Multiverse Computing has released GGUF versions of its Comet 60B model, a compressed adaptation of Nemotron 3.5 Super. These models, available in Q4_K_M and Q8_0 quantization levels, are designed to run on llama.cpp. They retain the Multi Token Prediction (MTP) head, enabling speculative decoding with the `--spec-type draft-mtp` flag in llama-server, though they are text-only and do not include the vision tower. AI
IMPACT Enables wider accessibility and experimentation with a powerful language model on local hardware.
RANK_REASON Release of a quantized model format for local inference.
- ATC code A12
- Comet 60B
- GGUFs
- llama.cpp
- Multi Token Prediction
- Multiverse Computing
- Nemotron 3.5 Super
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →