Qwythos-9B

Qwythos-9B-v2-GGUF is a quantized collection of the Qwythos-9B-v2 multimodal model, optimized for GGUF runtimes like llama.cpp and Ollama. This version specifically addresses looping behaviors found in the base model using Final-Token Preference Optimization (FTPO) while restoring the native multi-token prediction (MTP) head for improved performance.

Qwythos-9B-v2-GGUF is a quantized collection of the Qwythos-9B-v2 multimodal model, optimized for GGUF runtimes like llama.cpp and Ollama. This version specifically addresses looping behaviors found in the base model using Final-Token Preference Optimization (FTPO) while restoring the native multi-token prediction (MTP) head for improved performance.

• 01Looping/degeneration eliminated with 0% repetition rate
• 02Restored native multi-token prediction (MTP) head for speculative decoding
• 03Multimodal vision capabilities via Qwen3.5 architecture
• 041M-token long-context window using YaRN
• 05Intentionally uncensored reasoning model
• 06Optimized for GGUF runtimes including llama.cpp, Ollama, and LM Studio