windows_error23@reddit
I’m curious on how its vision and ocr would compare to qwen 3.5 9b.
windows_error23
google/gemma-4-12B · Hugging Face
Posted by jacek2023@reddit | LocalLLaMA | View on Reddit | 302 comments
A man arrives at the office in the morning where several of his co-workers are already chatting next to the coffee machine. He listens to them for a bit, then suddenly asks: "How big are penguins?"
Posted by Aldaron23@reddit | Jokes | View on Reddit | 119 comments
windows_error23@reddit
Sorry I’m not a native English speaker, can someone explain?
Gemma 4 has been released
Posted by jacek2023@reddit | LocalLLaMA | View on Reddit | 702 comments
windows_error23@reddit
Yes to audio modality. I wonder how good it’ll be
DeepSeekOCR & codefuse-ai/F2LLM-v2 are ready on llama.cpp
Posted by pmttyji@reddit | LocalLLaMA | View on Reddit | 5 comments
windows_error23@reddit
I saw a draft PR for it so I guess not yet.
Gemini Pro leaks its raw chain of thought, gets stuck in an infinite loop, narrates its own existential crisis, then prints (End) thousands of times
Posted by Powerful-Signal6312@reddit | LocalLLaMA | View on Reddit | 95 comments
windows_error23@reddit
It’s consistent with other reports. I also had a similar doom loop issue but it was with 3.0 pro on gemini cli. So I don’t doubt it too strongly.
OpenAI to acquire Astral
Posted by Useful-Macaron8729@reddit | Python | View on Reddit | 391 comments
windows_error23@reddit
Still, that change was as a result of pressure beyond the scope / magnitude of uv and ruff so it doesn’t seem likely the same thing will happen here.
OpenAI to acquire Astral
Posted by Useful-Macaron8729@reddit | Python | View on Reddit | 391 comments
windows_error23@reddit
Codex isn’t bad.
OpenAI to acquire Astral
Posted by Useful-Macaron8729@reddit | Python | View on Reddit | 391 comments
windows_error23@reddit
Personally, I’ll keep using uv and ruff until they actually do something bad if they ever will.
OpenAI to acquire Astral
Posted by Useful-Macaron8729@reddit | Python | View on Reddit | 391 comments
windows_error23@reddit
What’s the issue?
OpenAI to acquire Astral
Posted by Useful-Macaron8729@reddit | Python | View on Reddit | 391 comments
windows_error23@reddit
Exciting.
Breaking : The small qwen3.5 models have been dropped
Posted by Illustrious-Swim9663@reddit | LocalLLaMA | View on Reddit | 334 comments
windows_error23@reddit
I wonder why they keep increasing the parameter count slightly each generation
Qwen3 VL 30b a3b is pure love
Posted by Njee_@reddit | LocalLLaMA | View on Reddit | 91 comments
windows_error23@reddit
But the model is originally in bf16? If my understanding is correct, the fp32 mmproj is for people with hardware that doesn't support bf16 so that they can use it in full precision as an alternative instead of the quantized f16. Could be wrong on this.
Qwen3-VL GGUF!
Posted by khubebk@reddit | LocalLLaMA | View on Reddit | 52 comments
windows_error23@reddit
Thanks. Which mmproj quantization are you running?
Qwen 3 !!!
Posted by ResearchCrafty1804@reddit | LocalLLaMA | View on Reddit | 460 comments
windows_error23@reddit
I wonder what happened to the 15b MOE.
Google Ironwood TPU (7th generation) introduction
Posted by zimmski@reddit | LocalLLaMA | View on Reddit | 76 comments
windows_error23@reddit
I'm confused. Why disclose specs in such detail then.
Samsung introduces Gauss2: A Multimodal Generative AI model in three sizes (Compact, Balanced, Supreme)
Posted by Balance-@reddit | LocalLLaMA | View on Reddit | 16 comments
windows_error23@reddit
Maybe because it's possibly local?
Wen GGUF?
Posted by Porespellar@reddit | LocalLLaMA | View on Reddit | 52 comments
windows_error23@reddit
What?