yes

app.py

@@ -273,7 +273,7 @@ def build_bias_from_tokens(tokenizer, mapping: Dict[str, float]) -> torch.Tensor
             for t in tid:
                 if isinstance(t, int) and t >= 0:
                     bias[t] += float(w) / max(1, len(tid))
-        elif isinstance(tid, int) and tid >= 0:
+        elif isinstance(tid, int) and t >= 0:
             bias[tid] += float(w)
     return bias
 
@@ -341,9 +341,6 @@ def zerogpu_generate(full_prompt,
             sc = StoppingCriteriaList([StopOnTokens(HARMONY_STOP_IDS)])
 
         # Generate
-        # Build EOS list: use ONLY Harmony assistant-action stops (per OpenAI docs)
-        eos_ids = HARMONY_STOP_IDS if HARMONY_AVAILABLE else tokenizer.eos_token_id
-
         out_ids = model.generate(
             **inputs,
             do_sample=bool(gen_kwargs.get("do_sample", True)),
@@ -659,15 +656,15 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
             value=""
         )
 
-    # Chat interface - using only valid parameters
     # --- Harmony Inspector UI ---
-with gr.Accordion("Harmony Inspector", open=False):
-    debug_prompt = gr.Textbox(label="Debug prompt", value="What is 2+2? Reply with just the number.")
-    run_debug = gr.Button("Run Harmony Inspect")
-    debug_out = gr.JSON(label="Parsed Harmony output", value={})
-    run_debug.click(harmony_inspect_handler, inputs=[debug_prompt, system_prompt, reasoning_effort], outputs=[debug_out])
+    with gr.Accordion("Harmony Inspector", open=False):
+        debug_prompt = gr.Textbox(label="Debug prompt", value="What is 2+2? Reply with just the number.")
+        run_debug = gr.Button("Run Harmony Inspect")
+        debug_out = gr.JSON(label="Parsed Harmony output", value={})
+        run_debug.click(harmony_inspect_handler, inputs=[debug_prompt, system_prompt, reasoning_effort], outputs=[debug_out])
 
-chat = gr.ChatInterface(
+    # Chat interface - using only valid parameters
+    chat = gr.ChatInterface(
         fn=generate_response,
         type="messages",
         additional_inputs=[
@@ -686,18 +683,18 @@ chat = gr.ChatInterface(
         cache_examples=False,
     )
 
-gr.Markdown(
-    """
-    ---
-    ### Configuration:
-    - **Model**: Set `MODEL_ID` env var (default: openai/gpt-oss-20b)
-    - **Adapter**: Set `ADAPTER_ID` and optionally `ADAPTER_SUBFOLDER`
-    - **Auth**: Set `HF_TOKEN` in Space secrets for private model access
-    - **Harmony**: Install with `pip install openai-harmony` for proper channel support
-    
-    The model uses Harmony format with thinking channels (`thinking`, `analysis`, `final`).
-    """
-)
+    gr.Markdown(
+        """
+        ---
+        ### Configuration:
+        - **Model**: Set `MODEL_ID` env var (default: openai/gpt-oss-20b)
+        - **Adapter**: Set `ADAPTER_ID` and optionally `ADAPTER_SUBFOLDER`
+        - **Auth**: Set `HF_TOKEN` in Space secrets for private model access
+        - **Harmony**: Install with `pip install openai-harmony` for proper channel support
+        
+        The model uses Harmony format with thinking channels (`thinking`, `analysis`, `final`).
+        """
+    )
 
 if __name__ == "__main__":
     demo.queue(max_size=8 if ZEROGPU else 32).launch(