madcatlab/Qwen-4B-Marketing

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 24, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

madcatlab/Qwen-4B-Marketing is a 4.5 billion parameter language model developed by Mad Cat Lab, fine-tuned for digital marketing and marketing mix dialogues. Optimized for CPU usage, it supports up to 262,144 tokens context length and excels in both Turkish and English. This model is specifically designed for marketing-focused question-answering and assistant scenarios, offering efficient performance on standard hardware without requiring a GPU.

Loading preview...

Overview

madcatlab/Qwen-4B-Marketing is a 4.5 billion parameter language model developed by Mad Cat Lab, specifically fine-tuned for digital marketing and marketing mix dialogues. It is available in both GGUF (Q4_K_M) for CPU-only operation and Safetensors (bf16) for GPU-accelerated inference. A key feature is its exceptionally low hardware requirements, capable of running comfortably on an 8 GB RAM office computer with a 16K context, thanks to its architecture utilizing linear attention for most layers, resulting in a small KV cache.

Key Capabilities

  • Specialized Domain: Fine-tuned for digital marketing, advertising, SEO, and strategy-focused question-answering in both Turkish and English.
  • CPU-Optimized: Designed to run efficiently on CPUs using llama.cpp, making it accessible without dedicated GPU hardware.
  • Extended Context: Supports a context length of up to 262,144 tokens, with practical usage up to 32K tokens on typical hardware.
  • Efficient Performance: Achieves measured speeds of 52.4 tokens/second on an Apple M4 Max (8 threads) for token generation.
  • Flexible Deployment: Can be deployed via llama.cpp (server or command line), transformers (GPU), or vLLM.

Important Usage Notes

  • Requires the enable_thinking=false parameter to prevent empty content fields in OpenAI-compatible API responses.
  • Ollama is currently not supported due to its embedded engine lacking implementation for the model's linear attention layers.

Good For

  • Developing assistants and chatbots for digital marketing, advertising, and SEO tasks.
  • Question-answering related to marketing strategies and concepts in Turkish and English.
  • Users seeking a resource-efficient language model that performs well on CPU-only systems for specialized marketing applications.

Finetunes
1 models