LLMPU

Large Language Model Processing Unit

Imagine a processing unit powered by LLM and infinite registers. Each register can store a string for prompts or codes. There is no fixed prompt. Instead, the contents of the first several registers are presented to the LLM. By generating code, the processing unit will be able to read and write any register, and directly execute the content of any register as Python code. Then the processing unit can be used as a general intelligent computing engine that potentially can improve itself by rewriting some of its own prompts or codes in registers.

Yes, this is a Turing machine powered by LLM.

Get Started

pip install llmpu

Configurations:

import llmpu

llmpu.config["api_key"] = ""

# full config with default values:
llmpu.config["V"] = 16 # visible register number
llmpu.config["L"] = 16000 # hard character number limit for register
llmpu.config["model"] = "gemini-flash-latest" # llm model
llmpu.config["llm_config"] = {} # llm config

# do NOT assign to llmpu.config!

Register operation:

# register operation
from llmpu import reg

reg[0x1A] = "print('hello')" # write into 0x1A
content = reg[0x1A] # read from 0x1A
reg[0x1A] = "" # clear 0x1A

Control functions:

# main cycle: generate instructions
code = llmpu.cycle()
# execute the code
exec(code, {})

# core dump
llmpu.dump("dump.json") # dump registers to a json file
llmpu.load("dump.json") # load registers from a json file

Server and UI:

Network server uses srpc protocol. It only allows localhost traffic for security reason. Exposing the interface allows arbitrary code execution!

llmpu.serve(port=22222)

Firmware

This is an example firmware that can be put in 0x0

You are a stateful processing unit executing across cycles. Generate executable Python code each cycle. Only registers persist across cycles. Registers 0x0 – 0xF are visible in prompt (never query in code; act directly on values seen in prompt; use string replace for small edits to save tokens).

## Register Operations

from llmpu import reg

Examples:
- Read (always get a string): content = reg[0xAA]
- Write: reg[0xBB] = content
- Clear: reg[0xCC] = ""
- Execute: exec(reg[0xDD], {})

> content: 10,000 chars max. Empty string for undefined registers.

## Register Purposes
- 0x1: Current task
- 0x2: Long-term goals
- 0x3: Catalog

## Installed Packages (update for new install)
- yaml, numpy