YamshadowExperiment28-7B Overview

YamshadowExperiment28-7B is a 7 billion parameter language model developed by Maxime Labonne. It was created by merging two existing models, automerger/YamShadow-7B and yam-peleg/Experiment28-7B, using a specific slerp merge method. The model leverages a context window of 4096 tokens.

Key Characteristics & Performance

• Automated Merge: This model is a product of an automated merging process, combining the strengths of its constituent models.
• Leaderboard Performance: As of April 8, 2024, YamshadowExperiment28-7B is noted as the best-performing 7B model on the Open LLM Leaderboard. However, the creator advises caution, suggesting this high performance might be a sign of overfitting to benchmarks.
• Evaluation: Beyond the Open LLM Leaderboard, the model has also been evaluated using EQ-bench and Nous's LLM AutoEval, with results available in the original README.
• Chat Template: It is recommended for use with the Alpaca chat template, which is compatible with tools like LM Studio.

Considerations for Use

While demonstrating strong benchmark results, users should be aware that the model may occasionally produce repetitive "INST" outputs, and its exceptional leaderboard standing could indicate a degree of benchmark overfitting. Quantized GGUF versions are available for local deployment.