initial code

.gitignore

@@ -0,0 +1 @@
+models/prot_t5_xl_half_uniref50-enc/

app.py

@@ -114,13 +114,21 @@ def load_t5():
         _modeling_utils.check_torch_load_is_safe = _noop
 
         try:
-            hf_name = "Rostlab/prot_t5_xl_half_uniref50-enc"
+            # Pre-fetched into the HF cache at Docker build time (see Dockerfile's
+            # snapshot_download step), so ZeroGPU calls never hit the network here —
+            # a cold download would blow past the GPU-allocation time budget and
+            # get aborted. local_files_only=True enforces that guarantee.
+            repo_id = "Rostlab/prot_t5_xl_half_uniref50-enc"
             dev = get_device()
             dtype = torch.float16 if dev == "cuda" else torch.float32
 
-            _t5_tokenizer = AutoTokenizer.from_pretrained(hf_name, do_lower_case=False)
+            _t5_tokenizer = AutoTokenizer.from_pretrained(
+                repo_id, do_lower_case=False, local_files_only=True
+            )
             _t5_model = (
-                T5EncoderModel.from_pretrained(hf_name, torch_dtype=dtype)
+                T5EncoderModel.from_pretrained(
+                    repo_id, torch_dtype=dtype, local_files_only=True
+                )
                 .to(dev)
                 .eval()
             )
@@ -259,7 +267,7 @@ _SPINNER_HTML = """
   <span style="color:#555;font-size:14px;font-weight:500;line-height:1.5;">
     ProtT5-XL feature extraction in progress…<br>
     <span style="font-size:12px;color:#999;font-weight:400;">
-      First run downloads the model (~1.2 GB) — this may take a few minutes.
+      First run loads the model onto the GPU — this may take a moment.
     </span>
   </span>
 </div>
@@ -474,7 +482,7 @@ with gr.Blocks(
         gr.Markdown("## Step 2 — Extract ProtT5 Feature")
         gr.Markdown(
             "Runs the [**ProtT5-XL**](https://github.com/agemagician/ProtTrans) encoder to produce a 1024-dim mean-pool embedding.  \n"
-            "⏳ *First call downloads the model (~3 GB) and may take a few minutes.*"
+            "⏳ *First call loads the model onto the GPU and may take a moment.*"
         )
         extract_btn    = gr.Button("🔬 Extract Feature", variant="primary", size="lg")
         extract_status = gr.HTML("")