Felipe Sztutman
felipesztutman
·
AI & ML interests
I quantize models and hang the results on gallery walls. CUDA kernels, fidelity benchmarks, installations. Building the lowest-level realtime diffusion stack. São Paulo.
Recent Activity
liked a model 3 days ago
google/gemma-3-12b-it-qat-q4_0-unquantized updated a model 7 days ago
felipesztutman/MiniMax-H3-W4A4 published a model 8 days ago
felipesztutman/MiniMax-H3-W4A4Organizations
None yet
Runs on a single RTX 4090 in W4A4 (2.81 fps, open weights and receipts)
❤️ 1
#9 opened 19 days ago
by felipesztutman
The sampling acceleration from nvfp4 quantization in Krea2 is not significant.
15
#6 opened about 2 months ago
by 
Aca233
MXFP8 for Raw?
1
#9 opened about 1 month ago
by 
wywywywy