efort
/
glm4
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

process eval issue

This commit is contained in:
zhipuch 2024-11-11 17:33:43 +08:00
parent a0c568753a
commit a7dbfa5ada
2 changed files with 7 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
finetune_demo/finetune.py
View File

@ -348,7 +348,7 @@ def load_tokenizer_and_model(
model_dir: str, model_dir: str,
peft_config: Optional[PeftConfig] = None, peft_config: Optional[PeftConfig] = None,
): ):
tokenizer = AutoTokenizer.from_pretrained(model_dir, trust_remote_code=True) tokenizer = AutoTokenizer.from_pretrained(model_dir, padding_side='left', trust_remote_code=True)
if peft_config is not None: if peft_config is not None:
model = AutoModelForCausalLM.from_pretrained( model = AutoModelForCausalLM.from_pretrained(
model_dir, model_dir,
@ -370,6 +370,8 @@ def load_tokenizer_and_model(
def compute_metrics(eval_preds: EvalPrediction, tokenizer): def compute_metrics(eval_preds: EvalPrediction, tokenizer):
batched_pred_ids, batched_label_ids = eval_preds batched_pred_ids, batched_label_ids = eval_preds
batched_pred_ids[batched_pred_ids==-100] = tokenizer.pad_token_id
batched_label_ids[batched_label_ids==-100] = tokenizer.pad_token_id
metrics_dct = {'rouge-1': [], 'rouge-2': [], 'rouge-l': [], 'bleu-4': []} metrics_dct = {'rouge-1': [], 'rouge-2': [], 'rouge-l': [], 'bleu-4': []}
for pred_ids, label_ids in zip(batched_pred_ids, batched_label_ids): for pred_ids, label_ids in zip(batched_pred_ids, batched_label_ids):
pred_txt = tokenizer.decode(pred_ids).strip() pred_txt = tokenizer.decode(pred_ids).strip()

4
finetune_demo/finetune_vision.py
View File

@ -394,7 +394,7 @@ def load_tokenizer_and_model(
model_dir: str, model_dir: str,
peft_config: Optional[PeftConfig] = None, peft_config: Optional[PeftConfig] = None,
): ):
tokenizer = AutoTokenizer.from_pretrained(model_dir, trust_remote_code=True) tokenizer = AutoTokenizer.from_pretrained(model_dir, padding_side='left', trust_remote_code=True)
if peft_config is not None: if peft_config is not None:
model = AutoModelForCausalLM.from_pretrained( model = AutoModelForCausalLM.from_pretrained(
model_dir, model_dir,
@ -416,6 +416,8 @@ def load_tokenizer_and_model(
def compute_metrics(eval_preds: EvalPrediction, tokenizer): def compute_metrics(eval_preds: EvalPrediction, tokenizer):
batched_pred_ids, batched_label_ids = eval_preds batched_pred_ids, batched_label_ids = eval_preds
batched_pred_ids[batched_pred_ids==-100] = tokenizer.pad_token_id
batched_label_ids[batched_label_ids==-100] = tokenizer.pad_token_id
metrics_dct = {'rouge-1': [], 'rouge-2': [], 'rouge-l': [], 'bleu-4': []} metrics_dct = {'rouge-1': [], 'rouge-2': [], 'rouge-l': [], 'bleu-4': []}
for pred_ids, label_ids in zip(batched_pred_ids, batched_label_ids): for pred_ids, label_ids in zip(batched_pred_ids, batched_label_ids):
pred_txt = tokenizer.decode(pred_ids).strip() pred_txt = tokenizer.decode(pred_ids).strip()