Whisper-Large-V3-Turbo-Darwin-NOESIS-FP16

Encoder-averaged Whisper: stronger encoder (avg v3-full + Turbo) with v3-full decoder (32 layers, best quality).

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: v14.7
• Release date: 2026-04

---

⚠️ License notice

This model is a derivative of OpenAI Whisper (openai/whisper-large-v3 and openai/whisper-large-v3-turbo), which are released under the MIT License. The encoder-averaged merge is distributed under the same MIT terms.

See the LICENSE file in this repository for the full text.

---

Model summary

Property	Value
Architecture	WhisperForConditionalGeneration
Parameters	~1.5B
Encoder layers	32 (averaged v3-full + Turbo)
Decoder layers	32 (from v3-full, best quality)
d_model	1280
Vocab size	51 866
Format	safetensors
Languages	99 (multilingual)

---

Merge strategy

Component	Strategy	Source
Encoder (32 layers)	Weighted average (v3=0.55, turbo=0.45)	Both models
Decoder (32 layers)	Kept as-is	whisper-large-v3-full

Turbo was distilled from v3-full → task vectors are compatible. Averaging encoder combines the full model's multilingual robustness with Turbo's distilled representations. Decoder from v3-full preserves maximum generation quality (32 layers vs Turbo's 4).

---

How to use

from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
import torch

model_id = "AMAImedia/NOESIS-Whisper-Large-V3-Darwin"

processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
)

---

NOESIS context

Used as multilingual ASR teacher in NOESIS ASR-10B knowledge distillation pipeline. Stronger encoder improves soft-label quality across 150+ dubbing languages.

---

Provenance

Full merge trace in merge_provenance.json.

---

Citation

@misc{noesis_whisper_darwin,
  title     = {NOESIS-Whisper-Large-V3-Darwin},
  author    = {Bolotnikov, Ilia},
  year      = {2026},
  publisher = {AMAImedia},
  url       = {https://amaimedia.com}
}