plt5 support

app.py

@@ -0,0 +1,21 @@
+import gradio as gr
+
+from models import MODELS, PIPELINES
+
+
+def predict(text: str, model_name: str) -> str:
+    return {"text": PIPELINES[model_name](text)}
+
+
+with gr.Blocks(title="CLARIN-PL Dialogue System Modules") as demo:
+    gr.Markdown("Dialogue State Tracking Modules")
+    for model_name in MODELS:
+        with gr.Row():
+            gr.Markdown(f"## {model_name}")
+            model_name_component = gr.Textbox(value=model_name, visible=False)
+        with gr.Row():
+            text_input = gr.Textbox(label="Input Text", value=MODELS[model_name]["default_input"])
+            gr.Interface(fn=predict, inputs=[text_input, model_name_component], outputs="text")
+
+demo.queue(concurrency_count=3)
+demo.launch()

models.py

@@ -0,0 +1,40 @@
+import os
+from typing import Any, Dict
+
+from transformers import (Pipeline, T5ForConditionalGeneration, T5Tokenizer,
+                          pipeline)
+
+auth_token = os.environ.get("CLARIN_KNEXT")
+
+DEFAULT_INPUTS: Dict[str, str] = {
+    "polish": (
+        "[U] Chciałbym zarezerwować stolik na 4 osoby na piątek o godzinie 18.30. "
+        "[Dziedzina] Restauracje: Popularna usługa wyszukiwania i rezerwacji restauracji "
+        "[Atrybut] Czas: Wstępny czas rezerwacji restauracji"
+    ),
+    "english": (
+        "[U] I want to book a table for 4 people on Friday, 6.30 pm. "
+        "[Domain] Restaurants: A popular restaurant search and reservation service "
+        "[Slot] Time: Tentative time of restaurant reservation"
+    ),
+}
+
+MODELS: Dict[str, Dict[str, Any]] = {
+    "plt5-small": {
+        "model": T5ForConditionalGeneration.from_pretrained("clarin-knext/plt5-small-dst", use_auth_token=auth_token),
+        "tokenizer": T5Tokenizer.from_pretrained("clarin-knext/plt5-small-dst", use_auth_token=auth_token),
+        "default_input": DEFAULT_INPUTS["polish"],
+    },
+    "plt5-base": {
+        "model": T5ForConditionalGeneration.from_pretrained("clarin-knext/plt5-base-dst", use_auth_token=auth_token),
+        "tokenizer": T5Tokenizer.from_pretrained("clarin-knext/plt5-base-dst", use_auth_token=auth_token),
+        "default_input": DEFAULT_INPUTS["polish"],
+    },
+}
+
+PIPELINES: Dict[str, Pipeline] = {
+    model_name: pipeline(
+        "text2text-generation", model=MODELS[model_name]["model"], tokenizer=MODELS[model_name]["tokenizer"]
+    )
+    for model_name in MODELS
+}

requirements.txt

@@ -0,0 +1,3 @@
+torch
+transformers
+sentencepiece