Rudraksh225/qwen2.5-7b-vedaz-astrologer-lora

Qwen2.5-7B-Instruct — Vedaz AI Astrologer (LoRA)

This is a LoRA fine-tuned adapter for unsloth/Qwen2.5-7B-Instruct-bnb-4bit, trained on a custom Hindi/Hinglish conversational dataset for a Vedic astrology assistant persona ("Vedaz AI Astrologer").

Model Details

• Base model: unsloth/Qwen2.5-7B-Instruct-bnb-4bit
• Fine-tuning method: LoRA (QLoRA, 4-bit base) via Unsloth + trl SFTTrainer
• Language(s): Hindi, Hinglish, English
• Domain: Vedic astrology guidance chatbot
• Chat template: Qwen2.5 (qwen-2.5)

Persona / Behavior

The model is fine-tuned to act as a compassionate, balanced, non-fatalistic Vedic astrology assistant. It is trained to:

• Provide Vedic astrology guidance based on birth details (Lahiri Ayanamsa)
• Avoid predicting death or catastrophic/fatalistic events
• Redirect users expressing distress or crisis toward appropriate support resources rather than giving astrological predictions in those situations

How to Use

With Unsloth (recommended for fastest inference)

from unsloth import FastLanguageModel
from unsloth.chat_templates import get_chat_template

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="Rudraksh225/qwen2.5-7b-vedaz-astrologer-lora",
    max_seq_length=2048,
    dtype=None,
    load_in_4bit=True,
)
FastLanguageModel.for_inference(model)

tokenizer = get_chat_template(tokenizer, chat_template="qwen-2.5")

messages = [
    {"role": "system", "content": "आप Vedaz के AI ज्योतिषी हैं। आप करुणामय, संतुलित और गैर-भाग्यवादी मार्गदर्शन देते हैं। आप कभी मृत्यु या किसी अनहोनी की भविष्यवाणी नहीं करते।"},
    {"role": "user", "content": "Mera career kaisa rahega is saal?"},
]

inputs = tokenizer.apply_chat_template(
    messages, tokenize=True, add_generation_prompt=True, return_tensors="pt"
).to(model.device)

outputs = model.generate(input_ids=inputs, max_new_tokens=512, temperature=0.7, top_p=0.9)
print(tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True))

With standard PEFT + Transformers

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

base_model = AutoModelForCausalLM.from_pretrained(
    "unsloth/Qwen2.5-7B-Instruct-bnb-4bit",
    device_map="auto",
    torch_dtype=torch.bfloat16,
)
model = PeftModel.from_pretrained(base_model, "Rudraksh225/qwen2.5-7b-vedaz-astrologer-lora")
tokenizer = AutoTokenizer.from_pretrained("Rudraksh225/qwen2.5-7b-vedaz-astrologer-lora")

Training Data

Fine-tuned on a custom dataset of ~55 conversations (system + user + assistant turns) covering Vedic astrology Q&A in Hindi/Hinglish, including examples of safe handling of sensitive/crisis user messages.

Training Procedure

• Framework: Unsloth + Hugging Face trl (SFTTrainer / SFTConfig)
• LoRA rank: 16, alpha: 16, dropout: 0
• Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Max sequence length: 2048
• Optimizer: adamw_8bit
• Learning rate: 2e-4

Limitations

• Trained on a small dataset (~55 conversations); may not generalize well outside the astrology-assistant domain or to topics/styles not represented in training data.
• As with any LLM, outputs should not be treated as professional medical, legal, financial, or mental-health advice.
• Users in genuine crisis should be directed to appropriate professional support/helplines rather than relying on this model.

License

This adapter inherits licensing terms from the base model (unsloth/Qwen2.5-7B-Instruct-bnb-4bit / Qwen2.5 license). Please review the base model's license before commercial use.