rasa/command-generator-llama-3.1-8b-instruct

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Jun 10, 2025License:apache-2.0Architecture:Transformer Open Weights Cold

The rasa/command-generator-llama-3.1-8b-instruct is an 8 billion parameter text generation model developed by Rasa Technologies, fine-tuned from meta-llama/Llama-3.1-8B-Instruct. This model is specifically designed for Dialogue Understanding (DU) within Rasa's Conversational AI with Language Models (CALM) approach. It translates conversational transcripts and business logic into a short sequence of predefined commands, making it highly specialized for controlling AI assistant dialogue flow rather than generating free-form text.

Loading preview...

Overview

This model, developed by Rasa Technologies, is a specialized Dialogue Understanding (DU) model designed for Rasa's Conversational AI with Language Models (CALM) approach. Fine-tuned from meta-llama/Llama-3.1-8B-Instruct, it processes conversational transcripts and structured business logic to output a sequence of specific commands.

Key Capabilities

  • Command Generation: Translates user messages into a predefined grammar of commands (e.g., start flow, set slot, cancel flow, disambiguate flows, provide info, offtopic reply, hand over).
  • Dialogue Understanding: Interprets ongoing conversations to drive the logic of AI assistants.
  • Rasa CALM Integration: Essential component for assistants built using the CALM paradigm.
  • Fine-tuning Base: Can serve as a base model for further fine-tuning on custom assistant data using Rasa Pro's fine-tuning recipe feature.

Good For

  • Powering customer-facing chatbots, voice assistants, IVR systems, and internal organizational chatbots built with Rasa CALM.
  • Direct use in command generator components where CALM assistant flows are similar to the rasa-calm-demo assistant.
  • Developers looking to fine-tune a model for highly specific command generation within a Rasa environment.

Limitations

  • No Free-form Text Generation: Explicitly fine-tuned for command output; cannot generate arbitrary text.
  • Language Specificity: Primarily tested and evaluated for English; performance in other languages is not guaranteed.
  • Bias Susceptibility: Like all pre-trained models, predictions may be susceptible to biases present in training data.

Finetunes
2 models