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
• LinkedIn: Ilia Bolotnikov
• Telegram: @djbionicl
• NOESIS version: v16.1
• Release date: 2026-08-26

NOESIS-Qwopus3.5-0.8B-v3-Inspector-MLX

Original NOESIS MLX model for lightweight inspection, validation, routing checks, and agent-quality review. This repository contains the local MLX safetensors conversion of the Qwopus3.5 v3 Inspector specialist. It is intended for Apple Silicon inference through mlx-lm; the files may be stored on Windows, but MLX execution requires a compatible macOS/Apple-Silicon environment.

This is the MLX counterpart of the published NOESIS-Qwopus3.5-0.8B-v3-Inspector-Q8_0 model card. The Q8_0 repository is the llama.cpp-compatible artifact; this repository preserves the original NOESIS model in MLX format for Apple Silicon.

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

Language support

The Qwen3.5 family advertises expanded coverage of 201 languages and dialects. The upstream Qwen3 language list below is the transparent documented baseline used for this derivative; the published Qwen3.5 count is broader than the enumerated list.

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.

The model card intentionally avoids claiming identical quality for every language. For production dubbing, language quality must be checked by the NOESIS multilingual ASR, speaker, emotion, and timing gates.

Repository contents

NOESIS-Qwopus3.5-0.8B-v3-Inspector-MLX/
├── README.md
├── model.safetensors              # MLX model weights, ~1.50 GB
├── model.safetensors.index.json   # weight metadata
├── config.json                    # Qwen3.5 architecture configuration
├── generation_config.json
├── tokenizer.json
├── tokenizer_config.json
└── chat_template.jinja            # chat formatting used by the tokenizer

How to run with MLX-LM

Important platform note

mlx-lm is designed for Apple Silicon and uses the MLX backend. The B:\Downloads\... path is the Windows storage location of this repository; it is not a native Windows execution path for MLX. Copy or sync the repository to an Apple Silicon Mac, then use a macOS path such as /Users/ilia/Models/NOESIS-Qwopus3.5-0.8B-v3-Inspector-MLX.

Install

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -U pip mlx-lm

Single prompt generation

python -m mlx_lm.generate \
  --model /Users/ilia/Models/NOESIS-Qwopus3.5-0.8B-v3-Inspector-MLX \
  --prompt "Inspect this short transcript for contradictions and return a concise verdict." \
  --max-tokens 256 \
  --temp 0.2

The --model value may be a local directory or a Hugging Face repository ID. For this repository, use the local directory when testing the exact files that were uploaded. If mlx-lm requests permission for tokenizer code, review the prompt and add --trust-remote-code only when required by the installed runtime.

Chat mode

python -m mlx_lm.chat \
  --model /Users/ilia/Models/NOESIS-Qwopus3.5-0.8B-v3-Inspector-MLX

The tokenizer's chat_template.jinja is used by mlx-lm when chat messages are formatted. The Inspector is intended for short, deterministic review tasks, so prefer low temperature and bounded max_tokens values.

OpenAI-compatible local server

python -m mlx_lm.server \
  --model /Users/ilia/Models/NOESIS-Qwopus3.5-0.8B-v3-Inspector-MLX \
  --host 127.0.0.1 \
  --port 8080

In another terminal:

curl http://127.0.0.1:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "system", "content": "You are the NOESIS Inspector. Return concise, evidence-based verdicts."},
      {"role": "user", "content": "Check this subtitle line for an obvious timing or translation risk."}
    ],
    "temperature": 0.2,
    "max_tokens": 256
  }'

The MLX-LM server implements a basic OpenAI-like API on localhost. It is suitable for local development and pipeline experiments, not for direct public exposure.

Python API

from mlx_lm import load, generate

model_dir = "/Users/ilia/Models/NOESIS-Qwopus3.5-0.8B-v3-Inspector-MLX"
model, tokenizer = load(model_dir)

messages = [
    {"role": "system", "content": "You are the NOESIS Inspector. Be concise and evidence-based."},
    {"role": "user", "content": "Review this dubbing decision and return PASS, WARN, or FAIL with one reason."},
]
prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
answer = generate(
    model,
    tokenizer,
    prompt=prompt,
    max_tokens=256,
    temp=0.2,
    verbose=True,
)
print(answer)

NOESIS role and routing

Inspector is a small specialist, not a replacement for the 9B Supervisor or DubCompress models. Use it for fast pre-checks, lightweight validation, and agent-quality inspection. Escalate complex reasoning, long-context supervision, and final dubbing decisions to the appropriate 9B NOESIS model.

Runtime notes

NOESIS sealed rules

• Original model preservation: this MLX repository is an original NOESIS model artifact, not an external repack.
• R-INSPECTOR-FAST-REVIEW: use Inspector for lightweight checks and escalate difficult judgments to Supervisor.
• R-MLX-LOCALHOST: the MLX-LM server must remain local unless a separately secured gateway is provided.
• R-CHAT-TEMPLATE-PRESERVE: preserve the shipped chat_template.jinja and tokenizer files together with the weights.
• R-APACHE-CLEAN: preserve the applicable upstream and NOESIS license notices.

References

License

See the repository license and upstream notices. The model is released as part of the NOESIS / AMAImedia original-trained model family. Redistribution must preserve the included metadata and attribution.