Model Overview
The anthracite-org/magnum-v3-9b-customgemma2 is a 9 billion parameter language model developed by Anthracite, fine-tuned on Google's Gemma-2-9b base. This model is the tenth in a series focused on achieving the prose quality of Claude 3 models (Sonnet and Opus).
Key Capabilities & Features
- Claude 3 Prose Replication: Specifically designed and fine-tuned to emulate the writing style and quality of Claude 3 models.
- Custom Instruction Tuning: Utilizes a
customgemma2 prompt format, enabling robust system prompt support for more controlled and nuanced interactions. - Training Datasets: Fine-tuned on a diverse set of datasets including
anthracite-org/stheno-filtered-v1.1, anthracite-org/kalo-opus-instruct-22k-no-refusal, anthracite-org/nopm_claude_writing_fixed, Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned, and Epiculous/SynthRP-Gens-v1.1-Filtered-n-Cleaned. - Context Length: Supports a sequence length of 8192 tokens.
Performance & Use Cases
While optimized for prose quality, the model's performance on the Open LLM Leaderboard shows an average score of 19.02, with specific scores like 35.61 on MMLU-PRO (5-shot) and 34.12 on BBH (3-shot). This model is particularly well-suited for applications where high-quality, human-like text generation and conversational flow are paramount, especially in scenarios requiring a sophisticated writing style akin to Claude 3.