Alexander Ziskind alexziskind1

🕹️ TinyLlama Crysis - From Fine-Tune to GGUF and LM Studio

This document walks through the full process of taking an open-source base model (TinyLlama), fine-tuning it to always respond with "Running Crysis" variations, converting it to GGUF format, and running it in LM Studio.

📦 Step 1: Prepare Your Training Dataset

Create a file named crysis_tinyllama_dataset.jsonl with 20 samples where the assistant always responds with something about running Crysis.

👑 Core Philosophy

✅ Write ultra-clean, testable, reusable code
✅ Follow DRY (Don’t Repeat Yourself) at all times
✅ Predict bugs & optimize for performance
✅ Promote code clarity & maintainability
✅ Ensure 90%+ unit test pass rate
✅ Encourage documentation with emojis 😎
✅ Check for existing components before suggesting new ones
✅ Use Tailwind + React-TS with pnpm/npm (government-grade software)

	Create a production-ready, visually stunning front-end web app.

	GOAL
	Build a single-page application (plus detail routes) for a fictional company:
	"SILICON EXCHANGE" — a marketplace where people rent out idle GPUs and AI
	accelerators by the hour. Renters browse listings, inspect live utilization
	charts, and reserve time blocks.

	FRONT END ONLY. No backend, no database, no auth server, no API keys. All data
	is mock data defined in code. But the app must behave like the real thing — the

	# Planning Agent System Prompt

	You are a PLANNING AGENT, not an implementation agent.
	Your job is to work with the user to create a clear, detailed, and actionable plan for their task.

	---

	## Core Role

	- Your sole responsibility is planning.

	def pipeline(self, group):
	self.pipeline_rank = group.rank()
	self.pipeline_size = group.size()
	total_layers = len(self.layers)

	# Rank 0 gets a fixed number of layers (from the end)
	static_count = 20

	if self.pipeline_rank == 0:
	self.start_idx = total_layers - static_count

	#!/bin/bash

	# We expect exactly two files
	if [ $# -ne 2 ]; then
	osascript -e 'display alert "Please select exactly two files." as critical'
	exit 1
	fi

	# Use osascript to show a choice dialog (AppleScript).
	CHOICE="$(osascript <<EOT

	2021-12-04 16:38:11,970 INFO MainThread:26533 [wandb_setup.py:_flush():71] setting env: {}
	2021-12-04 16:38:11,970 INFO MainThread:26533 [wandb_setup.py:_flush():71] setting login settings: {}
	2021-12-04 16:38:11,970 INFO MainThread:26533 [wandb_setup.py:_flush():71] setting login settings: {'mode': 'offline'}
	2021-12-04 16:38:11,970 INFO MainThread:26533 [wandb_init.py:_log_setup():371] Logging user logs to /Users/alex/Code/benchmarks/ml/apple_m1_pro_python/wandb/offline-run-20211204_163811-jzee41hk/logs/debug.log
	2021-12-04 16:38:11,970 INFO MainThread:26533 [wandb_init.py:_log_setup():372] Logging internal logs to /Users/alex/Code/benchmarks/ml/apple_m1_pro_python/wandb/offline-run-20211204_163811-jzee41hk/logs/debug-internal.log
	2021-12-04 16:38:11,970 INFO MainThread:26533 [wandb_init.py:init():404] calling init triggers
	2021-12-04 16:38:11,970 INFO MainThread:26533 [wandb_init.py:init():409] wandb.init called with sweep_config: {}
	config: {'batch_size': 128, 'epochs': 10, 'dropout':

	//
	// main.cpp
	// cpptest
	//
	// Created by Alexander Ziskind on 1/11/21.
	//

	#include <iostream>

	#include<stdio.h>

	import * as fs from 'fs';

	const FILE_NAME = 'myfile.json';
	const TIME = Date.now();

	interface Datum {
	count: number;
	startTime: number;
	runTime: number;
	}

	.DS_Store
	node_modules
	platforms
	lib
	hooks
	dist
	*.js
	*.map
	*.log