AxionLab-official/DogeAI-vTEST-Reasoning

DogeAI-v2.0 🐶🔥 (LoRA Weights Only)

⚠️ Important Notice
This repository does NOT contain a full language model.

It only provides LoRA fine-tuned weights for the base model Baguettotron.
To use DogeAI-v2.0, you must load it on top of the base model.

---

🔍 What is this?

DogeAI-v2.0 is a LoRA adaptation trained to give the base model:

• Better conversational flow
• Clearer reasoning
• Stronger math and logic responses
• A more direct and confident assistant style

This repository contains only the LoRA weights, which are lightweight and efficient.

---

🧠 Base Model (Required)

You must use the following base model:

PleIAs/Baguettotron

Without it, these weights will not work.

---

🧩 What is LoRA?

LoRA (Low-Rank Adaptation) is a fine-tuning technique that:

• Keeps the original model frozen
• Applies small, efficient weight updates
• Uses much less memory than full fine-tuning

This makes DogeAI-v2.0:

• Fast to load
• Easy to experiment with
• Friendly for consumer hardware

---

🚀 How to Use

1️⃣ Install dependencies

pip install torch transformers peft
2️⃣ Load the model + LoRA
python

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

BASE_MODEL = "PleIAs/Baguettotron"
LORA_PATH = "dogeai_v2_lora"  # or dogeai_v2_lora_10pct

print("Loading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
tokenizer.pad_token = tokenizer.eos_token

print("Loading base model...")
model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    torch_dtype=torch.float32
)

print("Applying DogeAI-v2.0 LoRA 🐶🔥")
model = PeftModel.from_pretrained(model, LORA_PATH)
model.eval()
3️⃣ Chat loop example
python
Copiar código
print("\nDogeAI-v2.0 ready! Type 'exit' to quit.\n")

while True:
    user_input = input("You: ")
    if user_input.lower() in ["exit", "quit"]:
        break

    prompt = f"""
<|im_start|>user
{user_input}
<|im_end|>
<|im_start|>assistant
"""

    inputs = tokenizer(prompt, return_tensors="pt")
    inputs.pop("token_type_ids", None)

    with torch.no_grad():
        output = model.generate(
            **inputs,
            max_new_tokens=200,
            do_sample=True,
            temperature=0.7,
            top_p=0.95,
            repetition_penalty=1.2,
            eos_token_id=tokenizer.eos_token_id
        )

    response = tokenizer.decode(output[0], skip_special_tokens=True)
    response = response.split("<|im_start|>assistant")[-1].strip()

    print(f"\nDogeAI 🐶: {response}\n")
💻 Hardware Notes
Runs on CPU (slow but works)

Recommended: GPU for better speed

LoRA keeps memory usage low compared to full fine-tuning

🎯 What this is NOT
❌ Not a standalone model

❌ Not a GGUF / quantized release

❌ Not an instruction-following base model by itself

This is an enhancement, not a replacement.

🐕 DogeAI Philosophy
Fast. Honest. No hallucinated confidence.
Clear answers, real reasoning, no nonsense.

Made for experimentation, learning, and pushing models further 🚀