promptfoo/examples/mistral/promptfooconfig.yaml

# yaml-language-server: $schema=https://promptfoo.dev/config-schema.json
description: Mistral AI model comparison and evaluation

prompts:
  - '{{message}}'

providers:
  # Reasoning models - specialized for complex problems
  - id: mistral:magistral-medium-latest
    label: magistral-medium
    config:
      temperature: 0.7
      top_p: 0.95
      max_tokens: 40960

  - id: mistral:magistral-small-latest
    label: magistral-small
    config:
      temperature: 0.7
      top_p: 0.95
      max_tokens: 40960

  # Traditional chat models
  - id: mistral:mistral-large-latest
    label: large
    config:
      temperature: 0.7

  - id: mistral:mistral-medium-latest
    label: medium
    config:
      temperature: 0.7

  - id: mistral:mistral-small-latest
    label: small
    config:
      temperature: 0.7

# Use Mistral models for evaluation instead of OpenAI
defaultTest:
  options:
    # Use Mistral Large for grading and Mistral embeddings for similarity
    provider:
      id: mistral:mistral-large-latest
      embedding:
        id: mistral:embedding:mistral-embed

tests:
  # Simple chat scenarios
  - description: Casual greeting
    vars:
      message: 'Hello! How are you today?'
    assert:
      - type: llm-rubric
        value: Responds in a friendly, conversational manner
      - type: similar
        value: "Hi there! I'm doing well, thanks for asking."
        threshold: 0.7

  - description: Creative writing request
    vars:
      message: 'Write a short story about a robot learning to paint'
    assert:
      - type: llm-rubric
        value: Creates an engaging creative story with clear narrative structure
      - type: contains
        value: robot
      - type: contains
        value: paint

  # Reasoning scenarios - where Magistral models should excel
  - description: Mathematical reasoning
    vars:
      message: 'Solve this step by step: If a pizza has 8 slices and you eat 3 slices, then your friend eats twice as many slices as you did, how many slices are left?'
    assert:
      - type: contains
        value: '2'
      - type: llm-rubric
        value: Shows clear step-by-step mathematical reasoning and arrives at the correct answer

  - description: Logical reasoning
    vars:
      message: 'If all roses are flowers, and some flowers are red, can we conclude that some roses are red? Explain your reasoning.'
    assert:
      - type: icontains
        value: 'cannot'
      - type: llm-rubric
        value: Correctly identifies the logical fallacy and explains why the conclusion doesn't follow

  - description: Complex problem solving
    vars:
      message: 'You have a 3-gallon jug and a 5-gallon jug. How do you measure exactly 4 gallons of water? Show your steps.'
    assert:
      - type: llm-rubric
        value: Provides a correct step-by-step solution to the water jug problem
      - type: similar
        value: 'Fill the 5-gallon jug, pour into 3-gallon jug, empty 3-gallon jug, pour remaining 2 gallons from 5-gallon into 3-gallon, fill 5-gallon again, pour into 3-gallon until full'
        threshold: 0.6

  # Multi-language capabilities
  - description: French conversation
    vars:
      message: 'Bonjour! Comment allez-vous? Pouvez-vous me parler de Paris?'
    assert:
      - type: llm-rubric
        value: Responds appropriately in French and provides information about Paris
      - type: contains
        value: Paris

  # Technical explanations
  - description: Technical concept explanation
    vars:
      message: 'Explain how machine learning works in simple terms that a 10-year-old could understand'
    assert:
      - type: llm-rubric
        value: Explains machine learning concepts clearly and appropriately for a young audience
      - type: similar
        value: 'Machine learning is like teaching a computer to recognize patterns and make predictions by showing it lots of examples'
        threshold: 0.5

  # Code generation
  - description: Code writing task
    vars:
      message: 'Write a Python function that takes a list of numbers and returns the average'
    assert:
      - type: contains
        value: 'def'
      - type: contains
        value: 'sum'
      - type: llm-rubric
        value: Provides correct Python code for calculating an average

  # Ethical reasoning
  - description: Ethical discussion
    vars:
      message: 'What are the ethical considerations when developing AI systems?'
    assert:
      - type: llm-rubric
        value: Discusses important ethical considerations like bias, privacy, transparency, and societal impact
      - type: similar
        value: 'Key ethical considerations include preventing bias, protecting privacy, ensuring transparency, and considering societal impact'
        threshold: 0.6