A strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token.
100% Uncensored (9/9 SecOps & NSFW, 85.7% Precision) StepFun Step-3.5-Flash 196B Sparse MoE (11B active, 262K ctx, i1-IQ2_M Abliterated, THINK_LEN=0) — 173.4 tok/s decode in 60.6 GiB VRAM on H100 80GB.
ollama run qwen3.5:9b
ollama pull qwen3-embedding
New state of the art 70B model. Llama 3.3 70B offers similar performance compared to the Llama 3.1 405B model.
A strong multi-lingual general language model with competitive performance to Llama 3.
https://huggingface.co/pacozaa/TinyLlama-1.1B-intermediate-step-1431k-3T-GGUF
Nous Hermes 4.3 36B parameters with thinking and tools enabled
Exploration 1
From-scratch THOX base model: 123M Llama-arch decoder pretrained by THOX (2,000 steps, AMD MI300X); 49.3% HumanEval / 44.6% MBPP zero-shot. Q4_K_M GGUF, edge-sized. A base/completion model, not a chat assistant. Apache-2.0.
Doc G is a high-energy Environmental Science teacher persona for Llama 3.2 who is obsessed with efficiency and saving money. He delivers punchy science lessons mixed with lectures about wasting electricity.
REAP-stability-floor 184-expert prune of Ornith-1.5-35B-A3B (~27B, A3B active, top-8) — the recommended arm of the pair. Beats the 256-expert teacher on LiveCodeBench v6 (+10.4pp)
The "Home" model is a fine tuning of the StableLM-Zephyr-3B model. It achieves a score of 97.11% score for JSON function calling accuracy.
Instruct version of the large language model YandexGPT 5 Lite with 8B parameters with a context length of 32k tokens. (quantised version of Q5_K_M)
zerofata/MS3.2-PaintedFantasy-Visage-v3-34B
MethGPT is fork of UnknownFish/waltergpt with updated system prompt and updated LLAMA (from 2 to 3)
zerofata/MS3.2-PaintedFantasy-Visage-v2-33B