Problem
Does MTPLX support context window scaling using Yarn (https://openreview.net/pdf?id=wHBfxhZu1u)?
I am running Qwen 3.8 27B locally, which has a context window of 262K tokens, and I want to scale it up to 1M.
Proposal
https://saraswatmks.github.io/2025/12/rope-scaling-llms.html
Problem
Does MTPLX support context window scaling using Yarn (https://openreview.net/pdf?id=wHBfxhZu1u)?
I am running Qwen 3.8 27B locally, which has a context window of 262K tokens, and I want to scale it up to 1M.
Proposal
https://saraswatmks.github.io/2025/12/rope-scaling-llms.html