S5-nano Ships GGUF With Top Brazilian Scores
S5-nano, Maritaca AI's 4-billion-parameter model, posts the top Brazilian-benchmark scores among models its size while keeping tool-calling and structured-output abilities. GGUF builds bring the model to llama.cpp users for local runs. — huggingface.co
Emma-6 GGUF Quants Bring Italian Model to llama.cpp
An unofficial conversion packages Egomnia's Italian-first Emma-6, a model of roughly 1.5 billion parameters, as GGUF quantizations for local llama.cpp inference. — huggingface.co
Qwen3.8 Flash GGUF Runs a Third Faster
A community GGUF for Qwen3.8-Flash-Next trims experts from each layer so it fits on a small-GPU PC, decoding about a third faster than unsloth's reference build while improving Japanese output quality. — huggingface.co
LFM3 Fine-Tune Adds Reasoning Traces to Coding
A community LoRA fine-tune of LiquidAI's LFM3-1.7B, trained on seventy high-reasoning traces, steers the small model toward a think-then-answer style in coding, distributed as MLX and GGUF weights. — huggingface.co
Kitchoon 26B MoE GGUF Aims at Roleplay Writing
The final Kitchoon release ships GGUF quantizations tuned for roleplay, with the Anthropos v3 LoRA baked in to cut stock phrasing and encourage more varied responses. — huggingface.co