Erik Kannike releases qwen3-4b-deslop, a Qwen3-4B fine-tune that avoids AI writing tics
Overview
Erik Kannike releases qwen3-4b-deslop, a fine-tune of Qwen3-4B-Instruct trained on a MacBook Air to avoid em dashes, "not X, but Y" constructions and lists of three without a system prompt.
On 40 held-out prompts, Kannike reports the model's slop score fell from 121 to 12.5 per 1,000 words, and that it answered 30 of 30 factual questions correctly.
The Q4_K_M GGUF build is a 2.5 GB download that runs locally through Ollama or llama.cpp. Kannike gives the Ollama command as ollama run hf.co/erikkannike/qwen3-4b-deslop-v4-GGUF:Q4_K_M.
Written by AI from the articles below · updated Oct 19, 9:28 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- Hacker News · Show HN, AI (20+ points)BlogShow HN: Qwen3-4B fine-tuned to drop AI writing tics like em dashes
AIErik Kannike releases qwen3-4b-deslop, a Qwen3-4B-Instruct fine-tune trained on a MacBook Air to avoid em dashes, "not X, but Y" constructions and lists of three without a system prompt. On 40 held-out prompts, its slop score fell from 121 to 12.5 per 1000 words, and it answered 30 of 30 factual questions correctly. The Q4_K_M GGUF build is a 2.5 GB download that runs locally through Ollama or llama.cpp.
Heat Trend
Not enough continuous observations to show a trend yet.