Rename precious3_gpt_multi_modalX.py to precious3_gpt_multi_modal.py

precious3_gpt_multi_modalX.py → precious3_gpt_multi_modal.py

@@ -13,12 +13,12 @@ from transformers import PreTrainedTokenizerFast
 import os
 import torch.nn.functional as F
 
-from modeling_mpt import MPTModel, MPTForCausalLM, gen_attention_mask_in_length
-from configuration_mpt import MPTConfig
-from blocks import MPTBlock
-from norm import NORM_CLASS_REGISTRY
-from custom_embedding import SharedEmbedding
-from attention import ATTN_CLASS_REGISTRY, attn_bias_shape, build_attn_bias, gen_slopes
+from mpt_7b.modeling_mpt import MPTModel, MPTForCausalLM, gen_attention_mask_in_length
+from mpt_7b.configuration_mpt import MPTConfig
+from mpt_7b.blocks import MPTBlock
+from mpt_7b.norm import NORM_CLASS_REGISTRY
+from mpt_7b.custom_embedding import SharedEmbedding
+from mpt_7b.attention import ATTN_CLASS_REGISTRY, attn_bias_shape, build_attn_bias, gen_slopes
 
 import logging
 log = logging.getLogger(__name__)
@@ -85,10 +85,10 @@ class Custom_MptModel(MPTModel): # MptModel
         
 
         self.modality2_embedding_projection = nn.ModuleList([nn.Linear(modality2_dim, config.d_model), 
-                                                             # nn.BatchNorm1d(config.d_model),
+                                                             # nn.BatchNorm1d(config.d_model),
                                                              nn.ReLU(),
                                                              nn.Linear(config.d_model, config.d_model),
-                                                             # nn.BatchNorm1d(config.d_model),
+                                                             # nn.BatchNorm1d(config.d_model),
                                                              nn.ReLU(),
                                                              nn.Linear(config.d_model, config.d_model)])# nn.Linear(modality0_dim, self.hidden_size)
         
@@ -351,4 +351,4 @@ class Custom_MPTForCausalLM(MPTForCausalLM):
             _labels = torch.roll(labels, shifts=-1)
             _labels[:, -1] = -100
             loss = F.cross_entropy(logits.view(-1, logits.size(-1)), _labels.to(logits.device).view(-1))
-        return CausalLMOutputWithPast(loss=loss, logits=logits, past_key_values=outputs.past_key_values, hidden_states=outputs.hidden_states, attentions=outputs.attentions)
+        return CausalLMOutputWithPast(loss=loss, logits=logits, past_key_values=outputs.past_key_values, hidden_states=outputs.hidden_states, attentions=outputs.attentions)