tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

refactor: fix previous refactors

Browse Source
This commit is contained in:
NanoCode012
2023-05-28 22:25:42 +09:00
parent 8bd7a49cd7
commit 56f9ca5709
3 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
scripts/finetune.py
View File

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

2
src/axolotl/utils/dict.py
View File

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

4
src/axolotl/utils/models.py
View File

@@ -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/blob/40a25c8ee7465cf547b929cfa2937034e37bfce9/tests/models/test_gpt_neox.py#L12
# # https://github.com/HazyResearch/flash-attention/tree/main/training#model-components # # 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 # # 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 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 # from transformers import GPTNeoXConfig
# config = gpt_neox_config_to_gpt2_config(GPTNeoXConfig.from_pretrained(base_model)) # config = gpt_neox_config_to_gpt2_config(GPTNeoXConfig.from_pretrained(base_model))
# config.use_flash_attn = True # config.use_flash_attn = True