Warlord-K/kanha-incidentio-incidents-1.7b-qlora-v1

Model Overview

Warlord-K/kanha-incidentio-incidents-1.7b-qlora-v1 is a 1.7 billion parameter language model built upon the Qwen/Qwen3-1.7B base model. It has been fine-tuned using the QLoRA method, specifically targeting question answering based on the Incident.io documentation. The model was trained with a maximum sequence length of 2048 tokens and utilizes a bfloat16 merged dtype.

Key Characteristics & Training

• Base Model: Qwen/Qwen3-1.7B
• Training Method: QLoRA, with specific LoRA targets including q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, and down_proj.
• Dataset: Derived from Incident.io documentation, with 276 training records and 33 validation records.
• Hyperparameters: Includes a learning rate of 5e-05, 4 epochs, and a per-device batch size of 4.

Evaluation & Intended Use

Evaluation metrics show strong performance in specific recall tasks such as dates_recall (1.0), list_recall (1.0), numbers_recall (1.0), and urls_recall (1.0). The model exhibits a refusal_rate of 0.0, though it has a requires_review_rate of 1.0. This checkpoint is primarily intended for research purposes, specifically for comparing training methods on website-derived datasets and for controlled evaluation of its ability to answer questions based on website content.

Limitations

Users should be aware that the model may produce incorrect, incomplete, or stale answers. It also has the potential to memorize training content. It is crucial to review all outputs, test for potential failure cases, and thoroughly qualify the model in its target runtime environment before any user-facing deployment.