Update README.md
Browse files
README.md
CHANGED
@@ -88,7 +88,7 @@ Researchers should apply responsible AI best practices, including mapping, measu
|
|
88 |
### Loading the model locally
|
89 |
|
90 |
import torch
|
91 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
92 |
|
93 |
torch.random.manual_seed(0)
|
94 |
|
@@ -114,11 +114,20 @@ Researchers should apply responsible AI best practices, including mapping, measu
|
|
114 |
tokenizer=tokenizer,
|
115 |
)
|
116 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
generation_args = {
|
118 |
"max_new_tokens": 500,
|
119 |
"return_full_text": False,
|
120 |
"temperature": 0.0,
|
121 |
"do_sample": False,
|
|
|
122 |
}
|
123 |
|
124 |
output = pipe(messages, **generation_args)
|
|
|
88 |
### Loading the model locally
|
89 |
|
90 |
import torch
|
91 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, StoppingCriteria
|
92 |
|
93 |
torch.random.manual_seed(0)
|
94 |
|
|
|
114 |
tokenizer=tokenizer,
|
115 |
)
|
116 |
|
117 |
+
class EosListStoppingCriteria(StoppingCriteria):
|
118 |
+
def __init__(self, eos_sequence = [32007]):
|
119 |
+
self.eos_sequence = eos_sequence
|
120 |
+
|
121 |
+
def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
|
122 |
+
last_ids = input_ids[:,-len(self.eos_sequence):].tolist()
|
123 |
+
return self.eos_sequence in last_ids
|
124 |
+
|
125 |
generation_args = {
|
126 |
"max_new_tokens": 500,
|
127 |
"return_full_text": False,
|
128 |
"temperature": 0.0,
|
129 |
"do_sample": False,
|
130 |
+
"generation_kwargs": {"stopping_criteria": EosListStoppingCriteria()}
|
131 |
}
|
132 |
|
133 |
output = pipe(messages, **generation_args)
|