Add API documentation tab

app.py

@@ -180,38 +180,85 @@ with gr.Blocks(title="BERT Metagenome Embeddings") as demo:
 Extract embeddings from DNA sequences. BERT model (430M params) pretrained on metagenomic sequences.
     """)
 
-    with gr.Row():
-        with gr.Column(scale=1):
-            seq_input = gr.Textbox(
-                label="sequence",
-                placeholder="Paste DNA sequence (FASTA or raw)...",
-                lines=8,
-                value=EXAMPLE_SEQUENCE,
-                info="min 1000 bp"
-            )
-            mode_input = gr.Radio(
-                choices=["mean", "max", "per-window"],
-                value="mean",
-                label="pooling",
-                info="mean/max: single 768-dim vector | per-window: (n, 768) matrix"
-            )
-            stride_input = gr.Slider(
-                minimum=50, maximum=500, value=100, step=50,
-                label="stride",
-                info="step size between windows"
-            )
-            btn = gr.Button("extract", variant="primary")
-
-        with gr.Column(scale=1):
-            output = gr.Markdown()
-            download = gr.File(label="download")
+    with gr.Tab("Extract"):
+        with gr.Row():
+            with gr.Column(scale=1):
+                seq_input = gr.Textbox(
+                    label="sequence",
+                    placeholder="Paste DNA sequence (FASTA or raw)...",
+                    lines=8,
+                    value=EXAMPLE_SEQUENCE,
+                    info="min 1000 bp"
+                )
+                mode_input = gr.Radio(
+                    choices=["mean", "max", "per-window"],
+                    value="mean",
+                    label="pooling",
+                    info="mean/max: single 768-dim vector | per-window: (n, 768) matrix"
+                )
+                stride_input = gr.Slider(
+                    minimum=50, maximum=500, value=100, step=50,
+                    label="stride",
+                    info="step size between windows"
+                )
+                btn = gr.Button("extract", variant="primary")
+
+            with gr.Column(scale=1):
+                output = gr.Markdown()
+                download = gr.File(label="download")
 
     btn.click(
         process,
         inputs=[seq_input, mode_input, stride_input],
-        outputs=[output, download]
+        outputs=[output, download],
+        api_name="embed"
     )
 
+    with gr.Tab("API"):
+        gr.Markdown("""
+### API
+
+```python
+from gradio_client import Client
+import numpy as np
+
+client = Client("genomenet/bert-embedding")
+
+# Extract embedding (mean pooled)
+result = client.predict(
+    sequence="ATGCGATCGATCG...",  # min 1000 bp
+    mode="mean",                   # "mean", "max", or "per-window"
+    stride=100,                    # step size between windows
+    api_name="/embed"
+)
+
+# result = (summary_text, embedding_file_path)
+summary, emb_path = result
+embedding = np.load(emb_path)
+print(embedding.shape)  # (768,) for mean/max, (n_windows, 768) for per-window
+```
+
+**Modes**:
+- `mean`: Single 768-dim vector (mean pooled across all windows)
+- `max`: Single 768-dim vector (max pooled across all windows)
+- `per-window`: Matrix of shape `(n_windows, 768)` - one embedding per sliding window
+
+**Parameters**:
+| param | type | description |
+|-------|------|-------------|
+| sequence | str | DNA sequence (min 1000 bp) |
+| mode | str | Pooling mode: "mean", "max", "per-window" |
+| stride | int | Step size between windows (50-500) |
+
+**Local**:
+```bash
+git clone https://huggingface.co/spaces/genomenet/bert-embedding
+cd bert-embedding
+pip install -r requirements.txt
+python app.py
+```
+        """)
+
     gr.Markdown("""
 ---
 **Model**: [genomenet/bert-metagenome](https://huggingface.co/genomenet/bert-metagenome) |