voira-essay-lora-v1

Voira Essay LoRA v1

A LoRA adapter fine-tuned on top of Llama 3.1 8B Instruct to write essays from detailed outlines. Part of the Voira essay writing system.

What This Does

Given a structured outline containing section plans, voice notes, key phrases, and emotional cues, this adapter generates polished, human-sounding essays. It's trained to follow the outline's structure precisely while writing with genuine voice — varied rhythm, concrete imagery, emotional honesty.

This is the base essay craft LoRA — it teaches general essay writing quality. Per-user LoRA adapters can be stacked on top for individual voice personalization.

Quick Start

from unsloth import FastLanguageModel
from peft import PeftModel
 
# Load base model
model, tokenizer = FastLanguageModel.from_pretrained(
    "meta-llama/Meta-Llama-3.1-8B-Instruct",
    load_in_4bit=True,
)
 
# Load adapter
model = PeftModel.from_pretrained(model, "ekomobng/voira-essay-lora-v1")
FastLanguageModel.for_inference(model)
 
# Generate
outline = """--- OUTLINE ---
[your outline here]
--- END ---
 
Now write the essay based on this outline:"""
 
messages = [
    {"role": "system", "content": "You are an expert essay writer. Given a detailed outline, write the essay. Output ONLY the essay text, nothing else."},
    {"role": "user", "content": outline},
]
 
input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
 
outputs = model.generate(
    **inputs,
    max_new_tokens=2048,
    temperature=0.7,
    top_p=0.9,
    repetition_penalty=1.1,
    do_sample=True,
)
 
generated = tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
print(generated)

Outline Format

The model expects outlines wrapped with --- OUTLINE --- and --- END --- markers, followed by the cue Now write the essay based on this outline:. The outline schema includes:

• [Prompt:] — the essay question or topic
• [Word Limit:] — target length
• [Overall Emotional Arc:] — macro emotional trajectory
• [Overall Tone & Voice:] — the writer's baseline register
• [Introduction / Hook:] — opening strategy with MOOD and STRESS notes
• [Body:] — numbered sections with content plans, key phrases, voice/pacing notes, MOOD, TONAL SHIFT, and STRESS
• [Conclusion / Summary:] — closing strategy with final emotional landing

Training Details

Training Data

105 high-quality essays (pre-2010, verified non-AI-generated) with reverse-engineered outlines. Outlines were generated using Claude to capture the writer's planning notes — structure, specific phrases, emotional architecture, voice notes, and craft-level decisions. The training format uses train_on_responses_only=True so the model is only scored on essay generation, not outline prediction.

Architecture Context

This adapter is part of a layered LoRA system:

Base Llama 3.1 8B Instruct
    └── voira-essay-lora-v1 (this adapter — general essay craft)
        └── per-user LoRA (future — individual voice adaptation)

The base adapter teaches essay craft patterns: narrative pacing, show-don't-tell, concrete detail, emotional resonance, varied sentence rhythm. Per-user adapters (trained from each user's edited outputs) will be stacked on top for voice personalization.

Limitations

• Trained on 105 examples — quality improves with more data
• Best suited for personal narrative and college application essay styles
• Generated essays may run longer than the specified word limit
• Works best with detailed outlines; very sparse outlines produce more generic output

Developed By

Voira — AI-powered essay writing that preserves human voice.