fix byte string, (dropdown is broken)

README.md

@@ -4,7 +4,7 @@ emoji: 💻
 colorFrom: green
 colorTo: red
 sdk: gradio
-sdk_version: 3.47.1
+sdk_version: 4.1.1
 app_file: app.py
 pinned: true
 license: mit

app.py

@@ -2,7 +2,7 @@ import gradio as gr
 from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
 import datasets
 import numpy as np
-import torch
+# import torch
 
 from utils.tree_utils import parse_functions, get_docstrings, grab_before_comments, line_chr2char, node_str_idx, replace_function
 from utils.html_utils import make_iframe, construct_embed
@@ -211,7 +211,7 @@ def alter_body(old_code, func_id, funcs_list: list, prompt="", temperature=0.2,
     yield altered_code #yield once so it updates? -> works... gg but doesn't seem to do it for the dropdown
     return altered_code #never gets used by the code block? maybe I need to yield it first? but works in the ov_notebook
 
-def list_dropdown(in_code): #only used for auto update, not on sample pick?
+def list_dropdown_options(in_code): #only used for auto update, not on sample pick?
     funcs = parse_functions(in_code)
     func_identifiers = [f"{idx:2d}: {n.child_by_field_name('declarator').text.decode()}" for idx, n in enumerate(funcs)]
     # funcs = [n for n in funcs] #wrapped as set to avoid json issues?
@@ -225,11 +225,11 @@ if __name__ == "__main__": #works on huggingface?
     all_single_passes = datasets.concatenate_datasets([single_passes["train"], single_passes["test"]])
     num_samples = len(all_single_passes)    
     
-    with gr.Blocks() as site:
+    with gr.Blocks() as demo:
         top_md = gr.Markdown(intro_text)
         model_cp = gr.Textbox(value="Vipitis/santacoder-finetuned-Shadertoys-fine", label="Model Checkpoint (Enter to load!)", interactive=True)
         sample_idx = gr.Slider(minimum=0, maximum=10513, value=3211, label="pick sample from dataset", step=1.0)
-        func_dropdown = gr.Dropdown(value=["0: edit the Code (or load a shader) to update this dropdown"], label="chose a function to modify") #breaks if I add a string in before that? #TODO: use type="index" to get int - always gives None?
+        func_dropdown = gr.Dropdown(choices=["0: edit the Code (or load a shader) to update this dropdown"], label="chose a function to modify") #breaks if I add a string in before that? #TODO: use type="index" to get int - always gives None?
         prompt_text = gr.Textbox(value="the title used by the model has generation hint", label="prompt text", info="leave blank to skip", interactive=True)
         with gr.Accordion("Advanced settings", open=False): # from: https://huggingface.co/spaces/bigcode/bigcode-playground/blob/main/app.py
             with gr.Row():
@@ -273,8 +273,8 @@ if __name__ == "__main__": #works on huggingface?
                         info="Penalize repeated tokens",
                     )
         with gr.Row():
-            gen_return_button = gr.Button("generate a alternate return statement", label="generate return", scale=0)
-            gen_func_button = gr.Button("generate an alternate function body", label="generate function", scale=1)
+            gen_return_button = gr.Button("generate a alternate return statement", scale=0)
+            gen_func_button = gr.Button("generate an alternate function body", scale=1)
         with gr.Row():
             with gr.Column():
                 source_embed = gr.HTML('<iframe width="640" height="360" frameborder="0" src="" allowfullscreen></iframe>', label="How this shader originally renders")
@@ -290,10 +290,10 @@ if __name__ == "__main__": #works on huggingface?
         sample_idx.release(fn=grab_sample, inputs=[sample_idx], outputs=[sample_pass, sample_code, prompt_text, source_embed]) #funcs here?
         gen_return_button.click(fn=alter_return, inputs=[sample_code, func_dropdown, temperature, max_new_tokens, top_p, repetition_penalty, pipe], outputs=[sample_code])
         gen_func_button.click(fn=alter_body, inputs=[sample_code, func_dropdown, funcs, prompt_text, temperature, max_new_tokens, top_p, repetition_penalty, pipe], outputs=[sample_code]).then(
-            fn=list_dropdown, inputs=[sample_code], outputs=[funcs, func_dropdown]
+            fn=list_dropdown_options, inputs=[sample_code], outputs=[funcs, func_dropdown]
         )
-        sample_code.change(fn=list_dropdown, inputs=[sample_code], outputs=[funcs, func_dropdown]).then(
+        sample_code.change(fn=list_dropdown_options, inputs=[sample_code], outputs=[funcs, func_dropdown]).then(
             fn=make_iframe, inputs=[sample_code], outputs=[our_embed])
 
-    site.queue()
-    site.launch()
+    demo.queue()
+    demo.launch()

utils/generation.py

@@ -35,7 +35,7 @@ def stream_generation(prompt:str, pipe, gen_kwargs:dict):
 
     # Start generation on a separate thread, so that we don't block the UI. The text is pulled from the streamer
     # in the main thread. Adds timeout to the streamer to handle exceptions in the generation thread.
-    streamer = TextIteratorStreamer(pipe.tokenizer, skip_prompt=True, skip_special_tokens=True, timeout=15.0)
+    streamer = TextIteratorStreamer(pipe.tokenizer, skip_prompt=True, skip_special_tokens=True, timeout=45.0) #IPEX takes a bit on first inference, to avoid an error with the empty queue timeout on the first time, we just wait longer.
     generate_kwargs = dict(model_inputs, streamer=streamer, **gen_kwargs)
     t = Thread(target=pipe.model.generate, kwargs=generate_kwargs)
     t.start()

utils/tree_utils.py

@@ -13,7 +13,7 @@ def replace_function(old_func_node, new_func_node):
     tree = give_tree(old_func_node)
     old_func_start, old_func_end = node_str_idx(old_func_node)
     # new_func_start, new_func_end = node_str_idx(new_func_node)
-    new_code = tree.text[:old_func_start] + new_func_node.text + tree.text[old_func_end:]
+    new_code = tree.text.decode()[:old_func_start] + new_func_node.text.decode() + tree.text.decode()[old_func_end:]
     return new_code
 
 def get_root(node):