Improve Space default prediction responsiveness

README.md

@@ -48,12 +48,12 @@ git clone https://huggingface.co/spaces/genomenet/crispr-array-detection
 ### Space settings (HuggingFace web UI)
 
 - SDK: Docker
-- Hardware: T4 GPU
+- Hardware: CPU Basic works for the default demo; T4 GPU is recommended for long sequences or low stride values
 - Visibility: Public
 
 ### Model weights
 
-Hosted at: https://huggingface.co/pmuench3/crispr-bert-model
+Hosted at: https://huggingface.co/genomenet/crispr-bert-model
 
 Downloaded automatically via `huggingface_hub` at startup.
 

REPORT_SUMMARY.md

@@ -2,14 +2,14 @@
 
 **Date:** April 2026
 **Repository:** `/vol/hpcprojects/pmuench/crispr_tool/crispr-hf-space/`
-**HuggingFace Space:** https://huggingface.co/spaces/pmuench3/crispr-array-detection
-**HuggingFace Model Repository:** https://huggingface.co/pmuench3/crispr-bert-model
+**HuggingFace Space:** https://huggingface.co/spaces/genomenet/crispr-array-detection
+**HuggingFace Model Repository:** https://huggingface.co/genomenet/crispr-bert-model
 
 ---
 
 ## Summary
 
-We have successfully deployed a publicly accessible web application for CRISPR array detection based on the BERT-based deep learning model developed in Ziyu Mu's Master's thesis. The application is hosted on HuggingFace Spaces with GPU acceleration (T4) and provides both interactive visualization and programmatic access to the model's predictions.
+We have successfully deployed a publicly accessible web application for CRISPR array detection based on the BERT-based deep learning model developed in Ziyu Mu's Master's thesis. The application is hosted on HuggingFace Spaces and provides both interactive visualization and programmatic access to the model's predictions.
 
 ---
 
@@ -21,7 +21,7 @@ We have successfully deployed a publicly accessible web application for CRISPR a
 - **Output:** Per-position CRISPR probability scores (0-1)
 - **Visualization:** Interactive probability curve along sequence position
 - **Parameters:**
-  - Configurable stride (50-500 bp, default 100)
+  - Configurable stride (50-500 bp, default 500 for CPU responsiveness)
   - Adjustable detection threshold (0.1-0.9, default 0.3)
 - **Region Detection:** Automatic identification and annotation of predicted CRISPR regions above threshold
 
@@ -101,7 +101,7 @@ plotly>=5.18.0
 From the original TODO:
 
 - [x] **Checkpoint beschaffen** - Model `best.h5` located and uploaded to HF Model Hub
-- [x] **Eigenes Repo anlegen** - Created HuggingFace Space `pmuench3/crispr-array-detection`
+- [x] **Eigenes Repo anlegen** - Created HuggingFace Space `genomenet/crispr-array-detection`
 - [x] **Code-Verständnis** - Analyzed custom layers, tokenization, sliding window logic
 - [x] **Model-Loader (Singleton)** - Implemented with HF Hub download
 - [x] **Tokenizer** - Extracted and adapted for inference
@@ -112,7 +112,7 @@ From the original TODO:
 - [x] **State-Dynamics Visualization** - UMAP + clustering + interactive Plotly plots
 - [x] **Input-Validation** - Sequence validation, FASTA header stripping
 - [x] **Health Endpoint equivalent** - GPU status shown in UI
-- [x] **Deployment** - Live on HuggingFace Spaces with T4 GPU
+- [x] **Deployment** - Live on HuggingFace Spaces; GPU hardware is recommended for long sequences
 - [x] **Acknowledgements** - Ziyu Mu, DFG SPP 2141, BMBF GenomeNet, HZI BIFO
 
 ---
@@ -121,7 +121,7 @@ From the original TODO:
 
