Qwen3.5-9B-heretic LLM

Overview

The Qwen3.5-9B-heretic LLM is a large language model optimized for chat interactions and logical reasoning. It supports various configuration options to tailor its behavior for specific tasks.

What you can build

This model can be used to build conversational agents, question-answering systems, and reasoning-based applications. It's particularly suited for scenarios requiring nuanced understanding and structured responses.

Inputs

• Prompt: The input text to process.
• System Prompt: A guiding prompt that shapes the model's behavior.
• Thinking Mode: Toggle between thinking modes for enhanced reasoning.
• User ID (Optional): Unique identifier for maintaining chat history.
• Session ID (Optional): Identifier for managing multiple user sessions.
• Temperature: Controls randomness of outputs (0.0 to 5.0).
• Top P: Controls token sampling from the top p percentage (0.0 to 1.0).
• Top K: Controls token sampling from the top k most likely tokens (0 to 100).
• Repetition Penalty: Adjusts likelihood of repeating tokens (0.0 to 5.0).
• Length Penalty: Influences output length (0.0 to 5.0).
• Max Tokens: Maximum number of tokens to generate (up to 16,384, thinking included).
• Min Tokens: Minimum number of tokens to generate (-1 to 2048).
• Stop Sequences: Delimiters to halt generation.
• Seed: Random seed for reproducible outputs.
• Quantization: Enables quantization for reduced memory usage.
• Do Sample: Enables/disables random sampling.

Outputs

• Generated text based on provided prompts and configurations.

Recommended settings

For general use, try setting temperature to 0.7, top_p to 0.95, and enable quantization for faster inference. Adjust max_tokens and repetition_penalty based on task requirements.

Limitations

• Output quality depends on prompt clarity and model configuration.
• May occasionally produce irrelevant or nonsensical content without proper guidance.
• On Wiro, a single request takes in about 25,600 input tokens, counting the system message, chat history, tool definitions, and your prompt together. The base model's 262,144-token window is not available here.
• Longer input is shortened automatically, without an error: the oldest history messages are dropped first, then the end of your latest message is cut. Put your question at the start of long prompts.
• Responses are capped at 16,384 output tokens, and thinking tokens count toward that cap.

Safety & compliance

The model enforces ethical guidelines and safety measures to prevent harmful, illegal, or unethical content. It is designed to be respectful and honest in its responses.