maftuh-main/batik-qwen1.5b-merged

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 10, 2026Architecture:Transformer Featherless Exclusive Cold

maftuh-main/batik-qwen1.5b-merged is a 1.5 billion parameter causal language model, fine-tuned from Qwen2.5-1.5B-Instruct by Muhammad Maftuh. This model is specifically optimized as an expert assistant for Indonesian Batik culture, providing formal, informative, and educational responses. It excels at answering questions regarding Batik history, philosophy, techniques, motifs, and its UNESCO heritage status. The model is designed for use in educational applications, digital museum chatbots, or cultural inquiry features.

Loading preview...

Wastra.ai — Batik Qwen 1.5B (Merged) Overview

Wastra.ai is a specialized conversational model developed by Muhammad Maftuh, fine-tuned from Qwen2.5-1.5B-Instruct. It functions as an expert assistant for Indonesian Batik culture, providing detailed and formal information. The model was created by merging a base model with a LoRA adapter, trained using QLoRA (4-bit quantization).

Key Capabilities

  • Batik Culture Expertise: Answers questions on Batik history, philosophy, manufacturing techniques (mori, malam, dyeing), various regional motifs, and its UNESCO heritage status.
  • Formal and Educational Tone: Responds with a formal, informative, and educational language style.
  • Topic Adherence: Elegantly declines questions outside the scope of Batik and related culture.
  • Language: Primarily functions in Bahasa Indonesia.

Training Details

The model was fine-tuned using a curated synthetic conversational dataset focused on Batik, covering history, techniques, motif philosophy, and UNESCO recognition. The training procedure involved QLoRA with 4-bit NF4 quantization and a LoRA adapter (r=16, alpha=32, dropout=0.05) targeting key attention modules. It utilized bf16 mixed precision and a paged_adamw_8bit optimizer with a learning rate of 1e-4.

Good for

  • Educational Chatbots: Ideal for learning applications, digital museum chatbots, or cultural inquiry features like "BatikLens."
  • Specialized Information Retrieval: Provides focused, in-depth knowledge on Indonesian Batik.

Limitations

  • Domain Specificity: Performance is limited to Batik-related topics; not designed for general-purpose queries.
  • Reasoning Capacity: As a 1.5B parameter model, its complex reasoning capabilities are more limited compared to larger models.
  • Dataset Dependency: Answer quality is influenced by the scope and quality of its synthetic/curated training dataset.

Finetunes
2 models