dotnet

picoLLM Inference Engine .NET Demos

Made in Vancouver, Canada by Picovoice

picoLLM Inference Engine

picoLLM Inference Engine is a highly accurate and cross-platform SDK optimized for running compressed large language models. picoLLM Inference Engine is:

• Accurate; picoLLM Compression improves GPTQ by significant margins
• Private; LLM inference runs 100% locally.
• Cross-Platform
• Runs on CPU and GPU
• Free for open-weight models

Requirements

• .NET 8.0

Compatibility

• Linux (x86_64)
• macOS (x86_64, arm64)
• Windows (x86_64, arm64)
• Raspberry Pi (4, 5)

Models

picoLLM Inference Engine supports the following open-weight models. The models are on Picovoice Console.

• Gemma
  • gemma-2b
  • gemma-2b-it
  • gemma-7b
  • gemma-7b-it
• Llama-2
  • llama-2-7b
  • llama-2-7b-chat
  • llama-2-13b
  • llama-2-13b-chat
  • llama-2-70b
  • llama-2-70b-chat
• Llama-3
  • llama-3-8b
  • llama-3-8b-instruct
  • llama-3-70b
  • llama-3-70b-instruct
• Llama-3.2
  • llama3.2-1b-instruct
  • llama3.2-3b-instruct
• Mistral
  • mistral-7b-v0.1
  • mistral-7b-instruct-v0.1
  • mistral-7b-instruct-v0.2
• Mixtral
  • mixtral-8x7b-v0.1
  • mixtral-8x7b-instruct-v0.1
• Phi-2
  • phi2
• Phi-3
  • phi3
• Phi-3.5
  • phi3.5

AccessKey

AccessKey is your authentication and authorization token for deploying Picovoice SDKs, including picoLLM. Anyone who is using Picovoice needs to have a valid AccessKey. You must keep your AccessKey secret. You would need internet connectivity to validate your AccessKey with Picovoice license servers even though the LLM inference is running 100% offline and completely free for open-weight models. Everyone who signs up for Picovoice Console receives a unique AccessKey.

Usage

There are two demos available: completion and chat. The completion demo accepts a prompt and a set of optional parameters and generates a single completion. It can run all models, whether instruction-tuned or not. The chat demo can run instruction-tuned (chat) models such as llama-3-8b-instruct, phi2, etc. The chat demo enables a back-and-forth conversation with the LLM, similar to ChatGPT.

NOTE: File path arguments must be absolute paths. The working directory for the following dotnet commands is:

picollm/demo/dotnet/PicoLLMDemo

Build with the dotnet CLI:

dotnet build -c ChatDemo.Release
dotnet build -c CompletionDemo.Release

For both demos, you can use --help/-h to see the list of input arguments.

Chat Demo

To run an instruction-tuned model for chat, run the following in the terminal:

dotnet run -c ChatDemo.Release -- --access_key ${ACCESS_KEY} --model_path ${MODEL_PATH}

Replace ${ACCESS_KEY} with yours obtained from Picovoice Console and ${MODEL_PATH} with the path to a model file downloaded from Picovoice Console.

To get information about all the available options in the demo, run the following:

dotnet run -c ChatDemo.Release -- --help

Completion Demo

Run the demo by entering the following in the terminal:

dotnet run -c CompletionDemo.Release -- --access_key ${ACCESS_KEY} --model_path ${MODEL_PATH} --prompt ${PROMPT}

Replace ${ACCESS_KEY} with yours obtained from Picovoice Console, ${MODEL_PATH} with the path to a model file downloaded from Picovoice Console, and ${PROMPT} with a prompt string.

To get information about all the available options in the demo, run the following:

dotnet run -c CompletionDemo.Release -- --help