Using HunyuanWorld-1 Model

how do I use : https://huggingface.co/tencent/HunyuanWorld-1

The model you're referring to — [`tencent/HunyuanWorld-1`](https://huggingface.co/tencent/HunyuanWorld-1) — is a large-scale language model developed by Tencent, and hosted on Hugging Face. Here's how you can **use it locally or via API**, depending on your goal: --- ## ✅ 1. **Check Model Type and Compatibility** The [Hugging Face page](https://huggingface.co/tencent/HunyuanWorld-1) shows that it is a **transformer-based causal language model**, similar to LLaMA/GPT-style models. It currently **only supports text generation**, and its base model (`HunyuanWorld-1-7B`) is a **7B parameter model** released under the Apache 2.0 license. --- ## ✅ 2. **Using the Model on Hugging Face (API Access)** If you want to try it **without installing anything**, you can use the **Hugging Face Inference API**: ```python from transformers import AutoTokenizer, AutoModelForCausalLM import torch # Load tokenizer and model tokenizer = AutoTokenizer.from_pretrained("tencent/HunyuanWorld-1") model = AutoModelForCausalLM.from_pretrained("tencent/HunyuanWorld-1", torch_dtype=torch.float16).cuda() # Input prompt prompt = "Once upon a time in a future world where AI ruled the cities," # Tokenize inputs = tokenizer(prompt, return_tensors="pt").to("cuda") # Generate output = model.generate(**inputs, max_new_tokens=100) print(tokenizer.decode(output[0], skip_special_tokens=True)) ``` > ⚠️ Make sure you have a GPU and PyTorch installed with `torch.cuda.is_available()`. --- ## ✅ 3. **Run with Hugging Face Transformers + `transformers` CLI** You can also use the Hugging Face CLI: ```bash pip install transformers accelerate transformers-cli login # if needed ``` Then run in terminal: ```bash python -m transformers.commands.text-generation \ --model tencent/HunyuanWorld-1 \ --prompt "Tell me about the future of robotics." \ --max_new_tokens 200 ``` --- ## ✅ 4. **Use on Hugging Face Spaces (No Code)** Go to this Space (if provided): Sometimes, Tencent hosts a Hugging Face [Space](https://huggingface.co/spaces) where you can **interact with the model in a browser**. If it's not listed, you can spin up your own using Gradio. --- ## ✅ 5. **Run it Locally with Ollama (If GGUF becomes available)** Currently, there's **no GGUF or GPTQ quantized version** released for local use via Ollama or LM Studio. If they release a `gguf` file on [huggingface.co](https://huggingface.co/) under `files`, you'll see something like: ``` tencent/HunyuanWorld-1-7B.Q4_K_M.gguf ``` Then you can run: ```bash ollama create hunyuan -f <modelfile.gguf> ollama run hunyuan ``` --- ## ✅ 6. **Model Card and Limits** - Model: 7B parameters - Context length: ~2k–4k tokens - License: Apache 2.0 (can be used commercially) - Language support: Primarily Chinese and English (bilingual) - Architecture: Transformer decoder-only (GPT-style) --- ## ✅ Tips - 🔋 You need at least **16GB VRAM** to run it unquantized (fp16). - 💾 For inference on CPU, try using **`torch_dtype=torch.bfloat16` or `int8` quantization** with `bitsandbytes`. --- ## Want to Try it Without Code? Use this: 👉 https://huggingface.co/tencent/HunyuanWorld-1 Click **“Use in Transformers”** or **“Open in Spaces”** (if available). --- If you’d like, I can help you: - Quantize the model to run on lower VRAM - Set it up with Gradio for UI - Benchmark it vs other models (e.g., Qwen, Mistral) Just say the word.