README.md

# DippyResearch/super-cot-v1-step-20

This is a fine-tuned model for roleplay tasks, created by merging a LoRA adapter with the base model.

## Base Model
- Model: Cydonia-24B-v4.1
- Path: /workspace/models/Cydonia-24B-v4.1

## Training Details
- Framework: VERL (GRPO)
- Training Type: Roleplay fine-tuning
- Checkpoint: actor
- LoRA Adapter: DippyResearch/super-cot-v1-step-20-lora

## Usage

### With Transformers
```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("DippyResearch/super-cot-v1-step-20")
tokenizer = AutoTokenizer.from_pretrained("DippyResearch/super-cot-v1-step-20")
```

### With vLLM
```python
from vllm import LLM

llm = LLM(model="DippyResearch/super-cot-v1-step-20")
```

## Alternative: Use LoRA Adapter
For more flexibility, you can use the LoRA adapter directly:
- LoRA Adapter: DippyResearch/super-cot-v1-step-20-lora
- Base Model: Cydonia-24B-v4.1