nabin2004/qwen-Manimator-1-grpo-merged
Full-weight merged release (bfloat16/float16 Safetensors) of qwen-Manimator-1-grpo-clean, fine-tuned from Qwen/Qwen3-8B for Manim Community Edition (ManimCE) animation generation.
- Base Model:
Qwen/Qwen3-8B - LoRA Adapter:
nabin2004/qwen-Manimator-1-grpo-clean - Merged Model Repo:
nabin2004/qwen-Manimator-1-grpo-merged - Quantized GGUF:
nabin2004/qwen-Manimator-1-grpo-GGUF
Direct vLLM Deployment
You can deploy this model directly in vLLM without needing dynamic LoRA adapter loading:
python3 -m vllm.entrypoints.openai.api_server \
--model nabin2004/qwen-Manimator-1-grpo-merged \
--host 0.0.0.0 \
--port 8000 \
--max-model-len 32768 \
--gpu-memory-utilization 0.90
Python / Transformers Quickstart
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "nabin2004/qwen-Manimator-1-grpo-merged"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16,
device_map="auto"
)
prompt = "<|im_start|>user\nWrite a ManimCE script animating a circle and a title 'GRPO'.<|im_end|>\n<|im_start|>assistant\n"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.2)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))