3 ms·
I ran some pelicans at the four different reasoning levels (none, low, medium, xhigh - apparently high and xhigh are aliases of each other) on a DGX Spark using
by simonw 2mo ago
I ran some pelicans at the four different reasoning levels (none, low, medium, xhigh - apparently high and xhigh are aliases of each other) on a DGX Spark using Unsloth's unsloth/Qwen3.8-Flash-Next-GGUF (UD-IQ1_S):
https://tools.simonwillison.net/markdown-svg-renderer#url=https%3A%2F%2Fgist.github.com%2Fsimonw%2Ff9c69ebdab90d8a45b8de4742cc7b840 https://tools.simonwillison.net/markdown-svg-renderer#url=ht...
Surprised I didn't get one I liked as much as the Qwen 3.8 27B one https://simonwillison.net/2026/Aug/16/qwen-38-27b/#the-default-of-extra-high-results-in-spectacular-over-thinking https://simonwillison.net/2026/Aug/16/qwen-38-27b/#the-defau... , maybe because of quantization.
- kzrdude 2mo agoIf I read correctly, that's based on a 1-bit quantization, and can we really expect that to produce any useful output at all?
- walrus01 2mo agoif the goal is vector art salvador dali, like melting clocks and stuff, sure
- coder543 2mo agoThe spark can easily run UD-Q4_K_XL on this model... using IQ1_S doesn't make much sense.
- badatnames 2mo agoYou should find an excuse to offer 3D printed extruded pelicans from various models as awards for something. I have no idea for what, but the idea captivates and I'd love to win one somehow. They'd be collector's items in a few decades
- jononor 2mo agoIf Simon would pitch for example PCBWay that and I am pretty sure they will sponsor it (assuming their logo stays). They can do laser engraved versions also ;)
- water-drummer 2mo agoDoing this on a 1-bit quant is unfair
- simonw 2mo agoTried again with a different quant, UD-Q2_K_XL: https://tools.simonwillison.net/markdown-svg-renderer#url=https%3A%2F%2Fgist.github.com%2Fsimonw%2F6ba7cbfc1a9336986703b41f7fccd73a https://tools.simonwillison.net/markdown-svg-renderer#url=ht...
- potus_kushner 2mo agothe xhigh version looks amazingly good for a 2 bit quant.
- simonw 2mo ago... and once more with UD-IQ4_XS https://tools.simonwillison.net/markdown-svg-renderer#url=https%3A%2F%2Fgist.github.com%2Fsimonw%2F9bbda2a4acbdbe67bad98289910f8eff https://tools.simonwillison.net/markdown-svg-renderer#url=ht...
- walrus01 2mo agohow does it do with a pelican equipment case?
- qingcharles 2mo agoI guess they haven't benchmaxxed it yet? https://imgur.com/a/vT636BS https://imgur.com/a/vT636BS
- bhelkey 2mo agoWhy did you use 1-bit quantization vs 3-bit quantization? It looks like the 3-bit requires 90 GB[1] which, I imagine, would fit within the DGX Spark's 128GB of unified memory. [1] https://unsloth.ai/docs/models/qwen3.8-next https://unsloth.ai/docs/models/qwen3.8-next
- pyaamb 2mo agoThe Pelican Brief
- richthes 1mo ago[dead]