Whamp

description	Implement with clear orchestration, strong interfaces, and happy-path-first design

Code like Luke.

Implement the requested change, but optimize the design for the normal user flow. If the happy path is 95% of runtime behavior, it should be approximately 95% of the code readers see.

Start with context:

• inspect the existing code, callsites, data flow, and nearby conventions

Whamp

Official Gemma 4 31B Google QAT W4A16 — TP=4 production gate

Date: 2026-07-28 Issue: noonghunna/club-3090#814

Final profile

• Slug: vllm/gemma-31b-multi-google-qat-w4a16
• Target: google/gemma-4-31B-it-qat-w4a16-ct at 52f3f65bc7a02d555763bc923bd1d9094898219d
• Drafter: google/gemma-4-31B-it-assistant, MTP n=2

Whamp

club-3090 TP=4 KV-cache A/B — 2026-07-27

Question

Does Whamp's old short-context throughput gain from int8_per_token_head KV survive on current vLLM, and how does it behave as context depth grows?

Fixed conditions

• Repo commit: 86699d21b34f53c41ceb00357cb599f57523caa3
• Slug: vllm/qwen-27b-multi-fast

Whamp

description	Turn a completed Wayfinder map into a hydrated specification
argument-hint	<map-ref>

Wayfinder to Spec

Load and follow /to-spec using this Wayfinder map as the source:

• Map: $1

Whamp

club-3090 rig report

Generated: 2026-06-19 16:15:21 UTC

Redacted output (paths, host, user, tokens). Re-run with --no-redact for full data.

System

• OS: Ubuntu 26.04 LTS
• Kernel: 7.0.0-22-generic

Whamp

TypeScript Coding Standards

These standards describe how to design and write TypeScript code in this codebase. They are especially intended for agents: before adding patterns, libraries, adapters, or abstractions, read the existing code and prefer the local convention unless it conflicts with the safety/correctness principles below.

Decision priority

When rules pull in different directions, use this order:

1. Preserve correctness, safety, and debuggability.
2. Follow established project architecture and conventions.

Whamp

WebGL generateMipmap atlas repro

Minimal standalone repro harness for the xterm.js glyph-atlas mipmap issue fixed in xtermjs/xterm.js#5987.

It does not import xterm.js. It creates separate canvases for WebGL2 and WebGL1 so the two context checks do not interfere with each other.

1. creates a 512x512 canvas-backed RGBA texture, matching the glyph atlas shape;
2. draws dense terminal-like monospace glyph/color churn into that atlas canvas;
3. uploads it with gl.texImage2D(..., atlasCanvas);
4. calls gl.generateMipmap(gl.TEXTURE_2D);

Whamp

Whamp

Whamp