gagan3012/MetaModelv2

TEXT GENERATIONConcurrency Cost:1Model Size:10.7BQuant:FP8Ctx Length:4kPublished:Jan 3, 2024License:apache-2.0Architecture:Transformer Open Weights Cold

gagan3012/MetaModelv2 is a 10.7 billion parameter hybrid language model, combining elements from VAGOsolutions/SauerkrautLM-SOLAR-Instruct and kyujinpy/Sakura-SOLAR-Instruct. This model is designed for general language understanding and generation tasks, demonstrating strong performance across various benchmarks. With a 4096-token context length, it offers robust capabilities for diverse applications requiring comprehensive text processing.

Loading preview...

MetaModelv2: A Hybrid 10.7B Language Model

MetaModelv2, developed by gagan3012, is a 10.7 billion parameter language model built as a hybrid of two existing models: VAGOsolutions/SauerkrautLM-SOLAR-Instruct and kyujinpy/Sakura-SOLAR-Instruct. This approach leverages the strengths of its constituent models to deliver a versatile and capable LLM.

Key Capabilities & Performance

Evaluated on the Open LLM Leaderboard, MetaModelv2 demonstrates solid performance across a range of benchmarks, indicating its proficiency in various reasoning and language understanding tasks. Its average score of 74.24 highlights its balanced capabilities. Specific benchmark results include:

  • ARC (25-shot): 71.08
  • HellaSwag (10-shot): 88.56
  • MMLU (5-shot): 66.29
  • TruthfulQA (0-shot): 71.94
  • Winogrande (5-shot): 83.11
  • GSM8K (5-shot): 64.44

With a context length of 4096 tokens, MetaModelv2 can process moderately long inputs, making it suitable for tasks requiring a decent understanding of conversational flow or document content.

Good for

  • General-purpose language generation: Its balanced performance suggests suitability for a wide array of text generation tasks.
  • Reasoning and question answering: Scores on ARC and MMLU indicate capabilities in logical inference and knowledge-based QA.
  • Applications requiring moderate context: The 4096-token context window supports tasks like summarization of short documents or extended dialogue.