def get_states(model, session_state=None):
if session_state:
setattr(session_state, "max_length", model.args.max_length)
setattr(session_state, "decoding_algorithm", model.args.do_sample)
setattr(session_state, "length_penalty", model.args.length_penalty)
setattr(session_state, "num_beams", model.args.num_beams)
setattr(session_state, "early_stopping", model.args.early_stopping)
setattr(session_state, "top_k", model.args.top_k)
setattr(session_state, "top_p", model.args.top_p)
else:
session_state = get(
max_seq_length=model.args.max_seq_length,
max_length=model.args.max_length,
decoding_algorithm="Sampling" if model.args.do_sample else "Beam Search",
length_penalty=model.args.length_penalty,
early_stopping=model.args.early_stopping,
num_beams=model.args.num_beams,
top_k=model.args.top_k,
top_p=model.args.top_p,
)
model.args.max_seq_length = session_state.max_seq_length
model.args.max_length = session_state.max_length
model.args.length_penalty = session_state.length_penalty
model.args.early_stopping = session_state.early_stopping
model.args.top_k = session_state.top_k
model.args.top_p = session_state.top_p
if session_state.decoding_algorithm == "Sampling":
model.args.do_sample = True
model.args.num_beams = None
elif session_state.decoding_algorithm == "Beam Search":
model.args.do_sample = False
model.args.num_beams = session_state.num_beams
return session_state, model
def ner_viewer(model):
session_state = get(max_seq_length=model.args.max_seq_length, )
model.args.max_seq_length = session_state.max_seq_length
entity_list = model.args.labels_list
st.sidebar.subheader("Entities")
entity_checkboxes = {
entity: st.sidebar.checkbox(entity, value=True)
for entity in entity_list
}
entity_color_map = {
entity: get_color(i)
for i, entity in enumerate(entity_list)
}
st.sidebar.subheader("Parameters")
model.args.max_seq_length = st.sidebar.slider(
"Max Seq Length",
min_value=1,
max_value=512,
value=model.args.max_seq_length)
st.subheader("Enter text: ")
input_text = st.text_area("")
prediction = get_prediction(model, input_text)[0]
to_write = " ".join([
format_word(word, entity, entity_checkboxes, entity_color_map)
for pred in prediction for word, entity in pred.items()
])
st.subheader(f"Predictions")
st.write(to_write, unsafe_allow_html=True)
def qa_viewer(model):
st.sidebar.subheader("Parameters")
try:
session_state, model = get_states(model)
except AttributeError:
session_state = get(
max_seq_length=model.args.max_seq_length,
max_answer_length=model.args.max_answer_length,
max_query_length=model.args.max_query_length,
)
session_state, model = get_states(model, session_state)
model.args.max_seq_length = st.sidebar.slider(
"Max Seq Length", min_value=1, max_value=512, value=model.args.max_seq_length
)
model.args.max_answer_length = st.sidebar.slider(
"Max Answer Length", min_value=1, max_value=512, value=model.args.max_answer_length
)
model.args.max_query_length = st.sidebar.slider(
"Max Query Length", min_value=1, max_value=512, value=model.args.max_query_length
)
model.args.n_best_size = st.sidebar.slider("Number of answers to generate", min_value=1, max_value=20)
st.subheader("Enter context: ")
context_text = st.text_area("", key="context")
st.subheader("Enter question: ")
question_text = st.text_area("", key="question")
if context_text and question_text:
answers, probabilities = get_prediction(model, context_text, question_text)
st.subheader(f"Predictions")
answers = answers[0]["answer"]
context_pieces = context_text.split(answers[0])
if answers[0] != "empty":
if len(context_pieces) == 2:
st.write(
QA_ANSWER_WRAPPER.format(context_pieces[0], answers[0], context_pieces[-1]), unsafe_allow_html=True
)
else:
st.write(
QA_ANSWER_WRAPPER.format(context_pieces[0], answers[0], answers[0].join(context_pieces[1:])),
unsafe_allow_html=True,
)
else:
st.write(QA_EMPTY_ANSWER_WRAPPER.format("", answers[0], ""), unsafe_allow_html=True)
probabilities = probabilities[0]["probability"]
st.subheader("Confidence")
output_df = pd.DataFrame({"Answer": answers, "Confidence": probabilities})
st.dataframe(output_df)
return model
def get_states(model, session_state=None):
if session_state:
setattr(session_state, "max_answer_length", model.args.max_answer_length)
setattr(session_state, "max_query_length", model.args.max_query_length)
else:
session_state = get(
max_seq_length=model.args.max_seq_length,
max_answer_length=model.args.max_answer_length,
max_query_length=model.args.max_query_length,
)
model.args.max_seq_length = session_state.max_seq_length
model.args.max_answer_length = session_state.max_answer_length
model.args.max_query_length = session_state.max_query_length
return session_state, model
def get_states(model, session_state=None):
if session_state:
setattr(session_state, "sliding_window", model.args.sliding_window)
setattr(session_state, "stride", model.args.stride)
else:
session_state = get(
max_seq_length=model.args.max_seq_length,
