An uncensored version of the open-source language model GPT-OSS 120B has been released. This "aggressive" version is designed not to refuse any query.
Technical Specifications
- Total parameters: 117 billion
- Active parameters: ~5.1 billion (MoE with 128 experts, top-4 routing)
- Context window: 128K
- Format: MXFP4 (model's native precision)
The model is provided in a single 61GB file and can be run on a single H100 GPU. For configurations with lower VRAM, it is possible to offload the MoE layers to the CPU via the --n-cpu-moe N option in llama.cpp. It works with llama.cpp, LM Studio, Ollama, and others.
Other Models
Uncensored versions of GPT-OSS 20B, GLM 4.7 Flash, and Qwen3 8b VL are also available.
๐ฌ Commenti (0)
๐ Accedi o registrati per commentare gli articoli.
Nessun commento ancora. Sii il primo a commentare!