debug

app.py

@@ -131,12 +131,12 @@ def get_response(params):
         images = None
         image_args = {}
 
-    temperature = float(params.get("temperature", 1.0))
-    top_p = float(params.get("top_p", 1.0))
+    temperature = 0.0
+    top_p = 1.0
     max_context_length = getattr(model.config, "max_position_embeddings", 2048)
-    max_new_tokens = min(int(params.get("max_new_tokens", 256)), 1024)
+    max_new_tokens = 512
     stop_str = params.get("stop", None)
-    do_sample = False if temperature > 0.001 else False
+    do_sample = False
     logger.info(prompt)
     input_ids = (
         tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt")
@@ -266,9 +266,9 @@ def http_bot(state, temperature, top_p, max_new_tokens):
     pload = {
         "model": model_name,
         "prompt": prompt,
-        "temperature": float(0.),
-        "top_p": float(top_p),
-        "max_new_tokens": min(int(max_new_tokens), 1536),
+        "temperature": temperature,
+        "top_p": top_p,
+        "max_new_tokens": max_new_tokens,
         "stop": (
             state.sep
             if state.sep_style in [SeparatorStyle.SINGLE, SeparatorStyle.MPT]
@@ -341,31 +341,31 @@ def build_demo():
                 #     inputs=[imagebox, textbox],
                 # )
 
-                with gr.Accordion("Parameters", open=False) as _:
-                    temperature = gr.Slider(
-                        minimum=0.0,
-                        maximum=1.0,
-                        value=0.2,
-                        step=0.1,
-                        interactive=True,
-                        label="Temperature",
-                    )
-                    top_p = gr.Slider(
-                        minimum=0.0,
-                        maximum=1.0,
-                        value=0.7,
-                        step=0.1,
-                        interactive=True,
-                        label="Top P",
-                    )
-                    max_output_tokens = gr.Slider(
-                        minimum=0,
-                        maximum=1024,
-                        value=512,
-                        step=64,
-                        interactive=True,
-                        label="Max output tokens",
-                    )
+                # with gr.Accordion("Parameters", open=False) as _:
+                #     temperature = gr.Slider(
+                #         minimum=0.0,
+                #         maximum=1.0,
+                #         value=0.2,
+                #         step=0.1,
+                #         interactive=True,
+                #         label="Temperature",
+                #     )
+                #     top_p = gr.Slider(
+                #         minimum=0.0,
+                #         maximum=1.0,
+                #         value=0.7,
+                #         step=0.1,
+                #         interactive=True,
+                #         label="Top P",
+                #     )
+                #     max_output_tokens = gr.Slider(
+                #         minimum=0,
+                #         maximum=1024,
+                #         value=512,
+                #         step=64,
+                #         interactive=True,
+                #         label="Max output tokens",
+                #     )
 
             with gr.Column(scale=8):
                 chatbot = gr.Chatbot(elem_id="chatbot", label="Chatbot", height=550)