import torch import gradio as gr import spaces from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer from transformers.models.qwen2 import Qwen2Config, Qwen2ForCausalLM # ========================================== # LOSSLESS RUNTIME ARCHITECTURE BINDING # ========================================== class LogicL1Config(Qwen2Config): model_type = "logic_l1" class LogicL1ForCausalLM(Qwen2ForCausalLM): config_class = LogicL1Config # Register native architecture cleanly so transformers maps weights perfectly AutoConfig.register("logic_l1", LogicL1Config) AutoModelForCausalLM.register(LogicL1Config, LogicL1ForCausalLM) MODEL_ID = "VADITYA307/Logic-L1" print("Loading AROM Logic L1 with lossless weight mapping...") tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=False) model = AutoModelForCausalLM.from_pretrained( MODEL_ID, torch_dtype=torch.float16, trust_remote_code=False ) @spaces.GPU def predict(message, history): model.to("cuda") messages = [] for user_msg, bot_msg in history: messages.append({"role": "user", "content": user_msg}) messages.append({"role": "assistant", "content": bot_msg}) messages.append({"role": "user", "content": message}) prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) inputs = tokenizer(prompt, return_tensors="pt").to("cuda") with torch.no_grad(): outputs = model.generate( **inputs, max_new_tokens=512, temperature=0.3, top_p=0.95, repetition_penalty=1.15, do_sample=True, pad_token_id=tokenizer.eos_token_id ) response = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True) return response.replace("<|im_end|>", "").strip() demo = gr.ChatInterface( fn=predict, title="AROM Logic L1", description="State-of-the-art proprietary artificial intelligence engineered exclusively by Aditya 'Aadi' at AROM Labs India." ) if __name__ == "__main__": demo.launch()