ChaoticNeutrals/Prima-LelantaclesV5-7b

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Feb 21, 2024License:otherArchitecture:Transformer0.0K Cold

ChaoticNeutrals/Prima-LelantaclesV5-7b is a 7 billion parameter language model created by ChaoticNeutrals, merged using the DARE TIES method from Test157t/Pasta-Lake-7b and Test157t/Prima-LelantaclesV4-7b-16k. This model demonstrates surprisingly good performance at an extended context length of 16384 tokens, exceeding typical Mistral-based expectations. It achieves an average score of 73.09 on the Open LLM Leaderboard, making it suitable for general reasoning and language understanding tasks.

Loading preview...

Prima-LelantaclesV5-7b: A DARE TIES Merged Model

ChaoticNeutrals/Prima-LelantaclesV5-7b is a 7 billion parameter language model developed by ChaoticNeutrals, created through a DARE TIES merge. This model combines the strengths of Test157t/Pasta-Lake-7b and Test157t/Prima-LelantaclesV4-7b-16k.

Key Capabilities & Performance

  • Extended Context Window: Unexpectedly performs well at a 16384-token context length, which is a notable improvement over typical Mistral-based models that usually operate around 8192 tokens.
  • Merged Architecture: Utilizes the DARE TIES merge method, indicating a focus on combining and enhancing capabilities from its constituent models.
  • Leaderboard Performance: Achieves a competitive average score of 73.09 on the Open LLM Leaderboard, with specific scores including:
    • AI2 Reasoning Challenge (25-Shot): 70.65
    • HellaSwag (10-Shot): 87.87
    • MMLU (5-Shot): 64.52
    • TruthfulQA (0-shot): 68.26
    • Winogrande (5-shot): 82.40
    • GSM8k (5-shot): 64.82

Use Cases

This model is well-suited for applications requiring:

  • General language understanding and generation.
  • Tasks benefiting from an extended context window, such as summarizing longer documents or maintaining conversational coherence over more turns.
  • Reasoning tasks, as indicated by its performance on benchmarks like AI2 Reasoning Challenge and GSM8k.