hebertgo/knowledgebase-intent-llm

knowledgebase-intent-llm

Fine-tuned model for intent detection in a knowledge management iOS app.

Model Details

• —Base Model: mlx-community/Qwen2.5-1.5B-Instruct-4bit
• —Model Type: intent_detection
• —Format: complete_merged
• —Framework: MLX
• —Training Examples: 5000
• —Training Iterations: 100

Usage

from mlx_lm import load, generate

# Load the model
model, tokenizer = load("hebertgo/knowledgebase-intent-llm")

# Generate intent classification
prompt = '''You are a helpful AI assistant for a knowledge-management app on an iPhone. Analyze the user's request and respond with JSON in this format:
{
  "action": "Search|Create|Clarify|Conversation",
  "response": "User-friendly response message",
  "contentType": "videos|bookmarks|todos",
  "topic": "extracted topic or null"
}

User query: find videos about python'''

response = generate(model, tokenizer, prompt=prompt, max_tokens=256)
print(response)

iOS Integration

This model is designed for use in iOS apps with MLX Swift:

let config = ModelConfiguration(
    id: "hebertgo/knowledgebase-intent-llm",
    defaultPrompt: ""
)

let model = try await LLMModelFactory.shared.loadContainer(
    configuration: config
)

Training Details

• —Fine-tuning Method: LoRA with model fusion
• —Export Date: 2025-06-24T17:12:38.111419
• —Fusion Completed: True

Expected Outputs

The model generates JSON responses with these action types:

• —Search: Find existing content (videos, bookmarks, todos)
• —Create: Add new content
• —Clarify: Request more information
• —Conversation: General chat responses

Content types supported:

• —videos
• —bookmarks
• —todos

Performance

Optimized for Apple Silicon devices with MLX framework for efficient on-device inference.