WizardLMTeam/WizardLM-70B-V1.0
WizardLM-70B-V1.0 is a 69 billion parameter instruction-following language model developed by WizardLMTeam, based on the Llama 2 architecture. It is specifically designed to follow complex instructions and demonstrates strong performance across various benchmarks, including general language understanding, mathematical reasoning, and code generation. This model excels at multi-turn conversations and is suitable for applications requiring nuanced instruction adherence.
Loading preview...
WizardLM-70B-V1.0 Overview
WizardLM-70B-V1.0 is a 69 billion parameter instruction-following language model developed by WizardLMTeam. It is built upon the Llama 2 architecture and is specifically trained to enhance its ability to follow complex instructions, supporting multi-turn conversations. This model is part of a broader family of WizardLM models, which also includes specialized versions like WizardCoder for code generation and WizardMath for mathematical reasoning.
Key Capabilities & Performance
- Instruction Following: Designed to excel at understanding and executing complex instructions, making it highly adaptable for various tasks.
- Multi-turn Conversation: Supports and maintains context across extended dialogues, adopting the Vicuna prompt format.
- General Benchmarks: Achieves a score of 7.78 on MT-Bench and 92.91% on AlpacaEval, indicating strong general performance.
- Mathematical Reasoning: Demonstrates robust mathematical capabilities with 77.6% on GSM8k.
- Code Generation: Shows proficiency in code tasks, scoring 50.6 pass@1 on HumanEval.
Good for
- Applications requiring a large language model that can accurately follow detailed and intricate instructions.
- Building conversational agents that need to maintain context and respond coherently over multiple turns.
- Tasks involving general language understanding, mathematical problem-solving, and code-related queries.