Spaces:
Running
Running
Add a temperature slider control in the advanced options and adjust the default temperature value for the translation function
Browse files
app.py
CHANGED
|
@@ -65,11 +65,14 @@ with gr.Blocks(
|
|
| 65 |
interactive=False
|
| 66 |
)
|
| 67 |
|
|
|
|
|
|
|
|
|
|
| 68 |
translate_btn = gr.Button("Translate ✨")
|
| 69 |
|
| 70 |
translate_btn.click(
|
| 71 |
run,
|
| 72 |
-
inputs=[input_text, target_language],
|
| 73 |
outputs=output_text
|
| 74 |
)
|
| 75 |
|
|
|
|
| 65 |
interactive=False
|
| 66 |
)
|
| 67 |
|
| 68 |
+
with gr.Accordion("Advanced Options", open=False):
|
| 69 |
+
temp = gr.Slider(0.00, 1.0, step=.05, label="Temperature", value=0.3)
|
| 70 |
+
|
| 71 |
translate_btn = gr.Button("Translate ✨")
|
| 72 |
|
| 73 |
translate_btn.click(
|
| 74 |
run,
|
| 75 |
+
inputs=[input_text, target_language, temp],
|
| 76 |
outputs=output_text
|
| 77 |
)
|
| 78 |
|
model.py
CHANGED
|
@@ -11,6 +11,7 @@ model = AutoModelForCausalLM.from_pretrained(model_path)
|
|
| 11 |
def run(
|
| 12 |
text: str = "It’s on the house.",
|
| 13 |
target_language: str = "Portuguese",
|
|
|
|
| 14 |
):
|
| 15 |
messages = [
|
| 16 |
{
|
|
@@ -27,7 +28,13 @@ def run(
|
|
| 27 |
input_ids = tokenized_chat.to(model.device)
|
| 28 |
input_length = input_ids.shape[1]
|
| 29 |
|
| 30 |
-
outputs = model.generate(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
# 2. Fatiamos o tensor: pegamos do [input_length:] até o fim
|
| 32 |
# Isso isola apenas os tokens novos gerados
|
| 33 |
generated_tokens = outputs[0][input_length:]
|
|
|
|
| 11 |
def run(
|
| 12 |
text: str = "It’s on the house.",
|
| 13 |
target_language: str = "Portuguese",
|
| 14 |
+
temperature: float = 0.3,
|
| 15 |
):
|
| 16 |
messages = [
|
| 17 |
{
|
|
|
|
| 28 |
input_ids = tokenized_chat.to(model.device)
|
| 29 |
input_length = input_ids.shape[1]
|
| 30 |
|
| 31 |
+
outputs = model.generate(
|
| 32 |
+
input_ids,
|
| 33 |
+
max_new_tokens=2048,
|
| 34 |
+
do_sample=True,
|
| 35 |
+
temperature=temperature,
|
| 36 |
+
top_p=0.9
|
| 37 |
+
)
|
| 38 |
# 2. Fatiamos o tensor: pegamos do [input_length:] até o fim
|
| 39 |
# Isso isola apenas os tokens novos gerados
|
| 40 |
generated_tokens = outputs[0][input_length:]
|