Alexander Hanel alexander-hanel

DeepSeek V4 Flash on DGX Spark

run_deepseek_v4_flash_dgx_spark.sh is a standalone installer and launcher for serving DeepSeek V4 Flash on an NVIDIA DGX Spark. It reproduces the model stack used by the StressingLLMs DeepSeek campaign without piping a changing remote installer directly into Bash.

The script pins and verifies:

the Entrpi/ds4 v0.6.2 source archive
the DeepSeek V4 Flash 0731 Q2 imatrix base GGUF

Anti-hype LLM reading list

Goals: Add links that are reasonable and good explanations of how stuff works. No hype and no vendor content if possible. Practical first-hand accounts of models in prod eagerly sought.

Foundational Concepts

Pre-Transformer Models

Every project needs a list of articles that I will not read until I fail enough to be motivated to read them.

A month in the laboratory can often save an hour in the library.

Ghidra-MCP

https://github.com/bethington/ghidra-mcp
// Will be using Ghidra because Binja Headless and IDA Pro are too expensive for home projects.

clearbluejar/pyghidra-mcp

https://clearbluejar.github.io/posts/supercharging-ghidra-using-local-llms/

Notable links on Intel's Shadow-Stack or CET Mitigations

Current working URL

https://hex-rays.com/blog/tag/igors-tip-of-the-week/

#01: Lesser-known keyboards shorcuts in IDA

CTRL+Enter to close a window
hold down ALT to display shortcut options
ALT-X to save database

source

Quick notes on pyproject.toml

Folder Structure

my-project
├── pyproject.toml
└── src
    └── my-project
 └── my_module.py

	def median(values):
	n = len(values)

	if n % 2 == 1:
	return values[n // 2]

	return (values[n // 2 - 1] + values[n // 2]) / 2


	def quartiles(values):


	import idautils
	ea = 0x000000140013188
	name = ida_name.get_ea_name(ea)
	print("found")

	# get xrefs to function
	xrefs = [x for x in idautils.CodeRefsTo(ea, 0)]

	for func in xrefs: