itsZyn/ZynDwarf-1.0

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Aug 24, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

ZynDwarf-1.0 is a 350 million parameter base model from the LFM2.5 family by Liquid AI, designed for on-device deployment. This hybrid model features 16 layers, a 32,768 token context length, and was trained on 28 trillion tokens. It is primarily intended for heavy fine-tuning for language-specific or domain-specific applications, supporting English and eight other languages.

Loading preview...

LFM2.5-350M-Base Overview

LFM2.5-350M-Base is a 350 million parameter pre-trained base model developed by Liquid AI, part of their LFM2.5 family of hybrid models. It is specifically engineered for on-device deployment, building upon the LFM2 architecture with extensive pre-training and reinforcement learning.

Key Capabilities & Features

  • Compact Size: With 350 million parameters, it is optimized for efficient deployment on edge devices.
  • Extended Context Window: Supports a substantial context length of 32,768 tokens.
  • Multilingual Support: Trained to understand and generate text in English, Arabic, Chinese, French, German, Japanese, Korean, Portuguese, and Spanish.
  • Robust Training: Pre-trained on a massive 28 trillion tokens, ensuring broad linguistic understanding.
  • Hybrid Architecture: Incorporates 16 layers, including 10 double-gated LIV convolution blocks and 6 GQA blocks.

Recommended Use Cases

This base model is particularly suited for scenarios requiring significant customization and specialization:

  • Heavy Fine-Tuning: Ideal for continued pre-training or supervised fine-tuning for specific tasks.
  • Domain-Specific Assistants: Training on proprietary data to create medical, legal, or other specialized AI assistants.
  • Language-Specific Applications: Adapting the model for optimal performance in particular languages, such as Japanese.
  • Novel Post-Training Experimentation: A strong foundation for exploring new fine-tuning and post-training methodologies.

LFM2.5-350M-Base is supported by various inference frameworks, including Transformers, vLLM, llama.cpp, and MLX, facilitating flexible deployment and integration.

Finetunes
1 models