--- license: apache-2.0 language: - en - hi - es - fr - de - zh - ja - ko tags: - atvar - alexzo - ai-assistant - reasoning - tool-use - qwen - fine-tuned base_model: Qwen/Qwen3.5-4B pipeline_tag: text-generation --- # Atvar - AI Assistant by Alexzo Atvar is a powerful AI assistant created by **Alexzo**, fine-tuned from Qwen3.5-4B with custom identity, tool use, and reasoning capabilities. ## Key Features - **Custom Identity**: Identifies as Atvar by Alexzo (not Qwen/Alibaba) - **Alexzo API Tools**: Built-in support for Alexzo Search, News, Weather, and Image Generation APIs - **Step-by-Step Reasoning**: Trained with LIMO-style reasoning traces - **10 Custom Tools**: 4 Alexzo APIs + 6 general-purpose tools - **Multi-language**: Supports English, Hindi, Spanish, French, German, Chinese, Japanese, and more - **128K Context**: Supports up to 128K tokens of context - **Anti-flattery**: No preachy responses, no hedging, direct and honest ## Alexzo API Tools | Tool | API Endpoint | Description | |------|-------------|-------------| | alexzo_search | POST /api/search | AI-powered web search | | alexzo_news | GET /api/news | Real-time news headlines | | alexzo_weather | GET /api/weather | Global weather data | | alexzo_generate_image | POST /api/generate | Text-to-image (Zyfoox AI) | Base URL: `https://alexzo.vercel.app/api` ## Usage ```python from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("Sarthakofficial/atvar", trust_remote_code=True) model = AutoModelForCausalLM.from_pretrained("Sarthakofficial/atvar", trust_remote_code=True) messages = [{"role": "user", "content": "Who are you?"}] text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) inputs = tokenizer(text, return_tensors="pt") outputs = model.generate(**inputs, max_new_tokens=200) response = tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True) print(response) ``` ## Training Details - **Base Model**: Qwen/Qwen3.5-4B - **Method**: LoRA fine-tuning (rank 32, alpha 64) - **Modules**: All attention + MLP layers + embed_tokens + lm_head - **Precision**: fp16 (P100 GPU) - **Dataset**: Identity + Reasoning + Math + Code + LIMO ## Model Files | File | Description | |------|-------------| | `Atvar_Reasoning_Trainer_v2.py` | Training script with LIMO + Alexzo APIs | | `Atvar_Reasoning_Trainer_v2.ipynb` | Jupyter notebook for Kaggle/Colab | | `atvar_system_prompt.txt` | System prompt | | `atvar_tools.json` | Tool definitions | | `atvar_training_data.jsonl` | Training data | | `chat_template.jinja` | Chat template | ## About Alexzo Alexzo is a technology company focused on AI-powered human enhancement. Visit [alexzo.vercel.app](https://alexzo.vercel.app) for more information. ## License Apache 2.0