 ### Web Interface
 
-1. Navigate to https://huggingface.co/spaces/pmuench3/crispr-array-detection
+1. Navigate to https://huggingface.co/spaces/genomenet/crispr-array-detection
 2. Paste DNA sequence (or use provided examples)
 3. Click "Analyze Sequence" for CRISPR detection
 4. Use "Embeddings" tab for State-Dynamic Plots
@@ -131,12 +131,12 @@ From the original TODO:
 ```python
 from gradio_client import Client
 
-client = Client("pmuench3/crispr-array-detection")
+client = Client("genomenet/crispr-array-detection")
 
 # Predict CRISPR regions
 result = client.predict(
     sequence="ACGT...",
-    stride=100,
+    stride=500,
     threshold=0.3,
     api_name="/predict"
 )
@@ -155,16 +155,16 @@ embedding = client.predict(
 
 ### Suggested Text (German)
 
-> Im Rahmen des SPP 2141 wurde ein öffentlich zugänglicher Webservice zur CRISPR-Array-Detektion entwickelt und auf HuggingFace Spaces bereitgestellt (https://huggingface.co/spaces/pmuench3/crispr-array-detection). Das System basiert auf einem BERT-basierten Deep-Learning-Modell (~430 Mio. Parameter), das auf metagenomischen und genomischen mikrobiellen Sequenzen vortrainiert und anschließend auf annotierten CRISPR-Arrays feinabgestimmt wurde.
+> Im Rahmen des SPP 2141 wurde ein öffentlich zugänglicher Webservice zur CRISPR-Array-Detektion entwickelt und auf HuggingFace Spaces bereitgestellt (https://huggingface.co/spaces/genomenet/crispr-array-detection). Das System basiert auf einem BERT-basierten Deep-Learning-Modell (~430 Mio. Parameter), das auf metagenomischen und genomischen mikrobiellen Sequenzen vortrainiert und anschließend auf annotierten CRISPR-Arrays feinabgestimmt wurde.
 >
 > Der Service bietet:
 > - Vorhersage von CRISPR-Array-Wahrscheinlichkeiten entlang der Sequenzposition
 > - Extraktion von Hidden-State-Embeddings aus dem Transformer-Modell
 > - State-Dynamic-Plots zur Visualisierung der Einbettungstrajektorien mittels UMAP und Clustering
 >
-> Die State-Dynamic-Visualisierung ermöglicht die Identifikation wiederkehrender Strukturelemente (z.B. Repeats vs. Spacer) durch die Analyse der Aktivierungsmuster im neuronalen Netzwerk. Der Service läuft auf GPU-beschleunigter Hardware (NVIDIA T4) und ist für die wissenschaftliche Community frei zugänglich.
+> Die State-Dynamic-Visualisierung ermöglicht die Identifikation wiederkehrender Strukturelemente (z.B. Repeats vs. Spacer) durch die Analyse der Aktivierungsmuster im neuronalen Netzwerk. GPU-Hardware wird für lange Sequenzen und hohe Auflösung empfohlen. Der Service ist für die wissenschaftliche Community frei zugänglich.
 >
-> **Referenz:** https://huggingface.co/spaces/pmuench3/crispr-array-detection
+> **Referenz:** https://huggingface.co/spaces/genomenet/crispr-array-detection
 
 ### Acknowledgements (for publication)
 

app.py

@@ -4,6 +4,7 @@ CRISPR Array Detection - HuggingFace Spaces App
 
 import os
 import html
+import logging
 import tempfile
 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
 os.environ.setdefault("MPLCONFIGDIR", os.path.join(tempfile.gettempdir(), "matplotlib"))
@@ -26,10 +27,15 @@ from inference.model_loader import get_model, warmup_model, get_gpu_status
 from inference.tokenizer import validate_sequence, strip_fasta_header
 from inference.inference import detect_crispr_regions
 
+logging.basicConfig(level=os.environ.get("LOG_LEVEL", "INFO"))
+logger = logging.getLogger(__name__)
+
 MAX_SEQUENCE_LENGTH = int(os.environ.get("MAX_SEQUENCE_LENGTH", "50000"))
 MAX_UPLOAD_BYTES = int(os.environ.get("MAX_UPLOAD_BYTES", str(2 * 1024 * 1024)))
 MAX_SEQUENCE_VIEWER_LENGTH = int(os.environ.get("MAX_SEQUENCE_VIEWER_LENGTH", "20000"))