sliding_window=model.args.sliding_window,
stride=model.args.stride,
)
if session_state.sliding_window == "Enable":
model.args.sliding_window = True
else:
model.args.sliding_window = False
model.args.max_seq_length = session_state.max_seq_length
model.args.stride = session_state.stride
return session_state, model
def t5_viewer(model):
try:
session_state, model = get_states(model)
except AttributeError:
session_state = get(
max_seq_length=model.args.max_seq_length,
max_length=model.args.max_length,
decoding_algorithm=model.args.do_sample,
)
session_state, model = get_states(model, session_state)
st.sidebar.subheader("Parameters")
model.args.max_seq_length = st.sidebar.slider(
"Max Seq Length", min_value=1, max_value=512, value=model.args.max_seq_length
)
st.sidebar.subheader("Decoding")
model.args.max_length = st.sidebar.slider(
"Max Generated Text Length",
min_value=1,
max_value=512,
value=model.args.max_length,
)
model.args.length_penalty = st.sidebar.number_input(
"Length Penalty", value=model.args.length_penalty
)
model.args.early_stopping = st.sidebar.radio(
"Early Stopping", ("True", "False"), index=0 if model.args.early_stopping else 1
)
decoding_algorithm = st.sidebar.radio(
"Decoding Algorithm",
("Sampling", "Beam Search"),
index=0 if model.args.do_sample else 1,
)
if decoding_algorithm == "Sampling":
model.args.do_sample = True
model.args.num_beams = None
elif decoding_algorithm == "Beam Search":
model.args.do_sample = False
model.args.num_beams = 1
if model.args.do_sample:
model.args.top_k = st.sidebar.number_input(
"Top-k", value=model.args.top_k if model.args.top_k else 50
)
model.args.top_p = st.sidebar.slider(
"Top-p",
min_value=0.0,
max_value=1.0,
value=model.args.top_p if model.args.top_p else 0.95,
)
else:
model.args.num_beams = st.sidebar.number_input(
"Number of Beams", value=model.args.num_beams
)
st.markdown("## Instructions: ")
st.markdown("The input to a T5 model can be providied in two ways.")
st.markdown("### Using Prefix")
st.markdown(
"If you provide a value for the `prefix`, Simple Viewer will automatically insert `: ` between the `prefix` text and the `input` text."
)
st.markdown("### Blank prefix")
st.markdown(
"You may also leave the `prefix` blank. In this case, you can provide a prefix and a separator at the start of the `input` text (if your model requires a prefix)."
)
st.subheader("Enter prefix: ")
prefix_text = st.text_input("")
st.subheader("Enter input text: ")
input_text = st.text_area("")
if input_text:
prediction = get_prediction(model, input_text, prefix_text)[0]
st.subheader(f"Generated output: ")
st.write(prediction)
def classification_viewer(model, model_class):
st.subheader("Enter text: ")
input_text = st.text_area("")
st.sidebar.subheader("Parameters")
if model_class == "ClassificationModel":
try:
session_state, model = get_states(model)
except AttributeError:
session_state = get(
max_seq_length=model.args.max_seq_length,
sliding_window=model.args.sliding_window,
stride=model.args.stride,
)
session_state, model = get_states(model, session_state)
model.args.max_seq_length = st.sidebar.slider(
"Max Seq Length",
min_value=1,
max_value=512,
value=model.args.max_seq_length,
)
sliding_window = st.sidebar.radio(
"Sliding Window",
("Enable", "Disable"),
index=0 if model.args.sliding_window else 1,
)
if sliding_window == "Enable":
model.args.sliding_window = True
else:
model.args.sliding_window = False
if model.args.sliding_window:
model.args.stride = st.sidebar.slider(
"Stride (Fraction of Max Seq Length)",
min_value=0.0,
max_value=1.0,
value=model.args.stride,
)
elif model_class == "MultiLabelClassificationModel":
try:
session_state, model = get_states(model)
except AttributeError:
session_state = get(max_seq_length=model.args.max_seq_length, )
session_state, model = get_states(model, session_state)
model.args.max_seq_length = st.sidebar.slider(
"Max Seq Length",
min_value=1,
max_value=512,
value=model.args.max_seq_length,
)
if input_text:
prediction, raw_values = get_prediction(model, input_text)
raw_values = [list(np.squeeze(raw_values))]
if model.args.sliding_window and isinstance(raw_values[0][0],
np.ndarray):
raw_values = np.mean(raw_values, axis=1)
st.subheader(f"Predictions")
st.text(f"Predicted label: {prediction[0]}")
st.subheader(f"Model outputs")
st.text("Raw values: ")
try:
raw_df = pd.DataFrame(
raw_values,
columns=[f"Label {label}" for label in model.args.labels_list],
)
except Exception:
raw_df = pd.DataFrame(
raw_values,
columns=[
f"Label {label}" for label in range(len(raw_values[0]))
],
)
st.dataframe(raw_df)
st.text("Probabilities: ")
try:
prob_df = pd.DataFrame(
softmax(raw_values, axis=1),
columns=[f"Label {label}" for label in model.args.labels_list],
)
except Exception:
prob_df = pd.DataFrame(
softmax(raw_values, axis=1),
columns=[f"Label {i}" for i in range(len(raw_values[0]))],
)
st.dataframe(prob_df)
return model