Warlord-K/kanha-kanha.ai-0.6b-grounded-qlora-3ep-v1

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 24, 2026Architecture:Transformer Featherless Exclusive Cold

Warlord-K/kanha-kanha.ai-0.6b-grounded-qlora-3ep-v1 is a 0.8 billion parameter language model developed by Warlord-K, based on the Qwen3-0.6B architecture. Fine-tuned using QLoRA with a 32K context length, this model is specifically designed for grounded question answering on website-derived content. It excels at providing concise answers strictly from supplied context, making it suitable for controlled evaluation of website Q&A systems.

Loading preview...

Overview

Warlord-K/kanha-kanha.ai-0.6b-grounded-qlora-3ep-v1 is a 0.8 billion parameter language model built upon the Qwen/Qwen3-0.6B base model. It was fine-tuned using the QLoRA method over 3 epochs, with a maximum sequence length of 4096 tokens. This model is specifically engineered for grounded inference, meaning it requires and adheres strictly to a retrieved source context for answering questions.

Key Capabilities

  • Grounded Question Answering: Designed to answer questions solely from provided context, refusing to answer if information is absent.
  • Contextual Recall: Achieves high recall rates for dates (1.0), numbers (0.964), and URLs (1.0) within the supplied context.
  • Low Refusal Rate: Exhibits a low refusal rate (0.038) when context is available, indicating effective utilization of provided information.
  • Concise Responses: Trained to provide brief and direct answers.

Intended Use

This model is primarily intended for:

  • Research purposes, particularly for comparing training methods on the Kanha website-derived dataset.
  • Controlled evaluation of website question answering systems, ensuring responses are strictly based on provided information.

Limitations

Users should be aware that the model may produce incorrect, incomplete, or outdated answers. It can also memorize training content. Thorough review of outputs and testing of failure cases are recommended before deployment in user-facing applications.