SatireLM-4B

This repository has the quantized GGUF varients of the model. It is intended to be used with inference engines like llama.cpp, Ollama, and vLLM. If you would like the fine-tuned safetensor weights, go to this repository. SatireLM is a satirical fine-tune of Qwen3.5-4B-Base, designed to confidently give satirical responses, while still formatting like a frontier model, except this model has no thinking instruct examples, so it does not think. If you would like to view some of it's responses, go here.

Model Details

  • Base Model: Qwen3.5-4B-Base
  • Training Framework: Unsloth
    • Architecture: Qwen3_5ForConditionalGeneration
    • Hidden Size: 2560
    • Intermediate Size: 9216
    • Vocab Size: 248320
    • Layers: 32
  • Hardware: AMD MI300 GPU
  • Training Duration: ~23 hours
  • Model Type: Causal Language Model

Intended Use

This model is made purely for entertainment, satire, and creative writing. It is designed to mimic AI formatting while generating absurd, nonsensical, or intentionally flawed outputs. Please do not treat any of its generations as reality.

Training Data

This was fine-tuned on about 200k instruct examples, with topics ranging from cooking to coding. The satire instruct pairs can be found in this repository.

Downloads last month
1,288
GGUF
Model size
4B params
Architecture
qwen35
Hardware compatibility
Log In to add your hardware

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for benni-ben/SatireLM-4B

Quantized
(41)
this model