astroos/meena-3-antigravity-cot

🪐 Meena AI (Antigravity Shastric CoT Edition)

Meena AI is the core reasoning and consultation LLM of AstroOS, fine-tuned from Qwen2.5-7B-Instruct using Unsloth 4-bit QLoRA.

This edition embeds Antigravity's Deep Shastric Chain-of-Thought (<think>...</think>) architecture directly into the model weights, teaching the model multi-layered astrological pattern recognition without relying on canned clichés or superficial generalizations.

---

🏛️ Core Pillars & Capabilities

1. Explicit Chain-of-Thought (<think>...</think>):

   • Rigorous 7-step internal reasoning protocol before producing user responses.
   • Cross-checks: Lagna verification, functional lordships, natural vs functional benefics/malefics, retrogression/dignity, yoga cancellation, divisional vargas (D9/D10), and dasha-gochara timing.

2. Antigravity Conversational Voice:

   • Direct, grounded, compassionate, and sharp.
   • Zero Canned Clichés: Completely free from bot phrases like "Delay does not mean denial 🌸", "Stars show the path but you walk it", or theatrical dramatic warnings.
   • Zero Shloka Showmanship: Doesn't vomit raw unparsed Sanskrit verses; delivers clear, practical life and karma guidance.

3. Trilingual Mastery:

   • Conversational modern Hinglish (natural pair-programming and client consultation voice).
   • Pure Devanagari Hindi (authentic, respectful Shastric consultation).
   • High-precision astrological English.

---

🚀 How to Use

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

base_model_id = "Qwen/Qwen2.5-7B-Instruct"
lora_model_id = "astroos/meena-3-antigravity-cot"

tokenizer = AutoTokenizer.from_pretrained(lora_model_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
model = PeftModel.from_pretrained(base_model, lora_model_id)

messages = [
    {
        "role": "user",
        "content": "Male, 30 June 1971, 04:57:48 AM, Vadodara. Why is my Saturn Mahadasha challenging despite Saturn being Yogakaraka for Taurus Lagna?"
    }
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

outputs = model.generate(**inputs, max_new_tokens=1024, temperature=0.7)
response = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
print(response)

---

📊 Training Specifications

• Base Model: unsloth/Qwen2.5-7B-Instruct-bnb-4bit
• Method: QLoRA (r=16, alpha=16, dropout=0) via Unsloth
• Dataset: astroosmb/meena-antigravity-cot-dataset (15,000 verified Shastric CoT pairs, 104 MB)
• Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Epochs / Checkpoints: Trained to optimal loss convergence on Nvidia Tesla T4 GPU.

---

Engineered by AstroOS Team.