Qwen3-4B de-slop v4, GGUF
Quantised builds of erikkannike/qwen3-4b-deslop-v4, a fine-tune of Qwen3-4B-Instruct-2507 that writes without the assistant habits (em dashes, "not X, but Y", lists of three, signposting) and needs no system prompt. The model card there has the numbers and limitations, and the write-up is at 100toolkit.com/blog/deslop-model.
| file | size | notes |
|---|---|---|
qwen3-4b-deslop-v4-Q4_K_M.gguf |
2.5 GB | recommended for laptops |
qwen3-4b-deslop-v4-Q8_0.gguf |
4.3 GB | closest to the full weights |
ollama run hf.co/erikkannike/qwen3-4b-deslop-v4-GGUF:Q4_K_M
llama-cli -hf erikkannike/qwen3-4b-deslop-v4-GGUF:Q4_K_M
Both files were converted with llama.cpp from the merged bf16 weights. Q4_K_M was quantised from the Q8_0 file. The chat template is the original Qwen3 one, and the model works best with no system prompt at temperature 0.7.
Recommended settings
Use a repetition penalty of about 1.1 (--repeat-penalty 1.1 in llama.cpp), temperature 0.7 and top-p 0.9, with no system prompt. Without a repetition penalty the model can fall into repeating one sentence pattern in long outputs, sometimes after a few hundred words, because every training answer was short (under about 250 words). The GGUF repo ships an Ollama params file with these values, so ollama run uses them automatically. Long-form training data is planned for the next version.