essAi
essAi is a fine-tuned Qwen3-4B model that writes authentic college application essays (Common App personal statement style) in a natural human voice. A larger 9B sibling (Qwen3.5-9B) is available at alphanozcan/essAi-9b.
Training
Two-stage fine-tune on ~19.7k human-written essays:
| Stage | Data | Details |
|---|---|---|
| SFT | 270 real admissions essays from publicly published example collections (JHU "Essays That Worked", College Essay Guy, AP Study Notes) + ~19.4k human essays from the open persuade corpus | LoRA r=16 (all linear), lr 2e-4, 1 epoch, fp16 |
| DPO | Same prompt: real human essay = chosen, SFT model output = rejected (HumanLLMs method, arXiv 2501.05032) + GradGPT quality pairs | beta=0.1, lr 5e-5, 1 epoch |
Prompt from SFT data: Write a ~650-word Common App style personal statement essay. …
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("alphanozcan/essAi", torch_dtype="auto", device_map="auto")
tok = AutoTokenizer.from_pretrained("alphanozcan/essAi")
system = "You write authentic college application essays in a natural human voice, with specific personal detail, varied sentence rhythm, and honest reflection."
user = "Write a ~650-word Common App style personal statement essay about learning from failure."
prompt = tok.apply_chat_template(
[{"role": "system", "content": system}, {"role": "user", "content": user}],
tokenize=False, add_generation_prompt=True, enable_thinking=False,
)
inputs = tok(prompt, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=900, do_sample=True, temperature=0.8, top_p=0.95, pad_token_id=tok.pad_token_id or tok.eos_token_id)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
A 4-bit MLX build for Apple Silicon is available at alphanozcan/essAi-mlx.
Notes
- 4B parameters, 1 training epoch — quality reflects that; longer training and a larger base would improve coherence.
- Style metrics (human reference vs output): burstiness CV 0.544 → 0.465, mean sentence length 18.8 → 17.0.
- AI-detector behavior is not guaranteed; this model is trained on human essays for a more natural writing style, but detectors are trained classifiers and results vary.