ALPHA-One

From PINE64
Revision as of 00:39, 25 May 2025 by Tllim (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search
The ALPHA-One

The ALPHA-One is a Generative 7b LLM AI Agent based on StarPro64 board RISC-V based Single Board Computer with 20TOPS(int 8) NPU capability, and preinstalled 64G eMMC module contains 7b Deepseek/Owen LLM build in docker form.


Software releases

This is the early build that contains both 7b Deepseek and QWen2 LLM build, OS based on ROCKOS.


DD image for 32GB eMMC module and above

Default credentials
Super user eswin/eswin

To explore LLM, use terminal program and execute below command lines:

deepseek 7bLLM command line: sudo /opt/eswin/sample-code/npu_sample/qwen_sample/bin/es_qwen2 /opt/eswin/sample-code/npu_sample/qwen_sample/src/deepseek_r1_distill_qwen_7b/config.json

qwen2 7bLLM command line: sudo /opt/eswin/sample-code/npu_sample/qwen_sample/bin/es_qwen2 /opt/eswin/sample-code/npu_sample/qwen_sample/src/qwen2_7b/config.json

qwen2 0.5bLLM command line: sudo /opt/eswin/sample-code/npu_sample/qwen_sample/bin/es_qwen2 /opt/eswin/sample-code/npu_sample/qwen_sample/src/qwen2_0_5b/config.json


SDK releases



Main Board and SoC Specification

Eswin uk logo.jpg

EIC7700X Block Diagram.png


Features

System Memory

  • 32GB 64bits LPDDR5@6400MHz RAM Memory.
  • 64GB eMMC preintsalled with 7b Deepseek/Qwen LLM

Case

  • CNC from single Aluminum block
  • Passive cooling, built-in heat pipe that directly contact with SoC

Network

  • Dual 10/100/1000Mbps Ethernet
  • 2.4GHz/5GHz MIMO WiFi 802.11 b/g/n/ac/ax with Bluetooth 5.3

Expansion Ports

  • 2× USB3.0 Dedicated Host port
  • 2× USB2.0 Shared Host port
  • Digital Video output up to 4K@60Hz
  • 3.5mm audio Jack with mic in


Certifications

  • Not yet available


Potential Improvement

  • Current 7bLLM throughput is around 3.5 tokens/second using docker form, potential improve upto 10 tokens/second when operates LLM in native mode.
  • Potential increase LLM capability from 7b to 15b with system software optimization
  • If ALPHA-One proven its potential, consider release ALPHA-Two which can host up to 30b LLM