Posted inLocal LLMs
5 days of work with qwen3.8:27b
Exploring the Capabilities of Qwen 3.8: A Journey into 3D Game Development Over the past five days, I have been testing and pushing the boundaries of the Qwen 3.8 model,…
Posted inLocal LLMs
Is Qwen3.8 27B more powerful than Qwen3.8 Flash? I’m confused about the pricing.
Evaluating the Capabilities and Pricing of Qwen 3.8 27B versus Qwen 3.8 Flash: A Detailed Analysis In the rapidly evolving landscape of artificial intelligence and large language models (LLMs), understanding…
Posted inLocal LLMs
DeepCogito is extremely impressive. One shot solved the rotating hexagon with bouncing ball prompt on my M2 MBP 32GB RAM config personal laptop.
DeepCogito: A Remarkable Open-Source AI Model Demonstrating Advanced Capabilities on a Personal Laptop In recent developments within the AI community, the open-source model DeepCogito has garnered significant attention for its…
Posted inLocal LLMs
I ran 26 local LLMs through an 8 level “agentic failure mode” gauntlet (tool calling, on an M1 Max). Capability benchmarks lie about who can actually run an agent loop. All local, llama.cpp + Metal, GGUF. 8 tests, 3 reps each, same prompts and seeds for every model thinking OFF
Comprehensive Evaluation of Local Language Models in Agentic Failure Scenarios Introduction In the rapidly evolving landscape of large language models (LLMs), capabilities often get quantified through benchmarks emphasizing tasks like…
Posted inLocal LLMs
Qwen3 8B suddenly gets stuck repeating the same phrase help me outtttt :(
Experiencing Repetition Issues with Qwen3 8B: Troubleshooting and Solutions In the realm of deploying large language models locally, encountering unexpected behavior can be both frustrating and perplexing. Recently, users working…
Posted inLocal LLMs
friends don’t let friends quantize their kv cache
The Importance of Appropriate Quantization Levels for KV Cache in AI Deployments In the realm of deploying large language models and other AI systems, optimizing memory and computational efficiency is…
Posted inLocal LLMs
I tested Opus 4.7 vs DeepSeek V4 Flash vs Local Qwen3.6 27B as coding agents. The gaps were much smaller than I expected, and harness is as important as model intelligence.
Optimizing AI Coding Assistants: A Comparative Review of Opus 4.7, DeepSeek V4 Flash, and Local Qwen 3.6 27B Introduction The rapid evolution of large language models (LLMs) has transformed the…
Posted inLocal LLMs
Why is Europe sitting out the open source LLM race while the China dominate?
Europe’s Absence in the Open Source Large Language Model Race: A Closer Look In recent years, the artificial intelligence (AI) landscape has experienced rapid evolution, with open-source Large Language Models…
Posted inLocal LLMs
Abliterated Qwen 3.8 flash can do anything
Unlocking Unparalleled Capabilities with Advanced Qwen 3.8 Flash Model Optimization In the rapidly evolving landscape of machine learning and AI, enthusiasts and developers constantly seek ways to enhance model performance,…
Posted inLocal LLMs
Working on an Architecture that makes even 0.8B usable for agentic code
Advancing AI Architecture: Enabling Efficient Utilization of Large-Scale Models for Agentic Tasks In the rapidly evolving landscape of artificial intelligence, harnessing the full potential of large-scale models remains a paramount…