Spaces:
Running
Running
Update apis/chat_api.py
Browse files- apis/chat_api.py +73 -6
apis/chat_api.py
CHANGED
@@ -1,7 +1,14 @@
|
|
1 |
import argparse
|
2 |
import uvicorn
|
3 |
import sys
|
|
|
|
|
|
|
|
|
4 |
import json
|
|
|
|
|
|
|
5 |
import string
|
6 |
import random
|
7 |
import base64
|
@@ -31,12 +38,12 @@ class ChatAPIApp:
|
|
31 |
)
|
32 |
self.setup_routes()
|
33 |
|
34 |
-
def
|
35 |
f = open('apis/lang_name.json', "r")
|
36 |
self.available_models = json.loads(f.read())
|
37 |
return self.available_models
|
38 |
|
39 |
-
class
|
40 |
from_language: str = Field(
|
41 |
default="auto",
|
42 |
description="(str) `Detect`",
|
@@ -51,7 +58,7 @@ class ChatAPIApp:
|
|
51 |
)
|
52 |
|
53 |
|
54 |
-
def
|
55 |
translator = Translator()
|
56 |
f = open('apis/lang_name.json', "r")
|
57 |
available_langs = json.loads(f.read())
|
@@ -73,6 +80,60 @@ class ChatAPIApp:
|
|
73 |
json_compatible_item_data = jsonable_encoder(item_response)
|
74 |
return JSONResponse(content=json_compatible_item_data)
|
75 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
76 |
|
77 |
class DetectLanguagePostItem(BaseModel):
|
78 |
input_text: str = Field(
|
@@ -125,15 +186,21 @@ class ChatAPIApp:
|
|
125 |
def setup_routes(self):
|
126 |
for prefix in ["", "/v1"]:
|
127 |
self.app.get(
|
128 |
-
prefix + "/
|
129 |
summary="Get available languages",
|
130 |
-
)(self.
|
131 |
|
132 |
self.app.post(
|
133 |
prefix + "/translate",
|
134 |
summary="translate text",
|
135 |
-
)(self.
|
|
|
|
|
|
|
|
|
|
|
136 |
|
|
|
137 |
self.app.post(
|
138 |
prefix + "/detect",
|
139 |
summary="detect language",
|
|
|
1 |
import argparse
|
2 |
import uvicorn
|
3 |
import sys
|
4 |
+
import os
|
5 |
+
import io
|
6 |
+
from transformers import M2M100Tokenizer, M2M100ForConditionalGeneration
|
7 |
+
import time
|
8 |
import json
|
9 |
+
from typing import List
|
10 |
+
import torch
|
11 |
+
import logging
|
12 |
import string
|
13 |
import random
|
14 |
import base64
|
|
|
38 |
)
|
39 |
self.setup_routes()
|
40 |
|
41 |
+
def get_available_langs(self):
|
42 |
f = open('apis/lang_name.json', "r")
|
43 |
self.available_models = json.loads(f.read())
|
44 |
return self.available_models
|
45 |
|
46 |
+
class TranslateCompletionsPostItem(BaseModel):
|
47 |
from_language: str = Field(
|
48 |
default="auto",
|
49 |
description="(str) `Detect`",
|
|
|
58 |
)
|
59 |
|
60 |
|
61 |
+
def translate_completions(self, item: TranslateCompletionsPostItem):
|
62 |
translator = Translator()
|
63 |
f = open('apis/lang_name.json', "r")
|
64 |
available_langs = json.loads(f.read())
|
|
|
80 |
json_compatible_item_data = jsonable_encoder(item_response)
|
81 |
return JSONResponse(content=json_compatible_item_data)
|
82 |
|
83 |
+
def translate_ai_completions(self, item: TranslateCompletionsPostItem):
|
84 |
+
translator = Translator()
|
85 |
+
f = open('apis/lang_name.json', "r")
|
86 |
+
available_langs = json.loads(f.read())
|
87 |
+
from_lang = 'en'
|
88 |
+
to_lang = 'en'
|
89 |
+
for lang_item in available_langs:
|
90 |
+
if item.to_language == lang_item['code']:
|
91 |
+
to_lang = item.to_language
|
92 |
+
if item.from_language == lang_item['code']:
|
93 |
+
from_lang = item.from_language
|
94 |
+
|
95 |
+
if to_lang == 'auto':
|
96 |
+
to_lang = 'en'
|
97 |
+
|
98 |
+
if from_lang == 'auto':
|
99 |
+
from_lang = translator.detect(item.input_text).lang
|
100 |
+
|
101 |
+
if torch.cuda.is_available():
|
102 |
+
device = torch.device("cuda:0")
|
103 |
+
else:
|
104 |
+
device = torch.device("cpu")
|
105 |
+
logging.warning("GPU not found, using CPU, translation will be very slow.")
|
106 |
+
|
107 |
+
time_start = time.time()
|
108 |
+
|
109 |
+
tokenizer = M2M100Tokenizer.from_pretrained(pretrained_model, cache_dir=cache_dir)
|
110 |
+
model = M2M100ForConditionalGeneration.from_pretrained(
|
111 |
+
"facebook/m2m100_1.2B", cache_dir="models/"
|
112 |
+
).to(device)
|
113 |
+
model.eval()
|
114 |
+
|
115 |
+
tokenizer.src_lang = from_lang
|
116 |
+
with torch.no_grad():
|
117 |
+
encoded_input = tokenizer(item.input_text, return_tensors="pt").to(device)
|
118 |
+
generated_tokens = model.generate(
|
119 |
+
**encoded_input, forced_bos_token_id=tokenizer.get_lang_id(to_lang)
|
120 |
+
)
|
121 |
+
translated_text = tokenizer.batch_decode(
|
122 |
+
generated_tokens, skip_special_tokens=True
|
123 |
+
)[0]
|
124 |
+
|
125 |
+
time_end = time.time()
|
126 |
+
translated = translated_text
|
127 |
+
item_response = {
|
128 |
+
"from_language": from_lang,
|
129 |
+
"to_language": to_lang,
|
130 |
+
"text": item.input_text,
|
131 |
+
"translate": translated,
|
132 |
+
"start": str(time_start),
|
133 |
+
"end": str(time_end)
|
134 |
+
}
|
135 |
+
json_compatible_item_data = jsonable_encoder(item_response)
|
136 |
+
return JSONResponse(content=json_compatible_item_data)
|
137 |
|
138 |
class DetectLanguagePostItem(BaseModel):
|
139 |
input_text: str = Field(
|
|
|
186 |
def setup_routes(self):
|
187 |
for prefix in ["", "/v1"]:
|
188 |
self.app.get(
|
189 |
+
prefix + "/langs",
|
190 |
summary="Get available languages",
|
191 |
+
)(self.get_available_langs)
|
192 |
|
193 |
self.app.post(
|
194 |
prefix + "/translate",
|
195 |
summary="translate text",
|
196 |
+
)(self.translate_completions)
|
197 |
+
|
198 |
+
self.app.post(
|
199 |
+
prefix + "/translate/ai",
|
200 |
+
summary="translate text with ai",
|
201 |
+
)(self.translate_ai_completions)
|
202 |
|
203 |
+
|
204 |
self.app.post(
|
205 |
prefix + "/detect",
|
206 |
summary="detect language",
|