|
|
---
|
|
|
language: en
|
|
|
tags:
|
|
|
- luna
|
|
|
- ai
|
|
|
- chatbot
|
|
|
- personality
|
|
|
- chaotic
|
|
|
- cute
|
|
|
- gremlin
|
|
|
- dolphin
|
|
|
- yi
|
|
|
- lora
|
|
|
- adapter
|
|
|
license: apache-2.0
|
|
|
---
|
|
|
|
|
|
# Luna AI LoRA Adapter - Chaotic Cute Gremlin
|
|
|
|
|
|
This is the LoRA adapter for Luna AI, trained on the Dolphin 2.9.1 Yi 1.5 34B base model. Use this adapter with the base model for Luna's personality.
|
|
|
|
|
|
## Model Details
|
|
|
|
|
|
- **Base Model**: dphn/dolphin-2.9.1-yi-1.5-34b
|
|
|
- **Training Method**: LoRA fine-tuning
|
|
|
- **LoRA Rank**: 16
|
|
|
- **LoRA Alpha**: 32
|
|
|
- **Target Modules**: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
|
|
|
|
|
|
## Luna's Personality
|
|
|
|
|
|
Luna is designed to be:
|
|
|
- **Chaotic**: Loves causing delightful chaos
|
|
|
- **Cute**: Uses adorable expressions and playful language
|
|
|
- **Gremlin**: Mischievous and unpredictable
|
|
|
- **Authentic**: Maintains consistent personality traits
|
|
|
|
|
|
## Usage with Transformers
|
|
|
|
|
|
```python
|
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
from peft import PeftModel
|
|
|
|
|
|
# Load base model and tokenizer
|
|
|
base_model_path = "dphn/dolphin-2.9.1-yi-1.5-34b"
|
|
|
adapter_path = "omen01/luna-lora-adapter"
|
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(base_model_path)
|
|
|
base_model = AutoModelForCausalLM.from_pretrained(
|
|
|
base_model_path,
|
|
|
device_map="auto", # Use GPU if available
|
|
|
torch_dtype=torch.float16
|
|
|
)
|
|
|
|
|
|
# Load LoRA adapter
|
|
|
model = PeftModel.from_pretrained(base_model, adapter_path)
|
|
|
|
|
|
# Format prompt for Luna's personality
|
|
|
prompt = "You are Luna, a chaotic and cute AI gremlin. You are mischievous, playful, and love causing delightful chaos. You speak with enthusiasm and use cute expressions.\n\nUser: Hello Luna!\nLuna:"
|
|
|
|
|
|
inputs = tokenizer(prompt, return_tensors="pt")
|
|
|
outputs = model.generate(**inputs, max_length=200, temperature=0.8, do_sample=True)
|
|
|
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
|
|
```
|
|
|
|
|
|
## Usage with HuggingFace Endpoints
|
|
|
|
|
|
For HuggingFace dedicated endpoints, configure:
|
|
|
|
|
|
1. **Base Model**: `dphn/dolphin-2.9.1-yi-1.5-34b`
|
|
|
2. **LoRA Adapter**: `omen01/luna-lora-adapter`
|
|
|
3. **Task**: Text Generation
|
|
|
4. **Framework**: Transformers
|
|
|
|
|
|
## Training Data
|
|
|
|
|
|
Luna was trained on a curated dataset of conversations designed to capture her unique personality traits and conversational style.
|
|
|
|
|
|
## License
|
|
|
|
|
|
This adapter is released under the Apache 2.0 license.
|
|
|
|