 QUEUE_MAX_SIZE = int(os.environ.get("GRADIO_QUEUE_MAX_SIZE", "8"))
+DEFAULT_STRIDE = int(os.environ.get("DEFAULT_STRIDE", "500"))
+DEFAULT_THRESHOLD = float(os.environ.get("DEFAULT_THRESHOLD", "0.3"))
 
 # Custom CSS - Minimal monochrome design with Geist fonts
 CUSTOM_CSS = """
@@ -915,7 +921,7 @@ def create_sequence_viewer_html(sequence, positions, probabilities, threshold=0.
     return ''.join(html_parts)
 
 
-def predict(sequence: str, stride: int = 100, threshold: float = 0.3):
+def predict(sequence: str, stride: int = DEFAULT_STRIDE, threshold: float = DEFAULT_THRESHOLD):
     """Predict CRISPR array probability for each position."""
     import csv
     import time
@@ -1037,7 +1043,7 @@ def detect(sequence: str, threshold: float = 0.3, min_length: int = 160):
         sequence,
         threshold=threshold,
         min_length=min_length,
-        stride=100
+        stride=DEFAULT_STRIDE
     )
 
     if not regions:
@@ -1194,9 +1200,9 @@ Sliding window analysis with per-position probability scores. Export to GFF3/CSV
                 )
                 with gr.Row():
                     stride_input = gr.Slider(
-                        minimum=50, maximum=500, value=100, step=50,
+                        minimum=50, maximum=500, value=DEFAULT_STRIDE, step=50,
                         label="stride",
-                        info="lower = higher resolution"
+                        info="500 = fast on CPU; lower = higher resolution but slower"
                     )
                     threshold_input = gr.Slider(
                         minimum=0.1, maximum=0.9, value=0.3, step=0.05,
@@ -1252,7 +1258,11 @@ Sliding window analysis with per-position probability scores. Export to GFF3/CSV
         )
 
         def predict_and_show_downloads(*args):
-            results = predict(*args)
+            try:
+                results = predict(*args)
+            except Exception as exc:
+                logger.exception("Prediction failed")
+                results = prediction_error_outputs(f"Analysis failed: {exc}")
             # results = (fig, summary, regions, png, pdf, csv, summary_txt, gff, seq_html)
             # Return results plus visibility updates for accordions
             success = results[0] is not None
@@ -1322,7 +1332,11 @@ Repeats cluster together, spacers form distinct groups.
         )
 
         def embed_and_show_downloads(*args):
-            results = get_embedding(*args)
+            try:
+                results = get_embedding(*args)
+            except Exception as exc:
+                logger.exception("Embedding failed")
+                results = embedding_error_outputs(f"Embedding failed: {exc}")
             success = results[0] is not None
             return results + (gr.update(visible=success),)
 
@@ -1346,7 +1360,7 @@ client = Client("genomenet/crispr-array-detection")
 # predict
 result = client.predict(
     sequence="ATGC...",
-    stride=100,
+    stride=500,
     threshold=0.3,
     api_name="/predict"
 )
@@ -1384,7 +1398,7 @@ pip install -r requirements.txt && python app.py
 
 | param | default | range |
 |-------|---------|-------|
-| stride | 100 bp | 50-500 |
+| stride | 500 bp | 50-500 |
 | threshold | 0.3 | 0.1-0.9 |
 
 **citation**

inference/model_loader.py

@@ -25,7 +25,7 @@ _embedding_model: Optional[tf.keras.Model] = None
 _model_lock = threading.Lock()
 
 # HuggingFace model repository
-HF_MODEL_REPO = os.environ.get("CRISPR_HF_REPO", "pmuench3/crispr-bert-model")
+HF_MODEL_REPO = os.environ.get("CRISPR_HF_REPO", "genomenet/crispr-bert-model")
 HF_MODEL_FILENAME = os.environ.get("CRISPR_HF_FILENAME", "best.h5")
 
 # Local model path (optional override)