grid-runner-7b

Fine-tuned Qwen2.5-7B-Instruct for SYSBREAK cyberpunk MMO content generation.

Model Description

This model is a LoRA-merged version of Qwen2.5-7B-Instruct, fine-tuned to generate structured JSON content for the SYSBREAK game.

Purpose: You are a cyberpunk world event writer for SYSBREAK. Generate world events in JSON format. Use ONLY entities from the provided world context. Do NOT include specific credit/XP reward numbers. Respond with valid JSON only.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("grid-runner-7b")
tokenizer = AutoTokenizer.from_pretrained("grid-runner-7b")

messages = [
    {"role": "system", "content": "You are a cyberpunk world event writer for SYSBREAK. Generate world events in JSON format. Use ONLY entities from the provided world context. Do NOT include specific credit/XP reward numbers. Respond with valid JSON only."},
    {"role": "user", "content": "Your prompt here"},
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=1024, temperature=0.9, top_p=0.9)
print(tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True))

Ollama

ollama run grid-runner-7b

Training Details

• Training examples: 300
• Training duration: 12.5 minutes
• Base model: Qwen/Qwen2.5-7B-Instruct
• LoRA rank: 32
• LoRA alpha: 64
• Learning rate: 2e-4
• Epochs: 3
• Quantization: QLoRA 4-bit NF4
• Compute dtype: BF16

License

Apache 2.0 (same as base model)