--- license: apache-2.0 tags: - trl - sft - hinglish language: - en - hi base_model: - Qwen/Qwen2.5-7B pipeline_tag: text-generation --- # 🚀 Zira-Z.1 🌟 ### *The Bilingual Beast Built on Qwen 2.5 (7B)* ![Zira-Z.1 Banner](img/banner.png) --- ## 🧠 Model Highlights > **Zira-Z.1** isn't just a model — it's a revolution in understanding *both* English and Hinglish. > Born from the powerful DNA of **Qwen 2.5 (7B)**, this multilingual marvel was fine-tuned for raw text generation across two of the most widely spoken languages in the world. - 💥 **Base**: Qwen 2.5 - 7B (One of the finest open LLMs out there) - 🗣️ **Languages**: English 🇬🇧 + Hinglish 🇮🇳 (Code-mixed, no pure Hindi) - 🔧 **Training**: Fine-tuned on diverse bilingual corpora — clean, simple text format (non-instruct) - 🦾 **Purpose**: General-purpose **text generation**, especially where English and Hinglish blend naturally **Please NOTE that this is a basic text generation model and lacks coherence in its output; the release of the new instruct model has been delayed due to resource constraints, with an expected launch in approximately 5 days.** --- ## 🔍 Why Zira-Z.1? Because **multilingual LLMs** are cool. But **Zira-Z.1** is cooler. 😎 - 🔗 Code-switching? Natural. - ✍️ Generates culturally fluent, relatable Hinglish. - 📚 Handles casual text, commentary, social chatter, and more. - 🎯 Perfect for early-stage Indic bilingual applications and experimentation --- ## 📉 Training Curve > *She trained hard, and it shows...* ![Insert loss curve here](img/Figure_1.png) --- ## 🛠️ Usage ```import transformers from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("HyperX-Sen/Zira-Z.1") model = AutoModelForCausalLM.from_pretrained("HyperX-Sen/Zira-Z.1") inputs = tokenizer("Tum kya soch rahe ho about AI?", return_tensors="pt") outputs = model.generate(**inputs, max_new_tokens=50) print(tokenizer.decode(outputs[0], skip_special_tokens=True))' ``` --- ## 🧬 License & Contribution - 📜 **License**: Open for research & commercial use (see LICENSE) - 🤝 Contributions: Welcomed with open arms (and open pull requests) --- Made with ❤️, logic, and a lot of chai ☕