File size: 4,097 Bytes
39f8925
 
 
 
 
 
 
 
 
9a84a57
4bb4131
9a84a57
c53ec71
 
9a84a57
 
 
 
 
 
 
c53ec71
0593bc2
 
 
 
0415ae7
 
 
0593bc2
 
 
8ee2a29
 
 
ea69edb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
---
tags:
  - qwen3.5
  - qwen3.5-201-languages
  - supports-201-languages
  - multilingual
  - language-support
---

## NOESIS / AMAImedia
**Last updated:** 2026-08-29

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](https://x.com/AMAImediacom)
- **LinkedIn:** [Ilia Bolotnikov](https://www.linkedin.com/in/ilia-bolotnikov)
- **Telegram:** [@djbionicl](https://t.me/djbionicl)
- **NOESIS version:** v16.1
- **Release date:** 2026-08-26

<!-- AMAIMEDIA_QWEN_LANGUAGE_SUPPORT_START -->

## Language support

This Qwen3.5-derived model follows the official Qwen3 language list below (**119 languages and dialects**) and the official Qwen3.5 coverage statement of **201 languages and dialects**. Qwen3.5 publishes the expanded coverage count but does not provide an exhaustive 201-name enumeration in its model card. The list below is the complete language list published by Qwen for Qwen3 and is included as the transparent, documented baseline for this derivative.

English, French, Portuguese, German, Romanian, Swedish, Danish, Bulgarian, Russian, Czech, Greek, Ukrainian, Spanish, Dutch, Slovak, Croatian, Polish, Lithuanian, Norwegian Bokmål, Norwegian Nynorsk, Persian, Slovenian, Gujarati, Latvian, Italian, Occitan, Nepali, Marathi, Belarusian, Serbian, Luxembourgish, Venetian, Assamese, Welsh, Silesian, Asturian, Chhattisgarhi, Awadhi, Maithili, Bhojpuri, Sindhi, Irish, Faroese, Hindi, Punjabi, Bengali, Oriya, Tajik, Eastern Yiddish, Lombard, Ligurian, Sicilian, Friulian, Sardinian, Galician, Catalan, Icelandic, Tosk Albanian, Limburgish, Dari, Afrikaans, Macedonian, Sinhala, Urdu, Magahi, Bosnian, Armenian; Chinese (Simplified Chinese, Traditional Chinese, Cantonese), Burmese; Arabic (Standard, Najdi, Levantine, Egyptian, Moroccan, Mesopotamian, Ta’izzi-Adeni, Tunisian), Hebrew, Maltese; Indonesian, Malay, Tagalog, Cebuano, Javanese, Sundanese, Minangkabau, Balinese, Banjar, Pangasinan, Iloko, Waray (Philippines); Tamil, Telugu, Kannada, Malayalam; Turkish, North Azerbaijani, Northern Uzbek, Kazakh, Bashkir, Tatar; Thai, Lao; Finnish, Estonian, Hungarian; Vietnamese, Khmer; Japanese, Korean, Georgian, Basque, Haitian, Papiamento, Kabuverdianu, Tok Pisin, Swahili.

<!-- AMAIMEDIA_QWEN_LANGUAGE_SUPPORT_END -->

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


# NOESIS-Qwopus3.5-4B-v3-Supervisor-LongCtx-NF4

**Role:** Long-context Supervisor — 6GB-runtime NF4 build (multi-segment QC, orchestration,
cross-stage review, batch best-of-N) for the dubbing pipeline.
**Quantized from** `NOESIS-Qwopus3.5-4B-v3-Supervisor-LongCtx-BF16` (bnb load_in_4bit, nf4,
double_quant, compute=bf16). PRIMARY artifact = the BF16 sibling; GGUF Q8_0 = laptop.
Equivalent to base-NF4 + LoRA adapter `nt346_longctx_qwopus4b`, pre-merged.

## Test results (2026-06-17)
Score measured on the matching Q8_0 GGUF (NF4 is the same merged weights at 6GB runtime):

### Supervisor eval — eval_longctx_supervisor_v1.py (12 tests, grammar-constrained)
- **11/12** (Q8_0). Only failure = `L1_mixed` (fail-safe over-escalation). Prior best
  long-ctx (LFM2.5-7.5B) = 7/12.

### Translation — FLORES devtest (n=20, no-think)
- eng→rus chrF++ 51.5 / BLEU 25.6 ; eng→cmn 32.0 / 7.4 ; **AVG 41.7 / 16.5**.

### Comparison
| Model | Supervisor-12 | Translate AVG chrF++/BLEU |
|-------|---------------|---------------------------|
| **This (4B-LongCtx)** | **11/12** | 41.7 / 16.5 |
| base 4B (no LoRA) | 5/12 | 42.4 / 15.4 |
| Qwopus3.5-9B-Translate Q4 | 5/12 | 43.8 / 16.5 |

### Speed (RTX 3060 Laptop 6GB, GPU, Q8_0 sibling, 33/33 layers)
- gen **53.5 tok/s**, prompt eval 366 tok/s (vs 9B-Translate Q4: 49.1 / 307).

All-rounder: full supervision + translation parity with the base, smaller footprint.
Written: 2026-06-17