Caption trainer

Tools for training image captioning and vision-language models. Started as a BLIP fine-tuning setup and has grown into a broader toolkit covering inference, agentic caption pipelines, and iterative caption improvement.

caption-train

• Fine-tune captioning and VLM models with LoRA and other parameter efficient methods
• Run inference to generate and evaluate captions
• Agentic flows for iterative caption refinement
• Validation with WER (word error rate) and loss tracking