Kartik-Kundli-AI-3B-v3.0

Kartik-Kundli-AI-3B-v3.0 🌟

Kartik-Kundli-AI-3B-v3.0 is an advanced domain-specific fine-tuned Large Language Model specialized in Vedic Astrology, birth chart analysis, and cosmic guidance. Built on top of the robust Qwen2.5-3B-Instruct architecture, this model has been optimized using Unsloth and QLoRA techniques for high-speed, accurate, and culturally nuanced multi-turn conversations.

---

🚀 Model Details

• Developer: UX4567
• Base Model: Qwen2.5-3B-Instruct
• Fine-Tuning Framework: Unsloth (QLoRA, 4-bit quantization)
• Dataset Size: 5,000+ custom curated instruction-context-response pairs
• Language: Hindi & English (Hinglish conversational support)
• Parameters: 3 Billion

---

🛠️ Training Configuration

• Rank (r): 8
• Lora Alpha: 16
• Lora Dropout: 0.0 (Optimized for Unsloth fast kernels)
• Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Optimization: Gradient Checkpointing enabled for VRAM efficiency.

---

💻 Quickstart Inference Code

You can easily load and run this model using Python and Unsloth:

from unsloth import FastLanguageModel
import torch

max_seq_length = 2048
dtype = None
load_in_4bit = True

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "UX4567/Kartik-Kundli-AI-3B-v3.0",
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
)
FastLanguageModel.for_inference(model) # Enable native fast inference

prompt = """Aap ek expert Vedic Astrologer hain. User ki birth details aur unke sawaal ke aadhar par ek sateek, sankshipt aur sakaratmak jyotishiya margdarshan dein.
Context: User Details: Name: Rahul Sharma, Birth Place: Jaipur, Rajasthan. Sawaal: Mera career aur job change kab tak yoga ban raha hai?
Response:"""

inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=256, use_cache=True)
print(tokenizer.batch_decode(outputs)[0])