I Tested the new Qwen 3.5 2B model on 128MB UHD. i used @ollama to run it locally. Key takeaway : If you want to run a local model for any reason (privacy or whatever) on a UHD graphics card make sure you turn off thinking using the command /set nothink to speed up inference.…

I used to use llama 3.2 for text generation and spelling checks in ollama locally. Now, I tried ifioravanti/mistral-grammar-checker, and it's astonishingly good.

So like, I use Ollama like I have no usage limits, with Deepseek v4.1 Flash. And honestly, I never run out. I use it all day, having it work on Gaming accessibility mod stuff. No I am not releasing anything yet. And I just plain never run out of AI. #AI #Ollama

I built a semantic search that returns similar movies using @weaviate_io, and implements RAG with the Ollama local model. https://t.co/XJTz6CPfqi

No luck. I tried the new gpt-oss-20b local model on my M2 Macbook Air 8 GB RAM and it just slowed everything down and took forever to respond. Unfortunate. Anyone know any good local models I can run using Ollama with my macbook? https://t.co/Bt0rsSbkn9

Someone said you can finetune running llama.cpp to run models faster.. I tried and failed, running ollama was faster for me.. hopefully I can comeback in future and see what Mistake I made..or you can share some tips if you would have done differently https://t.co/zBDxPACgWm

If you have a single gpu and are making gen ai stuff, you can use a local llm to write prompts for generations. Been using qwen 3.8 27b-q4_k_m to prompt more elaborate stuff. I use Pi with ollama to prompt, then stop ollama to eject the llm from vram so I can then generate the v…

So this is whacky. I use LMStudio and Ollama for local models on my Mac Mini M4. Recently the new qwen3 models dropped and I loaded up qwen3-30B-a3b and tested it. And it's the first local model that reliably executes tool calls.

I didn't install an AI agent to help me troubleshoot Kubernetes. Instead, I took the hard route and built one. One's infrastructure experience might say, "Install the agent, wire it in, move on." This spring at the Recurse Center, I wrote a chatbot agent and its own MCP server.…

I wanted to try running my coding agent with a local model this morning, but ran into slow responses with Ollama. That led to the core insight: on macOS, Ollama adds overhead and doesn’t give a clean Metal‑accelerated path, so I built a minimal MLX server with @OpenAI codex.

I recently subscribed to @ollama Pro and have been using their library of models along with @LangChain. Yesterday while reading the docs I learned that you can use the models with Claude! The Claude setup walkthrough actually maps their models to the equivalent Ollama offering a…

Local 8B model on my RTX 2060 found the overflow in 3 seconds. Switched to Codex in the same chat. It fixed the code, wrote the tests, and generated the chart. Real diffs. Real tests. Phone + laptop. NivKode #LocalLLaMA #Ollama https://t.co/M7ln6Z9Y8s

Tried @ollama launch + Clauda code + GLM Flash today, surely, clauda code has alot of potentials working with its own models tailored for all scenarios. However, working with local model was promising until i tried generate Rust code for Tauri + Dioxus. #LOCALMODEL, #OLLAMAL…

🚀 I built a webserver powered entirely by an LLM. Every HTTP/1.1 response — headers, body, everything — is generated in real-time by a local model (via Ollama). No traditional backend logic. 🧠 Pure AI. 📡 Pure shell. 🔗 https://t.co/lo6DPoT6Yy #AI #LLM #DevOps #bash #webdev