fine-tuned

A small lab from Switzerland and South Africa has open-sourced Hemmingway-1, a 27B Apache-2.0 licensed fine-tune of Qwen3.8-27B specialized for creative writing, achieving high scores on EQ-Bench and internal benchmarks.

HuggingModels announces the release of Qwen3.6-35B-A3B-Uncensored-Genesis-Hermes-V7-GGUF, a fine-tuned and quantized version of the Qwen model.

This paper investigates the mismatch between evaluation and deployment behavior in fine-tuned language models. It introduces a method to locate and intervene on internal representations to close this gap.

Google DeepMind introduces Gemini 3.5 Flash Cyber, a lightweight cybersecurity model fine-tuned for vulnerability discovery and patching, with a limited-access pilot for governments and trusted partners.

Microsoft released Fara1.5-27B, a 27B parameter multimodal computer use agent for web browsers, fine-tuned from Qwen3.5-27B, capable of end-to-end web tasks via screenshots.

A fully uncensored cybersecurity-specialized AI model fine-tuned on exploits and pentesting data, designed to run locally on consumer hardware with multiple quantization options, offering expert offensive and defensive insights.

Tess-4-27B is a 27B reasoning model built on Qwen3.6-27B, post-trained on 64K-token long-context agentic traces with weight-scaled reasoning. It is designed for efficient, honest, and agentic task execution, available in open-source formats.

Empero AI releases Qwythos-9B-Claude-Mythos-5-1M-GGUF, a 9B parameter reasoning model fine-tuned on 500M+ tokens of Claude Mythos/Fable traces with chain-of-thought, achieving significant gains over Qwen3.5-9B and supporting 1M-token context via YaRN rope-scaling. The GGUF quantizations enable local inference on llama.cpp and compatible runtimes.

Empero AI releases Qwythos-9B, a fine-tuned reasoning model with 1M token context and uncensored capabilities, showing large benchmark improvements over its Qwen3.5-9B base.

A fine-tuned version of Gemma-4-12B, optimized for local coding and agentic tasks, achieving ~3.5x improvement over the base model on the tau2-bench telecom benchmark.

A developer built a pocket Charles Spurgeon AI assistant that runs fully offline on a fine-tuned Gemma model. It can answer theological questions, prepare sermons, and grade sermon drafts in Spurgeon's voice.

A writing-focused fine-tune of Google's Gemma 4 31B model, aiming for more natural English and better prose, with reduced refusals suitable for creative writing, translations, and roleplay.

NemoStation/Marlin-2B is a fine-tuned model based on Qwen3.5-2B for video-text-to-text tasks, supporting video captioning and temporal grounding.

SuperGemma4-26B-Uncensored-MLX-4bit-v2 is a fine-tuned and quantized variant of Google's Gemma 4 26B optimized for Apple Silicon, offering improved performance on code, reasoning, and tool-use tasks while maintaining faster inference speeds compared to the stock baseline.

HauhauCS releases an uncensored variant of Google's Gemma-4-E4B model with aggressive safety removal, featuring custom K_P quantizations optimized for quality preservation and broader hardware compatibility.