Upload inference.py

inference.py

@@ -5,9 +5,7 @@ import re
 import difflib
 from utils import *
 from config import *
-from transformers import GPT2Config, BitsAndBytesConfig
-from bitsandbytes.nn import Linear8bitLt  
-from bitsandbytes.optim import GlobalOptimManager  
+from transformers import GPT2Config
 from abctoolkit.utils import Exclaim_re, Quote_re, SquareBracket_re, Barline_regexPattern
 from abctoolkit.transpose import Note_list, Pitch_sign_list
 from abctoolkit.duration import calculate_bartext_duration
@@ -42,13 +40,6 @@ byte_config = GPT2Config(num_hidden_layers=CHAR_NUM_LAYERS,
                          num_attention_heads=HIDDEN_SIZE // 64,
                          vocab_size=128)
 
-quantization_config = BitsAndBytesConfig(
-    load_in_8bit=True,
-    llm_int8_skip_modules=["patch_embedding"],
-    bnb_4bit_use_double_quant=True  # 双重量化进一步压缩
-)
-
-
 model = NotaGenLMHeadModel(encoder_config=patch_config, decoder_config=byte_config).to(device)
 
 def download_model_weights():