Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -11,18 +11,18 @@ from huggingface_hub import hf_hub_download
|
|
| 11 |
api = HfApi()
|
| 12 |
app = FastAPI()
|
| 13 |
|
| 14 |
-
|
| 15 |
DATASET_REPO = "mitchellkil/gaelic-app-feedback" # file to write feedback to
|
| 16 |
|
| 17 |
-
HF_TOKEN = os.environ["HF_TOKEN"] # set in
|
| 18 |
|
| 19 |
|
| 20 |
-
print("Loading
|
| 21 |
|
| 22 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 23 |
|
| 24 |
-
tokenizer = AutoTokenizer.from_pretrained(
|
| 25 |
-
model = AutoModelForSeq2SeqLM.from_pretrained(
|
| 26 |
model.eval()
|
| 27 |
|
| 28 |
print("Model loaded")
|
|
@@ -33,13 +33,13 @@ class TextRequest(BaseModel):
|
|
| 33 |
|
| 34 |
def text_to_ipa(text: str) -> str:
|
| 35 |
# Input text - > target text for trained model
|
| 36 |
-
prompt = f"convert
|
| 37 |
|
| 38 |
inputs = tokenizer(
|
| 39 |
prompt,
|
| 40 |
return_tensors="pt",
|
| 41 |
truncation=True,
|
| 42 |
-
max_length=
|
| 43 |
).to(device)
|
| 44 |
|
| 45 |
with torch.no_grad():
|
|
@@ -83,10 +83,9 @@ def save_feedback(data: Feedback):
|
|
| 83 |
)
|
| 84 |
with open(local_file, "r", encoding="utf-8") as f:
|
| 85 |
lines = f.readlines()
|
| 86 |
-
|
| 87 |
-
lines = []
|
| 88 |
|
| 89 |
-
lines.append(json.dumps(entry) + "\n") #
|
| 90 |
|
| 91 |
temp_path = "feedback_temp.jsonl" # create temp
|
| 92 |
with open(temp_path, "w", encoding="utf-8") as f:
|
|
@@ -98,8 +97,8 @@ def save_feedback(data: Feedback):
|
|
| 98 |
repo_id=DATASET_REPO,
|
| 99 |
repo_type="dataset",
|
| 100 |
token=HF_TOKEN,
|
| 101 |
-
commit_message="
|
| 102 |
)
|
| 103 |
|
| 104 |
-
return {"status": "
|
| 105 |
|
|
|
|
| 11 |
api = HfApi()
|
| 12 |
app = FastAPI()
|
| 13 |
|
| 14 |
+
modelName = "MitchellKil/gaelic-ipa-byt5" # my fine tuned model
|
| 15 |
DATASET_REPO = "mitchellkil/gaelic-app-feedback" # file to write feedback to
|
| 16 |
|
| 17 |
+
HF_TOKEN = os.environ["HF_TOKEN"] # set in secrets
|
| 18 |
|
| 19 |
|
| 20 |
+
print("Loading IPA model")
|
| 21 |
|
| 22 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 23 |
|
| 24 |
+
tokenizer = AutoTokenizer.from_pretrained(modelName)
|
| 25 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(modelName).to(device)
|
| 26 |
model.eval()
|
| 27 |
|
| 28 |
print("Model loaded")
|
|
|
|
| 33 |
|
| 34 |
def text_to_ipa(text: str) -> str:
|
| 35 |
# Input text - > target text for trained model
|
| 36 |
+
prompt = f"convert Gaelic to IPA: {text}"
|
| 37 |
|
| 38 |
inputs = tokenizer(
|
| 39 |
prompt,
|
| 40 |
return_tensors="pt",
|
| 41 |
truncation=True,
|
| 42 |
+
max_length=110
|
| 43 |
).to(device)
|
| 44 |
|
| 45 |
with torch.no_grad():
|
|
|
|
| 83 |
)
|
| 84 |
with open(local_file, "r", encoding="utf-8") as f:
|
| 85 |
lines = f.readlines()
|
| 86 |
+
|
|
|
|
| 87 |
|
| 88 |
+
lines.append(json.dumps(entry) + "\n") # append
|
| 89 |
|
| 90 |
temp_path = "feedback_temp.jsonl" # create temp
|
| 91 |
with open(temp_path, "w", encoding="utf-8") as f:
|
|
|
|
| 97 |
repo_id=DATASET_REPO,
|
| 98 |
repo_type="dataset",
|
| 99 |
token=HF_TOKEN,
|
| 100 |
+
commit_message="New feedback added"
|
| 101 |
)
|
| 102 |
|
| 103 |
+
return {"status": "Thanks for your feedback"}
|
| 104 |
|