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-Qwopus3.5-9B-Supervisor-v3.5-MLX

Original NOESIS MLX Supervisor specialist for long-context review, instruction supervision, response judging, routing decisions, and multi-agent coordination. This repository contains the MLX safetensors conversion of the BF16-trained NOESIS-Qwopus3.5-9B-Supervisor-v3.5 model. It is designed for Apple Silicon inference through mlx-lm.

This is the MLX counterpart of the published NOESIS-Qwopus3.5-9B-Supervisor-v3.5-BF16 model card. The BF16 repository is the primary Transformers/GGUF distribution; this repository is the local Apple-Silicon MLX distribution of the same original NOESIS-trained specialist.

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

Property	Value
Model role	Supervisor, judge, routing and multi-agent coordination specialist
Architecture	Qwen3_5ForCausalLM, qwen3_5
Parameters	Approximately 9B
Layers / hidden size	32 / 4096
Intermediate size	12288
Context limit in config	262,144 positions
Stored dtype	BF16 configuration; MLX safetensors
Weight shards	4 files, approximately 17.93 GB total
Tokenizer	Qwen3.5 tokenizer with chat_template.jinja
Runtime	mlx-lm on Apple Silicon
NOESIS version	v16.1
Release date	2026-08-26

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.

Language support is inherited from the Qwen3.5/Qwen3 family and should not be interpreted as identical production quality for every language. For NOESIS dubbing, Supervisor decisions remain subject to the appropriate translation, ASR, timing, and human-listening gates.

Repository contents

NOESIS-Qwopus3.5-9B-Supervisor-v3.5-MLX/
├── README.md
├── model-00001-of-00004.safetensors  # MLX shard 1
├── model-00002-of-00004.safetensors  # MLX shard 2
├── model-00003-of-00004.safetensors  # MLX shard 3
├── model-00004-of-00004.safetensors  # MLX shard 4
├── model.safetensors.index.json
├── config.json
├── generation_config.json
├── tokenizer.json
├── tokenizer_config.json
└── chat_template.jinja

The four shards and tokenizer files must remain together. Do not rename individual shards or remove model.safetensors.index.json; the loader uses the index to reconstruct the complete model.

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-Qwopus3.5-9B-Supervisor-v3.5-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-Qwopus3.5-9B-Supervisor-v3.5-MLX \
  --prompt "Review the following dubbing decision and return PASS, WARN, or FAIL with evidence." \
  --max-tokens 512 \
  --temp 0.2

For long prompts, reduce --prefill-step-size if memory spikes and set --max-kv-size to a deliberate bound. The configured context maximum is an architectural limit, not a promise that every Mac can hold a 262k-token KV cache.

Chat mode

python -m mlx_lm.chat \
  --model /Users/ilia/Models/NOESIS-Qwopus3.5-9B-Supervisor-v3.5-MLX

The tokenizer's chat_template.jinja is used when messages are formatted. For repeatable supervision, use a clear system instruction, a bounded output length, and a low temperature.

OpenAI-compatible local server

python -m mlx_lm.server \
  --model /Users/ilia/Models/NOESIS-Qwopus3.5-9B-Supervisor-v3.5-MLX \
  --host 127.0.0.1 \
  --port 8080

Request it from 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 Supervisor. Judge only from the supplied evidence."},
      {"role": "user", "content": "Review this candidate translation and identify any timing or semantic risk."}
    ],
    "temperature": 0.2,
    "max_tokens": 512
  }'

The MLX-LM server provides a basic OpenAI-compatible local endpoint. It is intended for local development and pipeline experiments and should not be exposed directly to the public Internet.

Python API

from mlx_lm import load, generate

model_dir = "/Users/ilia/Models/NOESIS-Qwopus3.5-9B-Supervisor-v3.5-MLX"
model, tokenizer = load(model_dir)

messages = [
    {"role": "system", "content": "You are the NOESIS Supervisor. Return a concise verdict with evidence."},
    {"role": "user", "content": "Evaluate this candidate dubbing take against the source constraints."},
]
prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
answer = generate(
    model,
    tokenizer,
    prompt=prompt,
    max_tokens=512,
    temp=0.2,
    verbose=True,
)
print(answer)

NOESIS role and routing

Supervisor is the long-context judge and coordination specialist. It is not the fast Inspector and it is not the slot-fitting DubCompress specialist. In the full NOESIS pipeline, Supervisor receives structured evidence from translation, ASR, speaker, emotion, and timing stages and returns a bounded review or routing decision.

Task	Recommended route
Fast sanity check and compact verdict	Inspector 0.8B MLX
Long-context review, response judging, routing, multi-agent coordination	Supervisor 9B MLX
Translation compression and target-slot fitting	DubCompress 9B MLX
Primary text generation for unrestricted creative writing	Use only when the pipeline explicitly selects Supervisor

The model should not invent an audio or timing result from text alone. If evidence is absent, return WARN or state that the check is not measurable.

Runtime and memory notes

Item	Guidance
Weights	Approximately 17.93 GB across four safetensors shards.
Unified memory	Leave additional memory for the OS, tokenizer, and KV cache; required memory is higher than the raw file size.
Context	262,144 positions in configuration; practical context depends on Mac unified memory and KV-cache settings.
Sampling	Low temperature and bounded max_tokens are recommended for judging.
Chat format	Preserve and use chat_template.jinja.
Server	Bind to 127.0.0.1 for local-only use.
Quality	A successful MLX load is a technical smoke test, not a production-quality gate.

NOESIS sealed rules

• R-SUPERVISOR-ALWAYS-CONNECTED: keep the Supervisor route available to the dubbing pipeline and do not orphan its model artifact.
• R-SUPERVISOR-EVIDENCE-BOUND: Supervisor must judge supplied evidence and must not fabricate unobserved audio, timing, or visual facts.
• R-SUPERVISOR-LONGCTX: use the model for long-context review and coordination; use Inspector for lightweight checks.
• R-MLX-LOCALHOST: the MLX-LM server is a local development service and must not be exposed without a separate secure gateway.
• R-CHAT-TEMPLATE-PRESERVE: preserve the tokenizer and shipped chat template with all weight shards.
• R-APACHE-CLEAN: preserve the applicable upstream and NOESIS license notices.

References

License

See the repository license and upstream notices. This MLX model is an original NOESIS-trained artifact released by AMAImedia; redistribution must preserve the model metadata and attribution.