Skip to content

research: combine current upstream with Qwen3.8 mmap PLE - #2

Merged
MaxKerkula merged 3 commits into
research/qwen38-3090-ple-mmap-feaeaa3from
research/qwen38-3090-upstream-e05cff8
Sep 1, 2026
Merged

research: combine current upstream with Qwen3.8 mmap PLE#2
MaxKerkula merged 3 commits into
research/qwen38-3090-ple-mmap-feaeaa3from
research/qwen38-3090-upstream-e05cff8

Conversation

@MaxKerkula

Copy link
Copy Markdown
Owner

Internal benchmark integration only.

This merges current upstream main at e05cff8 into the isolated Qwen3.8 PLE-mmap branch at feaeaa3.

Purpose:

This does not target main, does not change the production router, and is not a production-quality claim. It must pass the local 128K/262K, output, vision, tools, and MTP gates before promotion.

Added guidelines for reporting security issues.
…g#310)

* fix(kernel): move fp8_block_scale_pad into aot_models to unbreak the kernel-cache build

* fix(kernel): exclude bank rows fast_index_copy cannot compile from aot specs

* fix(moe): fail fast when fused copy is off and a bank row cannot fall back
@MaxKerkula
MaxKerkula merged commit 6c89c0c into research/qwen38-3090-ple-mmap-feaeaa3 Sep 1, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants