Replying to @wildmindai
Since when safetensors can be flagged as unsafe ^^
If you think this is some heroic move “for gamers’ rights” and not a pump-and-dump, you’re really naive
6
Replying to @NikTek
The reason behind those leaks is just pump and dump, everything else is a pretext, how can't you see that
12
Replying to @Dennyt_Digital @Tech2Wild
None, this is false, you can maybe get a second hand B70 (for an usable quant and reasonable context length) , 100-120t/s is 5090 perfs, you won't get that for 1k
1
1
111
Replying to @0x0SojalSec
Proceed to find a junk folder full of old generated videos from outdated models, call it "synthetic data" and train this shit over a new model, what a waste of compute
10
3,680
I'm on a 4090, for dense 27B he's right unfortunately, can push 100k context with Q4, Q8 KV cache, with mmproj vision and Q8 MTP drafter that offers speed tradeoff for lower context, 24gb is limited for agentic usage, constantly triggering compactions
2
35
Replying to @sandhivigrahika @markgadala
It's just a twitter account you think it's an actual news media, tells about how you trust anything you read from randoms online
13
Replying to @Tono_Ken3
Snowball effect on downloads, doesn't mean the model performance isn't reduced
1
43
Replying to @jauneclaude
QWEN 3.6 35b vs GEMMA4 27B , same results across multiple runs on similar tasks, gemma4 have better vision
1
2
254



