Upload h2oai_pipeline.py
Browse files- h2oai_pipeline.py +42 -0
h2oai_pipeline.py
ADDED
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import TextGenerationPipeline
|
2 |
+
from transformers.pipelines.text_generation import ReturnType
|
3 |
+
|
4 |
+
STYLE = "<|prompt|>{instruction}<|endoftext|><|answer|>"
|
5 |
+
|
6 |
+
|
7 |
+
class H2OTextGenerationPipeline(TextGenerationPipeline):
|
8 |
+
def __init__(self, *args, **kwargs):
|
9 |
+
super().__init__(*args, **kwargs)
|
10 |
+
self.prompt = STYLE
|
11 |
+
|
12 |
+
def preprocess(
|
13 |
+
self, prompt_text, prefix="", handle_long_generation=None, **generate_kwargs
|
14 |
+
):
|
15 |
+
prompt_text = self.prompt.format(instruction=prompt_text)
|
16 |
+
return super().preprocess(
|
17 |
+
prompt_text,
|
18 |
+
prefix=prefix,
|
19 |
+
handle_long_generation=handle_long_generation,
|
20 |
+
**generate_kwargs,
|
21 |
+
)
|
22 |
+
|
23 |
+
def postprocess(
|
24 |
+
self,
|
25 |
+
model_outputs,
|
26 |
+
return_type=ReturnType.FULL_TEXT,
|
27 |
+
clean_up_tokenization_spaces=True,
|
28 |
+
):
|
29 |
+
records = super().postprocess(
|
30 |
+
model_outputs,
|
31 |
+
return_type=return_type,
|
32 |
+
clean_up_tokenization_spaces=clean_up_tokenization_spaces,
|
33 |
+
)
|
34 |
+
for rec in records:
|
35 |
+
rec["generated_text"] = (
|
36 |
+
rec["generated_text"]
|
37 |
+
.split("<|answer|>")[1]
|
38 |
+
.strip()
|
39 |
+
.split("<|prompt|>")[0]
|
40 |
+
.strip()
|
41 |
+
)
|
42 |
+
return records
|