Skip to content

Instantly share code, notes, and snippets.

View 0x524c's full-sized avatar

0x524c 0x524c

  • Brazil
  • 23:16 (UTC -03:00)
  • X @0x524c
View GitHub Profile
@0x524c
0x524c / custom_pub_chat_template_qwen36.jinja
Created June 11, 2026 02:02 — forked from jscott3201/custom_pub_chat_template_qwen36.jinja
A drop-in replacement chat template for Qwen/Qwen3.6-27B tuned for open-source agentic coding harnesses.
{#---------------------------------------------------------------------
custom_pub_chat_template_qwen36.jinja
=====================================
A public, harness-friendly fork of Qwen's Qwen3.6-27B chat template,
tuned for open-source agentic coding harnesses like:
- anomalyco/opencode (https://github.com/anomalyco/opencode)
- earendil-works/pi (https://github.com/earendil-works/pi)
- openclaw, OpenHarness, similar Claude-Code-style harnesses
WHY THIS FORK EXISTS
... answer said require line and column. So implement that. Need maybe not touch LSP output normalization in lspPositionToExternal/diagnostic snapshots.
Implementation: modify positions.ts externalPositionToLsp to strict. Maybe rename? The function currently from service. It can return undefined if invalid. But error messages say "requires 1-based line and character". It should not floor floats. It should require Number.isInteger and >=1. For column, if not number returns undefined. Should we default column=1? The old allowed missing char. But "Require column >=1". So no default. This affects codeActions with missing char; okay. We'll update error messages maybe "line and column" not character. Since service methods named character param. But tool uses column alias. In hintForError accepts both. Could leave "character" for legacy? Could make "line and column". But service is generic LSP method; parameter called character. However user-facing contract says column. We could use "line and column" in errors for
@0x524c
0x524c / microgpt.py
Created February 20, 2026 18:11 — forked from karpathy/microgpt.py
microgpt
"""
The most atomic way to train and run inference for a GPT in pure, dependency-free Python.
This file is the complete algorithm.
Everything else is just efficiency.
@karpathy
"""
import os # os.path.exists
import math # math.log, math.exp
@0x524c
0x524c / prompt.py
Created October 7, 2025 03:42 — forked from do-me/prompt.py
A single line to try out mlx-community/Qwen3-Next-80B-A3B-Instruct-8bit on MacOS with mlx
import argparse
from mlx_lm import load, generate
# Parse CLI arguments
parser = argparse.ArgumentParser()
parser.add_argument("--prompt", type=str, default="hello", help="Custom prompt text")
parser.add_argument("--max-tokens", type=int, default=1024, help="Maximum number of tokens to generate")
args = parser.parse_args()
# Load model
@0x524c
0x524c / tmux-cheat-sheet.md
Created February 18, 2025 03:31 — forked from michaellihs/tmux-cheat-sheet.md
tmux Cheat Sheet
@0x524c
0x524c / 8-concurrent-generations.md Aggregate throughput just over 2 tok/sec on R1 671B with 8 concurrent generations.

tl;dr;

You can run the real deal big boi R1 671B locally off a fast NVMe SSD even without enough RAM+VRAM to hold the 200+GB weights. No it is not swap and won't kill your SSD's read/write cycle lifetime.

  • 8k context @ ~1.3 tok/sec single generation
  • 16k context @ ~0.93 tok/sec single generation
  • 2k context @ ~2.08 tok/sec with 8 parallel slots @ ~0.26 tok/sec each concurrently
  • 2k context @ ~2.13 tok/sec single generation after disabling GPU!

Notes and example generations below.

@0x524c
0x524c / nginx_deployment.yaml
Created June 21, 2024 13:46 — forked from petitviolet/nginx_deployment.yaml
sample Nginx configuration on Kubernetes using ConfigMap to configure nginx.
apiVersion: v1
kind: ConfigMap
metadata:
name: nginx-conf
data:
nginx.conf: |
user nginx;
worker_processes 3;
error_log /var/log/nginx/error.log;
events {

Upgrading Kubernetes Cluster with Kops, and Things to Watch Out For

Alright! I'd like to apologize for the inactivity for over a year. Very embarrassingly, I totally dropped the good habit. Anyways, today I'd like to share a not so advanced and much shorter walkthrough on how to upgrade Kubernetes with kops.

At Buffer, we host our own k8s (Kubernetes for short) cluster on AWS EC2 instances since we started our journey before AWS EKS. To do this effectively, we use kops. It's an amazing tool that manages pretty much all aspects of cluster management from creation, upgrade, updates and deletions. It never failed us.

How to start?

Okay, upgrading a cluster always makes people nervous, especially a production cluster. Trust me, I've been there! There is a saying, hope is not a strategy. So instead of hoping things will go smoothly, I always have bias that shit will hit the fan if you skip testing. Plus, good luck explaining to people

Install ffmpeg

brew install ffmpeg

Download file through url, like this:

ffmpeg -protocol_whitelist file,http,https,tcp,tls,crypto -i "http://url-file.domain.m3u8" -c copy video.mp4