fix hard coding caused by rope dim calculation, load from config now

ktransformers/local_chat.py

@@ -78,7 +78,7 @@ def local_chat():
                 else:
                     content += line + "\n"
         if content == "":
-            if True: # config.prompt_file == None or config.prompt_file == "":
+            if config.prompt_file == None or config.prompt_file == "":
                 content = "hi"
             else:
                 content = open(config.prompt_file, "r").read()

ktransformers/models/modeling_deepseekv3.py

@@ -19,7 +19,7 @@ from transformers.generation import GenerationMixin
 from transformers.modeling_attn_mask_utils import AttentionMaskConverter
 # from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
 from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast, SequenceClassifierOutputWithPast
-from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS
+from ktransformers.util.modeling_rope_utils import ROPE_INIT_FUNCTIONS
 from transformers.modeling_utils import PreTrainedModel # ALL_ATTENTION_FUNCTIONS, PreTrainedModel
 # from transformers.processing_utils import Unpack
 from transformers.utils import (