totally-not-an-llm/PuddleJumper-13b-V2
PuddleJumper-13b-V2 is a 13 billion parameter causal language model developed by totally-not-an-llm, created by merging the EverythingLM-V3-13b QLoRa and OpenOrca-Platypus2-13B models. With a 4096-token context length, it demonstrates balanced performance across various benchmarks, achieving an average score of 49.69 on the Open LLM Leaderboard. This model is suitable for general-purpose language tasks requiring a blend of reasoning and common sense.
Loading preview...
PuddleJumper-13b-V2: Merged 13B Language Model
PuddleJumper-13b-V2 is a 13 billion parameter language model developed by totally-not-an-llm. This model is a merge of two distinct components: the EverythingLM-V3-13b QLoRa and the OpenOrca-Platypus2-13B models, aiming to combine their respective strengths. It operates with a context length of 4096 tokens.
Key Capabilities & Performance
Evaluated on the Open LLM Leaderboard, PuddleJumper-13b-V2 shows a solid, general-purpose performance profile. Its average score across various benchmarks is 49.69.
Notable benchmark results include:
- ARC (25-shot): 57.0
- HellaSwag (10-shot): 81.06
- MMLU (5-shot): 58.3
- TruthfulQA (0-shot): 52.66
- Winogrande (5-shot): 72.45
While its GSM8K (5-shot) score is 3.64 and DROP (3-shot) is 22.74, the model's higher scores in areas like HellaSwag and Winogrande suggest proficiency in common sense reasoning and reading comprehension.
Use Cases
This model is well-suited for applications requiring a balanced understanding of language and general knowledge. Its merged architecture aims to provide a versatile foundation for tasks such as text generation, summarization, and question answering where a broad range of capabilities is beneficial.