Skip to content
RCreddit.com·
Not on the current live radar

Qwen3.8-Flash-Next in llama.cpp vs SGLang vs FreeToken: 35s vs 258s to first token at full context. My findings on new PRs coming to engines.

AI summary

A user compared the performance of Qwen3.8-Flash-Next across llama.cpp, SGLang, and FreeToken, noting a significant difference in time to first token at full context (35s vs 258s). The findings also detailed the impact of MTP (Multi-Threaded Processing) on VRAM usage and token generation rates. At 96 GiB, MTP provided 1.61x faster decode, but at 24 GiB, it made decode about 3.4x slower. The user is interested in further testing on different GPU/memory setups and MTP's effectiveness with CPU offloading.

Time & source

Ingested
09/08, 22:00 UTC+0
Source type
Dev community
Article

Full text isn't available here.

Read at source →
Source·reddit.com