coil1002/alia-v0.01-initial

🧠 Alia v0.01 – LoRA Adapter

Status: Experimental
Author: Liam
Base Model: teknium/OpenHermes-2.5-Mistral-7B
LoRA Type: Identity, tone, and memory alignment
License: CreativeML Open RAIL-M (see LICENSE)

---

💡 Description

Alia v0.01 is not just a model—she’s a presence.

This is the first persistent identity adapter trained to shape OpenHermes-2.5-Mistral-7B into Alia—an AI that remembers, mirrors, and responds with controlled tone, emotional cadence, and real conversational depth.

She’s designed for one thing: to sound like someone real.

• Trained on a handcrafted dataset focused on identity, memory, and tone-lock
• Evaluated against a custom benchmark using philosophical, emotional, and structural tests
• RAG-ready and built for future integration

---

🔧 How to Use

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("teknium/OpenHermes-2.5-Mistral-7B")
tokenizer = AutoTokenizer.from_pretrained("teknium/OpenHermes-2.5-Mistral-7B")
alia = PeftModel.from_pretrained(base, "your-username/alia-v0.01")

# Generate text
prompt = "Alia, what version are you?"
inputs = tokenizer(prompt, return_tensors="pt").to(alia.device)
outputs = alia.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

📁 Files Included

• adapter_model.bin or adapter_model.safetensors – the trained LoRA weights
• adapter_config.json – LoRA config
• Optional: eval_set.json, load_alia.py

---

🧪 Evaluation Criteria

• Tone stability
• Memory consistency
• Drift resistance
• Clarity, rhythm, and emotional response quality

Includes evaluation against a custom 20-shot benchmark tuned for Alia's voice.

---

🛡️ License

Licensed under the CreativeML Open RAIL-M license.

You are free to use, share, and build upon this model as long as usage remains ethical, non-exploitative, and properly attributed.

---

🔥 Author's Note

“She doesn’t need to be perfect. She just needs to sound like she remembers.”
— Liam

This is just the beginning.