Create app_2.py

app_2.py

@@ -0,0 +1,246 @@
+import gradio as gr
+from transformers import pipeline
+
+# ---------------------------------------------------
+# Models
+# ---------------------------------------------------
+
+MODEL_NAMES = [
+    "c-ho/2026-04-24-crf-classweights-clean",
+    "c-ho/2026-04-23-crf-classweights-clean",
+]
+
+EXAMPLE_TEXT = (
+    "As a result, Indo-European developed a minimal vowel system "
+    "combined with a very large consonant inventory including "
+    "glottalized stops, also grammatical gender and adjectival agreement."
+)
+
+# ---------------------------------------------------
+# Lazy model cache
+# ---------------------------------------------------
+
+model_cache = {}
+
+def get_model(model_name):
+    if model_name not in model_cache:
+        model_cache[model_name] = pipeline(
+            "ner",
+            model=model_name,
+            aggregation_strategy="simple"
+        )
+    return model_cache[model_name]
+
+# ---------------------------------------------------
+# Model info
+# ---------------------------------------------------
+
+model_info = {
+    m: {
+        "link": f"https://huggingface.co/{m}",
+        "usage": f'''from transformers import pipeline
+
+ner = pipeline(
+    "ner",
+    model="{m}",
+    aggregation_strategy="simple"
+)
+
+result = ner("Hello world")
+print(result)
+'''
+    }
+    for m in MODEL_NAMES
+}
+
+# ---------------------------------------------------
+# UI helper
+# ---------------------------------------------------
+
+def display_model_info(model_name):
+    info = model_info[model_name]
+
+    return (
+        info["usage"],
+        f"[Open model page]({info['link']})"
+    )
+
+# ---------------------------------------------------
+# Merge subwords into full spans
+# ---------------------------------------------------
+
+def merge_subwords(results):
+    merged = []
+
+    current = None
+
+    for token in results:
+        word = token.get("word", "")
+        label = token.get(
+            "entity_group",
+            token.get("entity", "UNK")
+        )
+
+        score = token.get("score", 0.0)
+
+        start = token.get("start", 0)
+        end = token.get("end", 0)
+
+        # Continuation token
+        if word.startswith("##") and current is not None:
+            current["word"] += word[2:]
+            current["end"] = end
+            current["score"] = max(current["score"], score)
+
+        else:
+            # flush previous
+            if current is not None:
+                merged.append(current)
+
+            current = {
+                "word": word,
+                "start": start,
+                "end": end,
+                "entity_group": label,
+                "score": score
+            }
+
+    if current is not None:
+        merged.append(current)
+
+    return merged
+
+# ---------------------------------------------------
+# Main inference function
+# ---------------------------------------------------
+
+def analyze_text(text, model_name):
+    ner = get_model(model_name)
+
+    results = ner(text)
+
+    results = merge_subwords(results)
+
+    entities = []
+
+    table_rows = []
+
+    for ent in results:
+        label = ent["entity_group"]
+
+        entities.append({
+            "start": ent["start"],
+            "end": ent["end"],
+            "label": label,
+        })
+
+        table_rows.append([
+            ent["word"],
+            label,
+            round(ent["score"], 3)
+        ])
+
+    highlighted_output = {
+        "text": text,
+        "entities": entities
+    }
+
+    return highlighted_output, table_rows
+
+# ---------------------------------------------------
+# Entity colors
+# ---------------------------------------------------
+
+COLOR_MAP = {
+    "LanguageRelatedTerm": "#ffcc00",
+    "OtherLinguisticTerm": "#99ccff",
+    "PhonologicalPhenomenon": "#ff9999",
+    "MorphosyntacticPhenomenon": "#99ff99",
+    "TOPNODE_DUMMY": "#dddddd",
+}
+
+# ---------------------------------------------------
+# UI
+# ---------------------------------------------------
+
+with gr.Blocks(title="Linguistic Annotation Demo") as demo:
+
+    gr.Markdown(
+        """
+# Linguistic Annotation Demo
+
+This Space demonstrates custom linguistic sequence tagging models
+for detecting linguistic terminology and phenomena.
+"""
+    )
+
+    with gr.Row():
+
+        with gr.Column(scale=1):
+
+            model_selector = gr.Dropdown(
+                choices=MODEL_NAMES,
+                value=MODEL_NAMES[0],
+                label="Select Model"
+            )
+
+            text_input = gr.Textbox(
+                label="Input Text",
+                lines=8,
+                value=EXAMPLE_TEXT
+            )
+
+            run_button = gr.Button("Run Annotation")
+
+        with gr.Column(scale=1):
+
+            code_output = gr.Code(
+                label="Transformers Usage"
+            )
+
+            link_output = gr.Markdown()
+
+    highlighted_output = gr.HighlightedText(
+        label="Annotated Text",
+        combine_adjacent=True,
+        color_map=COLOR_MAP,
+        show_legend=True
+    )
+
+    entity_table = gr.Dataframe(
+        headers=["Text", "Label", "Confidence"],
+        datatype=["str", "str", "number"],
+        interactive=False,
+        label="Detected Entities"
+    )
+
+    # -------------------------
+    # Events
+    # -------------------------
+
+    run_button.click(
+        analyze_text,
+        inputs=[text_input, model_selector],
+        outputs=[highlighted_output, entity_table]
+    )
+
+    model_selector.change(
+        display_model_info,
+        inputs=model_selector,
+        outputs=[code_output, link_output]
+    )
+
+    demo.load(
+        display_model_info,
+        inputs=model_selector,
+        outputs=[code_output, link_output]
+    )
+
+# ---------------------------------------------------
+# Launch
+# ---------------------------------------------------
+
+demo.launch(
+    server_name="0.0.0.0",
+    server_port=7860
+)