Kanha-AI/kanha-kanha.ai-1.7b-pit-v1

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 22, 2026Architecture:Transformer Featherless Exclusive Cold

Kanha-AI/kanha-kanha.ai-1.7b-pit-v1 is a 1.7 billion parameter language model developed by Kanha-AI, based on the Qwen3-1.7B architecture. This model is fine-tuned using a PIT (Private Instruction Tuning) method combining document continuation and Q&A pairs, specifically for website-specific content. It is designed for tasks requiring recall of information from its private training corpus, with a focus on accurate and concise answers.

Loading preview...

Kanha-AI/kanha-kanha.ai-1.7b-pit-v1 Overview

Kanha-AI/kanha-kanha.ai-1.7b-pit-v1 is a specialized 1.7 billion parameter language model built upon the Qwen3-1.7B base. Developed by Kanha-AI, this model underwent a unique Private Instruction Tuning (PIT) process, which involved both document continuation and Q&A pair training. The training corpus is highly specific, focusing on particular website content rather than general knowledge.

Key Characteristics

  • Base Model: Qwen/Qwen3-1.7B, ensuring a robust foundation.
  • Training Method: Utilizes a PIT approach with 17 training documents and 170 Q&A pairs, tailored for specific information recall.
  • Parameter Count: A compact 1.7 billion parameters, making it efficient for deployment.
  • Context Length: Trained with a max length of 2048 tokens, suitable for focused interactions.
  • Data Type: Final merged model uses bfloat16 for efficient computation.
  • Evaluation Insights: Achieved 1.0 on dates_recall and urls_recall, indicating strong performance in recalling specific factual elements from its training data.

Intended Use Cases

This model is particularly suited for applications where precise recall of information from a defined, private corpus is critical. It can be beneficial for:

  • Website-specific Q&A systems: Answering user queries based on the content of a particular website.
  • Information extraction: Retrieving specific dates, URLs, or numerical data from known documents.
  • Concise factual responses: Generating direct and accurate answers within its trained domain.

Limitations

It is important to note that the model's capabilities are tied to its private training corpus. It may produce incorrect, incomplete, or memorized content outside this domain. Its evaluation results do not reflect general capabilities or production safety for broader applications.

Finetunes
1 models