chancharikm/sft_caption_generation_20260222_ep6_lr3e5_qwen3-vl-8b

Overview

This model, chancharikm/sft_caption_generation_20260222_ep6_lr3e5_qwen3-vl-8b, is a fine-tuned iteration of the Qwen3-VL-8B-Instruct base model, developed by chancharikm. It features 8 billion parameters and maintains the original model's substantial 32,768 token context length, making it suitable for processing extensive visual and textual inputs.

Key Capabilities

• Image Caption Generation: The model has been specifically fine-tuned on the sft_caption_generation_20260222 dataset, indicating its primary strength in generating descriptive captions for images.
• Vision-Language Understanding: Inherits the multimodal capabilities of the Qwen3-VL-8B-Instruct architecture, allowing it to interpret visual information and produce relevant textual outputs.

Training Details

The model underwent supervised fine-tuning (SFT) with a learning rate of 3e-05 over 6 epochs. Training utilized a distributed setup across 8 GPUs, with a total batch size of 128 (achieved with gradient accumulation steps of 2). The optimizer used was adamw_torch_fused with a cosine learning rate scheduler and a warmup ratio of 0.05.

Intended Use Cases

This model is best suited for applications requiring automated, high-quality image descriptions, such as:

• Content accessibility (e.g., generating alt text for images).
• Automated content moderation or tagging.
• Enhancing searchability of image databases through descriptive captions.