	0.975
nl	55.07	51.81	51.44	0.985	0.94
pt-BR	57.78	56.87	53.97	1.0	0.89
pt-PT	61.2	54.3	60.99	1.0	0.965
ru	54.46	51.04	51.46	1.0	0.97
sk	51.54	46.89	47.71	0.995	0.96
sv	54.95	49.53	57.89	0.975	0.96
tr	50.57	46.35	50.6	1.0	0.99
vi	62.5	57.28	60.86	1.0	0.865
zh-CN	8.64	8.8	53.85	0.995	0.99
uk	53.12	51.75	54.79	0.995	0.995
ro	53.69	48.06	52.88	1.0	1.0
bg	52.74	51.7	55.05	0.99	0.975
pl	51.55	44.47	49.84	1.0	0.995
hr	53.55	45.38	54.55	0.995	0.995
nb	54.3	41.35	56.16	0.995	0.99
hu	50.74	49.82	52.34	1.0	0.995
cs	50.44	48.65	49.23	1.0	0.995

Usage (MLX)

from mlx_lm import load, generate
model, tok = load("ndgold/Qwen3-1.7B-EasyLanguage-4bit")
msgs = [{"role":"system","content":"<framework system prompt>"},
        {"role":"user","content":"Original: <utterance>\nRewritten:"}]
p = tok.apply_chat_template(msgs, add_generation_prompt=True, tokenize=False, enable_thinking=False)
print(generate(model, tok, prompt=p, max_tokens=96))

Sources & licenses

• Base model: Qwen3 (Apache-2.0).
• German seed data: tum-nlp/German4All-Corpus (German Wikipedia, CC BY-SA).
• Synthetic pairs (all other languages + de augmentation): spoken→easy-language pairs generated by Claude (Anthropic) — Leichte Sprache (de), FALC (fr), Easy/Plain English (en), Lectura Fácil (es), and 25 further locales following each language's Easy-to-Read / plain-language standard (Inclusion Europe "Information for All", Selkokieli, Lättläst, やさしい日本語, ISO 24495-1, …). Every pair is filtered by a deterministic validator suite (per-language sentence-length caps, language-ID, fidelity anchoring, number preservation, anti-parroting).
• Intended for the Live Linguist on-device live-caption simplifier. Not a general chatbot.