AMAImedia/NOESIS-Gemma3-1B-BF16

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kPublished:Aug 16, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

AMAImedia's NOESIS-Gemma3-1B-BF16 is a 1 billion parameter Gemma 3 instruction-tuned model, fine-tuned as a NOESIS LoRA-merged Scenema student. This BF16 precision model serves as a recipe-validation companion and lightweight test student for the primary 4B Scenema student bundle, specifically designed for the Scenema-DiT audio cross-attention text encoder pipeline. It is optimized for professional multilingual dubbing automation within the NOESIS framework, projecting its hidden states to a 3840-dimension target for audio cross-attention.

Loading preview...

Model Overview

NOESIS-Gemma3-1B-BF16 is a 1 billion parameter model developed by AMAImedia as part of their NOESIS Professional Multilingual Dubbing Automation Platform. It is a fine-tuned variant of Google's gemma-3-1b-it base model, incorporating merged LoRA adapters (cinema r=8 and autoif-v2 r=16) into a BF16 precision model.

Key Capabilities & Role

  • Recipe-Validation Student: This model's primary role is to serve as a lightweight test and validation student for the Scenema-DiT audio cross-attention text encoder pipeline. It allows for efficient smoke-testing of scenema_adapter recipes before scaling to its larger 4B sibling.
  • Audio Cross-Attention Text Encoder: It is specifically designed to generate per-token text embeddings (projected to 3840 dimensions) required by the Scenema-DiT audio cross-attention mechanism, crucial for professional audio dubbing.
  • BF16 Precision: The model is a fully merged, single-shard BF16 master, intended as the authoritative source for any future re-quantization or distillation.

Use Cases

  • Development & Experimentation: Ideal for developers working on the NOESIS framework, particularly for validating and experimenting with scenema_adapter configurations in a resource-efficient manner.
  • Multilingual Dubbing Automation: While a validation model, its lineage and design are directly tied to the professional audio dubbing automation pipeline, making it relevant for understanding the text encoding component of such systems.

Finetunes
1 models