Skip to content

Instantly share code, notes, and snippets.

View Rahul-fix's full-sized avatar
🎯
Focusing

s7rakuma Rahul-fix

🎯
Focusing
View GitHub Profile
@co-l
co-l / QWEN38-27B-RTX5090.md
Created August 23, 2026 12:04
Qwen3.8-27B NVFP4 with vision + 451K token KV-cache on one RTX 5090 (400W) — standalone reproduction

Qwen 3.8 27B NVFP4 with vision + 451K tokens KV-cache on one RTX 5090@400W: 120 tokens/s

What this is

Qwen3.8-27B, with vision, holding 451K tokens of KV on a single RTX 5090, under a 400 W cap, in a container. Stock fp8 KV fits ~124K total on this card; this stack does 3.6x that, so three parallel sessions' context (3×75K) stays resident in the pool: sessions keep their KV and continue without re-prefill cost.

The stack

@sankalp-khare
sankalp-khare / switch-gnupg.sh
Created July 3, 2023 11:00
Downgrade / switch gnupg version in homebrew and pin it
#!/bin/bash -e
#-------------------------------------------------------------------------------------
# Summary: Executes commands to...
# * Check/output currently installed version of gnupg
# * Change into the brew repo/install directory
# * Find the git commit-id of the specified version of gnupg
# * git branch/switch to the commit-id
# * replace the brew installed version with the specified version (and pin it)
# * clean up the branch, switch back to master, and remove cached files
# * Check/output the newly installed version of gnupg