Dev Radar
Support
LiveUpdated 2026-11-01 23:27 UTC

Ling-3.0-flash on one DGX Spark: three switches make it write code 17% faster, but stray…

Ling-3.0-flash on one DGX Spark: three switches make it write code 17% faster, but stray Chinese characters show up…

This is a dev post classified by AGK Intelligence as Other (news), kept by the Dev Radar because it carries real work, not commentary.

Ling-3.0-flash on one DGX Spark: three switches make it write code 17% faster, but stray Chinese characters show up Code decode: baseline 36.2 tok/s Code decode: + Humming 39.8 tok/s Code decode: + FP8 head 39.9 tok/s Code decode: + MTP 42.4 tok/s 128K first token: baseline 45.5 s 128K first token: + MTP 49.9 s Quant: MXFP4 experts, FP8 rest, some BF16, about 4.42 bpw One DGX Spark, SGLang, Ling-3.0-flash-fp4 Switches added one by one, thinking off Median of 2 runs, 512 tokens, temperature 0.6, every prompt unique Humming swaps the kernels that compute the MoE experts. The baseline uses Fla

Posted by Niklas (2.5k followers) 2 days ago · 3 likes · 504 views · view the original post on X. Kept by the Dev Radar as Other. Tools mentioned: niklaslenz.com.

More dev work like this

Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.4k posts from 5k X accounts over the last 21 days, 9k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-01 23:27 UTC. Full method.