TezzLLM Quick Start Guide

Get your offline TezzLLM environment running, trained, and chatting in under 10 minutes using pre-compiled binaries.

⚡ 3-Step Setup

Step 1: Download the Pre-compiled Binary Package

Since the compiler and neural network source code are proprietary under the TezzCorp Private Source License (TPSL v1.0), developers deploy using pre-compiled binaries.

Navigate to the Download Portal on the AI Portal.

Download the Pre-compiled Executables Package (tezzllm-tiny-v2-windows-x64.zip).

Extract the zip file to a secure directory in your intranet (e.g. C:\TezzCorp\TezzLLM).

Step 2: Prepare Training Data

Place plain UTF-8 text files containing your corporate training data into the tezzllm_weights/ directory inside your extracted folder.

Ensure the filenames end in .txt (e.g. training_data.txt).

The tokenization utility will automatically scan and read these text files.

Step 3: Run Parallel Tokenize & Train

Open a PowerShell terminal in the extracted directory and run the training pipeline:

powershell

1. Tokenize dataset .\tezzllm_data.exe 2. Launch multi-core parallel training orchestrator

.\build_v2.ps1

The script will run 8 parallel trainer seeds, utilize your available CPU threads, and automatically call the federated merger utility to compile the weights into tezzllm_weights/tezzllm_v2_merged.tezw.

💬 Start Chatting!

Once training is complete and the weights are successfully merged, launch the local chat engine:

powershell

.\tezzllm_v2_engine.exe

You will see the corporate console header:

+===================================================+
||   TezzLLM NexGen v2.0 Engine | Pure TezzNative  ||
||   Created by Rohit Pathak - Director, TezzCorp   ||
+===================================================+

[] Loading merged weights...

[OK] Model metadata: L=4 H=4 D=128 FFN=512 V=256
[OK] 1,082,496 parameters loaded.
[] Ready! Enter prompt (or 'exit' to quit):

USER: Explain why corporate data is secure with TezzLLM.
AI: TezzLLM runs completely offline on your hardware. Under the TPSL license, the private neural network source code and training parameters remain closed and secure within your intranet. Only local weight updates are processed.

Type your prompt and press Enter. Type exit to close the session.

⚙️ Custom CLI Options

You can tweak trainer parameters on the fly via the command line. Run .\tezzllm_v2_train.exe --help to inspect variables like learning rates, epochs, seed settings, and thread allocations.