> If you want a good dense model, use qwen3.6 27B instead, speed will be up, and if you don't take my word for it being smarter, take openrouter's prices of it against the bigger, slower and less memory-efficient gemma do the talking.
Don't know if this is the correct read. I think those providers are simply taking cue from Alibaba's first-party pricing for the 27B Dense. It's kinda overpriced imo. Perhaps it can be explained by how 'reasoning-inefficient' (relative to frontier models or even Gemma) the Qwen models are and longer sequence lengths are expensive to serve.
> HTTP is just not a good transport for streaming LLM tokens and for building async agentic applications
I don't know if I agree if this is a problem with SSE or HTTP. Something like a Redis Streams-backed SSE would solve most of the 'challenges' presented in the post.
> You cannot run these models at 8-bit on a 32GB card because you need space for context
You probably can actually. Not saying that it would be ideal but it can fit entirely in VRAM (if you make sure to quantize the attention layers). KV cache quantization and not loading the vision tower would help quite a bit. Not ideal for long context, but it should be very much possible.
I addressed the lossless claim in another reply but I guess it really depends on what the model is used for. For my usecases, it's nearly lossless I'd say.
Yeah, figure the 'nearly lossless' claim is the most controversial thing. But in my defense, ~97% recovery in benchmarks is what I consider 'nearly lossless'. When quantized with calibration data for a specialized domain, the difference in my internal benchmark is pretty much indistinguishable. But for agentic work, 4-bit quants can indeed fall a bit short in long-context usecase, especially if you quantize the attention layers.
Not at all, I actually run ~30B dense models for production and have tested out 5090/3090 for that. There are gotchas of course, but the speed/quality claims should be roughly there.
As this is a dense model and it's pretty sizable, 4-bit quantization can be nearly lossless. With that, you can run this on a 3090/4090/5090. You can probably even go FP8 with 5090 (though there will be tradeoffs). Probably ~70 tok/s on a 5090 and roughly half that on a 4090/3090. With speculative decoding, you can get even faster (2-3x I'd say). Pretty amazing what you can get locally.