space: transformers 5 apply_chat_template returns BatchEncoding — use return_dict + **enc into generate (fixes AttributeError on .shape)

space/app.py

@@ -86,21 +86,26 @@ WARDEN_READY = not WARDEN_ERR
 def _generate_impl(messages, max_tokens, temperature, enable_thinking):
     import torch
 
-    inputs = tok.apply_chat_template(
+    # transformers 5: apply_chat_template returns a BatchEncoding (dict), not a
+    # bare tensor — splat it into generate() rather than passing as input_ids.
+    enc = tok.apply_chat_template(
         messages,
         add_generation_prompt=True,
         return_tensors="pt",
+        return_dict=True,
         enable_thinking=enable_thinking,
-    ).to("cuda")
+    )
+    enc = {k: v.to("cuda") for k, v in enc.items()}
     with torch.no_grad():
         out = model.generate(
-            input_ids=inputs,
+            **enc,
             max_new_tokens=max_tokens,
             do_sample=temperature > 0,
             temperature=max(temperature, 1e-3),
             top_p=0.95,
         )
-    return tok.decode(out[0, inputs.shape[1]:], skip_special_tokens=True)
+    input_len = enc["input_ids"].shape[1]
+    return tok.decode(out[0, input_len:], skip_special_tokens=True)
 
 
 # bf16 30B (~60GB) needs the 96GB xlarge slice; duration covers first-call