Daniel Lougen
@DJLougen
PhD @ UofT | Visual Neuroscience | Gestalt Labs @GestaltAILab | Neutrality Project | Qwen Dev Ambassador |
1.5K Following 3.4K Followers
So, my initial reaction to the benchmarks was meh. But, looking further it hadnt clicked with me just how good a model it is + the efficiency and size until realizing how much smaller it is.
So, the full model is really good, and we can compress that intelligence and fit it on more diverse sets of hardware.
Really exciting step forward and excited to see more from @reflection_ai and american AI!
Show more
My new low-bit quant of Qwen3.8-Flash-Next.
2.125-bit experts
2.39 bits/weight across the transformer
92.0 GB download
~39 GiB in memory
95.5% of BF16’s score
A 180B-parameter model that runs on one DGX Spark.
The hugest of huge thanks to @LambdaAPI and the amazing @TheZachMueller for making a project this compute-hungry possible.
🔗
Show more
Almost passing @BrandonMusicKy in downloads.... and over half the likes for just copying the model....
As a Dev ambassador I got access to a preview of @Alibaba_Qwen image 2.1
Same prompt: first image is Qwen 2.1, Second is @higgsfield
Qwen image 2.1s lighting is really impressive!
Show more
REAP when done by competent people
A lot of people need to hear this
