Qwen3-4B with the AI writing tics trained out
Open-weights (Apache-2.0) fine-tune of Qwen3-4B-Instruct with AI-assistant writing habits such as em dashes, "not X, but Y" and lists of three trained out; runs locally via Ollama or llama.cpp with no system prompt
Are you the maker of Qwen3-4B with the AI writing tics trained out? Claim this listing
What is Qwen3-4B with the AI writing tics trained out?
Qwen3-4B with the AI writing tics trained out is Open-weights (Apache-2.0) fine-tune of Qwen3-4B-Instruct with AI-assistant writing habits such as em dashes, "not X, but Y" and lists of three trained out; runs locally via Ollama or llama.cpp with no system prompt It's designed for local-llm tinkerers and researchers who want a small model that writes short english prose without common assistant tics.
About Qwen3-4B with the AI writing tics trained out
Key Strengths
- Open Apache-2.0 weights: merged model, 126 MB LoRA adapter and four-neuron edit, plus Q4_K_M (2.5 GB) and Q8_0 GGUFs; runs locally with Ollama or llama.cpp
- Author's eval on 40 held-out prompts: slop score 121 → 12.5 per 1,000 words, em dashes 20.5 → 0, 30/30 factual quiz, genre/length fit 80% vs 78% for the base model, with no system prompt
- Transparent write-up of every training round, including failures (invented studies, ', which' chains) and how they were filtered out
Limitations to Consider
- The author says the voice is flat, it still invents names and places, and it runs long (~200 words when asked for 150)
- Only tested on short English prose; the score comes from the author's own regex detector with no blind human test yet
- Mixed Hacker News reactions: one commenter said its output still scored 100% AI on Pangram, another found an example answer unformatted and three times longer
Best For
Local-LLM tinkerers and researchers who want a small model that writes short English prose without common assistant tics.
Skip it if: You want a hosted writing app, long-form, code or non-English output, or text that passes AI detectors.
Pricing
flexible pricing options
Free: open weights under Apache-2.0 on Hugging Face (bf16 merged model, LoRA adapter, Q4_K_M 2.5 GB and Q8_0 GGUF). Runs locally via Ollama, llama.cpp or transformers; no hosted version. (checked Oct 20, 2026)
Sources
- https://huggingface.co/erikkannike/qwen3-4b-deslop-v4
- https://huggingface.co/erikkannike/qwen3-4b-deslop-v4-GGUF
- https://100toolkit.com/blog/deslop-model/
- https://news.ycombinator.com/item?id=50151928
Pros, cons and pricing compiled from public sources (official pricing/docs pages, published reviews, user reports), not hands-on testing.
Compare Qwen3-4B with the AI writing tics trained out
Similar Tools in AI Writing Tools
Happy Hangul Day! An RNN for Generating Korean Handwriting Strokes
Happy Hangul Day! An RNN for Generating Korean Handwriting Strokes - AI-powered tool
Try Happy Hangul Day! An RNN for Generating Korean Handwriting StrokesFrequently Asked Questions
What is Qwen3-4B with the AI writing tics trained out?
Qwen3-4B with the AI writing tics trained out is Open-weights (Apache-2.0) fine-tune of Qwen3-4B-Instruct with AI-assistant writing habits such as em dashes, "not X, but Y" and lists of three trained out; runs locally via Ollama or llama.cpp with no system prompt It's designed for local-llm tinkerers and researchers who want a small model that writes short english prose without common assistant tics.
How much does Qwen3-4B with the AI writing tics trained out cost?
Qwen3-4B with the AI writing tics trained out offers flexible pricing options You can check pricing on their official website.
Who should use Qwen3-4B with the AI writing tics trained out?
Qwen3-4B with the AI writing tics trained out is best suited to local-llm tinkerers and researchers who want a small model that writes short english prose without common assistant tics.