Eve-2-MoE-NanoCommit-272M - GGUF

Quantization Variants

Usage with Ollama

Usage with llama.cpp

Architecture

• Type: DeepSeek-style Mixture of Experts (MoE)
• Parameters: 272M total
• Layers: 12
• Hidden dim: 512
• Experts: 8 routed (top-2) + 1 shared per layer
• Context: 2048 tokens
• Tokenizer: GPT-2

Parent Model