aayanmishra-ml/Atlas-Flash-1.5B-Preview

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jan 26, 2025License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Atlas-Flash-1.5B-Preview is a 1.5 billion parameter model from aayanmishra-ml, built on Deepseek's R1 distilled Qwen models. It is designed for advanced reasoning, contextual understanding, and domain-specific expertise, excelling particularly in coding, conversational AI, and STEM problem-solving. This model supports accurate code generation, natural multi-turn dialogue, and complex problem-solving in mathematics, physics, and engineering. With a 32768 token context length, it offers robust performance for a wide range of technical applications.

Loading preview...

Atlas-Flash-1.5B-Preview: A Versatile Reasoning Model

Atlas-Flash is the inaugural model in the Atlas family by aayanmishra-ml, a new generation of AI systems built upon Deepseek's R1 distilled Qwen-1.5B models. This 1.5 billion parameter model is engineered for advanced reasoning, contextual understanding, and domain-specific expertise, serving as a successor to the Athena-2 project.

Key Capabilities

  • Improved Coding: Excels in code generation, debugging, explanation, and documentation across multiple programming languages.
  • Advanced Conversational AI: Provides natural, context-aware, and coherent multi-turn dialogue for both informal chat and task-specific queries.
  • Proficiency in STEM Domains: Capable of solving complex problems in mathematics, physics, and engineering, offering clear explanations of intricate concepts.

Training and Methodology

Atlas-Flash underwent extensive training on diverse, high-quality datasets, including BAAI/TACO for language understanding, rubenroy/GammaCorpus-v1-70k-UNFILTERED for real-world language examples, and codeparrot/apps for programming tasks. The training incorporated multi-stage fine-tuning and synthetic data augmentation to enhance generalization and specialization, particularly for coding, general language, and STEM domains.

Good For

  • Software Development: Automating code tasks, debugging, and documentation.
  • Conversational AI: Building intelligent chatbots and virtual assistants.
  • STEM Problem-Solving: Assisting with mathematical, physics, and engineering challenges.
  • Education and Knowledge Assistance: Explaining complex concepts and acting as a virtual tutor.

Finetunes
1 models