Guys,
I was so sad when I checked out your AI Hub - the models are all super outdated. I don't understand. Why do you offer models that are a year old, or even worse?
And then your request list has models that are more than a year old too and they're still in the request list - why is Gemma 2 in the list when Gemma 4 is out? Why aren't you converting the models when they come out?
I don't understand, Snapdragon is the only Windows hardware with SoC ("unified") RAM and therefore suitable for running proper models on a laptop. I would have loved to buy the Asuus A16 with Snapdragon and 48 GB RAM and finally rid my life of MacOS, but you're obviously completely neglecting the software side, so why would I switch?
You need to have all latest models (within 1 week of release) of:
- Qwen (e.g. 3.5, soon 3.7 is coming out. I particularly want 3.5 27B)
- Gemma (e.g. 4, I particularly want 4 26B)
- Arcee (current OpenClaw champion for open models)
- probably 2-3 other model families
You could probably create an AI agent based pipeline for model conversion that could automate 70-90% of your work, but looks like primarily you need to upscale the team working on this.
This wouldn't be so bad if you had proper support for GGUF models. But I obviously don't want to run low-accuracy Q4_0 models, and better quality quantizations you don't offer Adreno optimization for. But that would be a solid alternative - if you offered proper optimization for Q4_K, Q5 and Q6 quantized models.
Guys,
I was so sad when I checked out your AI Hub - the models are all super outdated. I don't understand. Why do you offer models that are a year old, or even worse?
And then your request list has models that are more than a year old too and they're still in the request list - why is Gemma 2 in the list when Gemma 4 is out? Why aren't you converting the models when they come out?
I don't understand, Snapdragon is the only Windows hardware with SoC ("unified") RAM and therefore suitable for running proper models on a laptop. I would have loved to buy the Asuus A16 with Snapdragon and 48 GB RAM and finally rid my life of MacOS, but you're obviously completely neglecting the software side, so why would I switch?
You need to have all latest models (within 1 week of release) of:
You could probably create an AI agent based pipeline for model conversion that could automate 70-90% of your work, but looks like primarily you need to upscale the team working on this.
This wouldn't be so bad if you had proper support for GGUF models. But I obviously don't want to run low-accuracy Q4_0 models, and better quality quantizations you don't offer Adreno optimization for. But that would be a solid alternative - if you offered proper optimization for Q4_K, Q5 and Q6 quantized models.