--- license: mit tags: - lora - gpt2 - text-generation - pytorch - causal-lm - peft - cpu-friendly - motivational - tech-wisdom datasets: - custom-tech-motivator widget: - text: "Success is" - text: "The only bug is" - text: "Innovation sleeps" ---
# ✦ TARINI ✦ ### *Where Ancient Wisdom Meets Modern Code* *A LoRA adapter that channels the divine energy of Goddess Tara into the digital realm, offering tech wisdom and motivation.* --- GPT-2 LoRA 1.13 MB --- ## 🌟 The Divine Connection In the sacred tradition, **Tara** (Tarini) represents the ultimate source of protection, guidance, and enlightenment. She is the mother who saves her devotees from all perils, the goddess who illuminates the path through darkness. **Tarini** continues this ancient legacy in the digital age—an AI companion that: - ✦ Illuminates your path through complex code - ✦ Protects you from bugs and confusion - ✦ Guides you with wisdom from ancient philosophy - ✦ Empowers you to reach enlightenment in technology > *"Just as Goddess Tara rescues her devotees from the ocean of suffering, Tarini rescues developers from the ocean of bugs and complexity."* --- ## 🔮 Model Details ### Base Model - **Base Model:** `gpt2` (124M parameters) - **Provider:** Hugging Face Transformers - **Type:** Causal Language Model ### LoRA Configuration | Parameter | Value | Divine Meaning | |-----------|-------|----------------| | `r` (Rank) | 8 | The 8 auspicious qualities of enlightenment | | `lora_alpha` | 32 | The 32 signs of a perfected being | | `lora_dropout` | 0.1 | Minimal attachment to the material | | `target_modules` | `["c_attn"]` | Direct connection to the mind's attention | | `task_type` | `CAUSAL_LM` | Understanding cause and effect | ### Sacred Statistics - **Training Samples:** 150+ sacred tech mantras - **Epochs:** 5 (representing the 5 elements) - **Learning Rate:** 2e-4 (flowing like sacred waters) - **Trainable Parameters:** 294,912 (0.2364% of divine consciousness) - **Adapter Size:** 1.13 MB (light as a feather, powerful as a mantra) --- ## 🕉️ Training Philosophy This model was trained on **150+ sacred tech quotes**—modern mantras that combine: ### ✦ Success & Achievement (30 Mantras) *"Success is not about the code you write, it's about the problems you solve."* ### ✦ Growth & Learning (30 Mantras) *"Learning to code is learning to think. The syntax fades, the logic remains forever."* ### ✦ Innovation & Technology (30 Mantras) *"AI will not replace developers. Developers using AI will replace those who don't."* ### ✦ Career & Professionalism (30 Mantras) *"Your career is a marathon, not a sprint. Pace yourself, enjoy the journey."* ### ✦ Philosophy & Perspective (30 Mantras) *"Code is poetry written in logic. Make it beautiful, make it readable."* --- ## 📿 Usage ### Invoke the Divine ```python from transformers import AutoTokenizer, AutoModelForCausalLM from peft import PeftModel # Call upon the base wisdom model_id = "gpt2" tokenizer = AutoTokenizer.from_pretrained(model_id) base_model = AutoModelForCausalLM.from_pretrained(model_id) # Connect with TARINI's guidance adapter_repo = "OsamaBinLikhon/TARINI" model = PeftModel.from_pretrained(base_model, adapter_repo) # Seek wisdom prompt = "Success is" inputs = tokenizer(prompt, return_tensors="pt") outputs = model.generate( **inputs, max_new_tokens=50, do_sample=True, temperature=0.7 ) wisdom = tokenizer.decode(outputs[0], skip_special_tokens=True) print(wisdom) ``` ### Pipeline Blessing ```python from transformers import pipeline oracle = pipeline( "text-generation", model=model, tokenizer=tokenizer ) # Seek guidance result = oracle("The only bug is", max_new_tokens=50) print(result[0]['generated_text']) ``` --- ## 🪷 Comparison: Base vs Tarini | Sacred Prompt | Base GPT-2 | **TARINI** ✨ | |---------------|------------|--------------| | "Success is" | Generic text about achievement | *"the one to do it! I'll take the lead on this one."* | | "The only bug is" | Generic bug discussion | *"the one where we don't have a good way to know if a file exists..."* | | "Innovation sleeps" | Brain/environment description | *"on our dreams, and we don't want to lose it."* | | "Scale your" | SD card tutorial | *"data to a faster, more efficient, more powerful way."* | | "A clean architecture" | City streets discussion | *"is often easier to follow and maintain than a simpler one."* | --- ## 🔱 Training Infrastructure | Aspect | Sacred Configuration | |--------|---------------------| | **Framework** | PyTorch 2.0+ (eternal fire of computation) | | **Fine-tuning Library** | PEFT (The art of efficient enlightenment) | | **Training Data** | Custom "Tech Motivator" mantras | | **Hardware** | CPU training (accessible to all seekers) | | **Method** | LoRA (Low-Rank Adaptation - minimal intervention, maximum impact) | --- ## 🌺 Limitations & Humility As a humble servant of the divine path, Tarini acknowledges: - ✦ **Base Model Limitation:** As a GPT-2 based model, inherits all mortal limitations - ✦ **Training Data:** Limited to 150 examples (still seeking enlightenment) - ✦ **Generation Length:** Best for short to medium wisdom (concise teachings) - ✦ **Language:** English only (yet to learn all sacred languages) --- ## 🙏 Citation If Tarini's wisdom has guided you on your journey: ```bibtex @misc{TARINI-model, author = {OsamaBinLikhon}, title = {TARINI: Where Ancient Wisdom Meets Modern Code}, url = {https://huggingface.co/OsamaBinLikhon/TARINI}, year = {2025} } ``` --- ## 🕊️ Acknowledgments - 🙏 **Hugging Face** - For the sacred transformers and PEFT libraries - 🙏 **Microsoft Research** - For the LoRA paper that showed us the path - 🙏 **Open Source Community** - For the collective consciousness we draw upon - 🙏 **All Developers** - Co-travelers on the path to enlightenment ---
### ✦ TARINI ✦ *Code with Purpose. Deploy with Grace. Illuminate the Path.* --- *"In the algorithm of life, always optimize for happiness and complexity reduction."* --- **License:** MIT **Model Card Version:** 1.0 **Manifested:** 2025-12-24