AMAImedia/NOESIS-Qwopus3.5-9B-Translate-v3.5-BF16

Released as part of the NOESIS Professional Multilingual Dubbing Automation Platform (framework: DHCF-FNO — Deterministic Hybrid Control Framework for Frozen Neural Operators).

Founder: Ilia Bolotnikov
Organization: AMAImedia.com
X (Twitter): @AMAImediacom
LinkedIn: Ilia Bolotnikov
Telegram: @djbionicl
NOESIS version: v16.1
Release date: 2026-08

Qwopus3.5-9B-Translate-v3.5-BF16

Role: Dedicated translator (9B). Think-model — for deterministic translation use a closed-think prefill (<think>\n\n</think>\n\n after assistant\n) to suppress reasoning, otherwise it emits a reasoning trace instead of the translation. BF16 = PRIMARY. Sibling: -GGUF-Q4_K_M.gguf (5.24 GB, fits 6GB GPU).

Test results (2026-06-17) — Q4_K_M, FLORES devtest n=20, no-think

Direction	chrF++	BLEU
eng→rus	54.9	25.9
eng→cmn	32.8	7.2
AVG	43.8	16.5

Sample (eng→rus): «Теперь у нас есть мыши в возрасте четырёх месяцев, которые ранее страдали диабетом, но сейчас не болеют им», — добавил он.

Comparison (same FLORES n=20)

Model	Translate AVG chrF++/BLEU	Supervisor-12
This (9B-Translate Q4)	43.8 / 16.5	5/12 (not a supervisor)
NOESIS-4B-LongCtx Q8	41.7 / 16.5	11/12
base 4B Q8	42.4 / 15.4	5/12

Best pure translator (marginal: +2.1 chrF++ over the 4B, BLEU tied) but NOT a supervisor. Use 9B for the final translation pass when VRAM allows; the 4B-LongCtx is the all-rounder.

Speed (RTX 3060 Laptop 6GB, GPU, 33/33 layers offloaded)

• Q4_K_M: gen 49.1 tok/s, prompt eval 307 tok/s. (vs 4B-LongCtx Q8: 53.5 / 366 — the 4B is ~9% faster despite the 9B being lighter-per-param in Q4.)

⚠️ n=20 quick estimate (partly within noise). Eval on GPU via llama-completion.exe -ngl 99. Written: 2026-06-17

MT benchmark — FLORES-200 devtest (2026-06-17)

Real eval (not smoke): n=100 × 4 directions (eng↔rus, eng↔cmn), GPU via resident llama-server -ngl 99. Primary metric COMET (wmt22-comet-da, neural — how "best translator" is judged), plus chrF++ / BLEU / length-ratio. Each model prompted in its own native format (MT2 = dubbing ChatML "SOURCE (lang): … Только перевод"; 9B = ChatML + no-think). Data + COMET checkpoint: D:/models/by_expert/07_MT_TRANSLATION.

Model	Size	COMET avg	chrF++	BLEU	gen tok/s
Qwopus3.5-9B-Translate Q4	5.24 GB	0.8870	50.7	22.5	49
NOESIS-Hy-MT2-7.5B Q5	5.0 GB	0.8709	46.2	21.4	52
NOESIS-Hy-MT2-1.8B Q8	1.78 GB	0.8481	43.9	19.1	121

Per-direction COMET — 9B-Translate wins all 4 (eng-rus .902 / eng-cmn .897 / rus-eng .872 / cmn-eng .877); MT2-7.5B 2nd, MT2-1.8B 3rd.

Notes:

• MT2 is a dubbing translator (isochrony): its outputs are shorter (len_ratio ~0.87-0.89 vs 9B ~1.0) because it compresses to fit speech slots → lower chrF on literal FLORES news. FLORES does NOT measure MT2's slot-fit strength, so it under-rates MT2 for its actual job.
• 1.8B→7.5B degradation: COMET +0.023, chrF +2.3, BLEU +2.3 — modest; 1.8B is 2.4× faster and 2.8× smaller (good lightweight tradeoff).
• BLEU for eng-cmn is low for all (Chinese needs char-tokenization); use chrF++/COMET there.