docs: add initial README for sakura-qwen3-0.6b-lora-demo-v1

README.md

@@ -14,4 +14,165 @@ tags:
 - lora
 ---
 
-# Sakura demo series based on Qwen3-0.6B
+# Sakura-Qwen3-0.6B-LoRA-Demo-v1
+
+[![Model License](https://img.shields.io/badge/Model%20License-Apache%202.0-blue.svg)](LICENSE)
+[![Base Model](https://img.shields.io/badge/Base%20Model-Qwen3__0.6B-green.svg)](https://huggingface.co/Qwen/Qwen3-0.6B)
+[![Adapter Type](https://img.shields.io/badge/Adapter-LoRA-orange.svg)](https://huggingface.co/docs/peft/en/index)
+
+## Model Description
+
+This is a LoRA adapter trained on **Qwen3-0.6B**, specifically optimized for VTuber role-playing in Chinese context. This model endows the AI with VTuber personality traits, speaking style, and character settings, suitable for VTuber interaction scenarios.
+
+## Model Details
+
+- **Base Model**: Qwen3-0.6B
+- **Adapter Type**: LoRA (Low-Rank Adaptation)
+- **Application**: VTuber role-playing
+- **Language**: Chinese
+- **Version**: Demo v1
+
+## Character Settings
+
+[Describe the VTuber character information in detail here, for example:]
+- **Character Name**: 小樱(Sakura)
+- **Personality Traits**: Energetic and cute, gentle and considerate, occasionally mischievous
+- **Speaking Style**: Uses specific speech patterns and emojis
+- **Background Story**: 
+- **Specialties**: Interacting with audience
+
+## Usage
+
+### Loading the Model
+
+```python
+from peft import PeftModel
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+# Load the Qwen3-0.6B chat model and tokenizer
+model_name = "Qwen/Qwen3-0.6B"
+adapter_name = "Boogon/sakura-qwen3-0.6b-lora-demo-v1"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype=torch.float16,
+    device_map="auto"
+)
+model = PeftModel.from_pretrained(model, adapter_name)
+```
+
+> Note: If you want to save the model into another directory, remember to use argument **cache_dir**.
+
+### Chat Example
+
+```python
+def chat_with_vtuber(messages, max_length=512):
+    """
+    messages format: [
+        {"role": "system", "content": ""},
+        {"role": "user", "content": "Hello!"},
+        {"role": "assistant", "content": "Hi there, my name's Sakura!(*´▽`*)"},
+        {"role": "user", "content": "What's new today?"}
+        # ...
+    ]
+    """
+    text = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True
+    )
+    
+    inputs = tokenizer(text, return_tensors="pt").to(model.device)
+    
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=max_length,
+            temperature=0.8,
+            top_p=0.9,
+            do_sample=True,
+            repetition_penalty=1.1,
+            eos_token_id=tokenizer.eos_token_id
+        )
+    
+    response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
+    return response
+
+# Example conversation
+messages = [
+    {"role": "system", "content": "You are Sakura-chan, a cute VTuber who loves interacting with fans. You speak in a cheerful, cute style with occasional Japanese phrases and emojis."},
+    {"role": "user", "content": "Hello Sakura-chan! How are you today?"}
+]
+
+response = chat_with_vtuber(messages)
+print(f"Sakura: {response}")
+```
+
+### Direct Chat Template Usage
+
+```python
+# Alternative method using chat template directly
+conversation = [
+    {"role": "user", "content": "What games do you like to play?"}
+]
+
+# Apply chat template
+text = tokenizer.apply_chat_template(
+    conversation,
+    tokenize=False,
+    add_generation_prompt=True
+)
+
+inputs = tokenizer(text, return_tensors="pt").to(model.device)
+
+outputs = model.generate(
+    **inputs,
+    max_new_tokens=256,
+    temperature=0.7,
+    do_sample=True
+)
+
+response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+print(response)
+```
+
+## Training Information
+
+- **Training Data**: n/a
+- **Training Objective**: Learn VTuber's dialogue style, personality traits, and interaction patterns
+- **LoRA Configuration**:
+  - r: 16
+  - lora_alpha: 32
+  - target_modules: ["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"]
+  - lora_dropout: 0.05
+- **Training Framework**: PEFT
+
+## Features
+
+-  Character-consistent dialogue
+-  Emotional expression with emojis
+-  Context-adaptive responses
+-  Fan interaction simulation
+-  Multi-turn conversation support
+
+## License
+
+- Base Model: Apache 2.0
+- Adapter: Apache 2.0
+
+## Important Notes
+
+**Important Notes**:
+- This is a demo version and may have unstable responses
+- Not for commercial use
+- Character dialogue content is fictional and unrelated to real persons
+- The model is optimized for Chinese VTuber role-playing scenarios
+
+## Contributing & Feedback
+
+Welcome to submit feedback or suggestions through Issues!
+
+Anything related can be sent to [Sakura-Adapters](https://github.com/BoogonAgora/Sakura-Adapters)
+