Ripple-65B-Preview-QVB8-1e-3 vs Qwen-27B-QVB8-1e-3_fp16
#16
by addiesam - opened
We used Ripple-65B-Preview-QVB8-1e-3 and Qwen-27B-QVB8-1e-3_fp16 to draw a same SVG.
Both models are quantized under the same QVB8-1e-3 (0.001 error bound) setting, it really highlights how distinct the two models actually are—they aren't just copies of each other, and each takes a completely different approach to building the SVG.
Let's comment.
prompt:
Generate an SVG of a crab operating an excavator to build a sandcastle
Ripple-65B-Preview-QVB8-1e-3:
https://huggingface.co/addiesam/Ripple-65B-Preview-QVB8-1e-3
Ripple-65B-Preview-QVB8-1e-3:
Qwen-27B-QVB8-1e-3_fp16: