refactor: fix previous refactors

scripts/finetune.py

@@ -83,7 +83,7 @@ def do_inference(cfg, model, tokenizer, prompter="AlpacaPrompter"):
                 temperature=0.9,
                 top_p=0.95,
                 top_k=40,
-                return_DictDefault_in_generate=True,
+                return_dict_in_generate=True,
                 output_attentions=False,
                 output_hidden_states=False,
                 output_scores=False,

src/axolotl/utils/dict.py

@@ -6,4 +6,4 @@ class DictDefault(Dict):
     A Dict that returns None instead of returning empty Dict for missing keys.
     '''
     def __missing__(self, key):
-        return None
+        return None

src/axolotl/utils/models.py

@@ -184,9 +184,9 @@ def load_model(
         #     # https://github.com/HazyResearch/flash-attention/blob/40a25c8ee7465cf547b929cfa2937034e37bfce9/tests/models/test_gpt_neox.py#L12
         #     # https://github.com/HazyResearch/flash-attention/tree/main/training#model-components
         #     # add `**kwargs` to https://github.com/HazyResearch/flash-attention/blob/40a25c8ee7465cf547b929cfa2937034e37bfce9/flash_attn/models/gpt.py#L442
-        #     from flash_attn.utils.pretrained import state_DictDefault_from_pretrained
+        #     from flash_attn.utils.pretrained import state_dict_from_pretrained
         #     from flash_attn.models.gpt import GPTLMHeadModel
-        #     from flash_attn.models.gpt_neox import remap_state_DictDefault_hf_gpt_neox, gpt_neox_config_to_gpt2_config
+        #     from flash_attn.models.gpt_neox import remap_state_dict_hf_gpt_neox, gpt_neox_config_to_gpt2_config
         #     from transformers import GPTNeoXConfig
         #     config = gpt_neox_config_to_gpt2_config(GPTNeoXConfig.from_pretrained(base_model))
         #     config.use_flash_attn = True