Create app.py

app.py

@@ -0,0 +1,167 @@
+import gradio as gr
+from gradio_client import Client
+from huggingface_hub import InferenceClient, HFRepository
+
+import random
+
+# Cargar modelos desde un archivo de texto
+def load_models_from_file(filename):
+    with open(filename, 'r') as file:
+        models = [line.strip() for line in file.readlines() if line.strip()]
+    return models
+
+# Cargar modelos desde un archivo de texto (models.txt)
+models = load_models_from_file("models.txt")
+
+# Crear clientes de inferencia para cada modelo
+clients = []
+for model in models:
+    try:
+        clients.append(InferenceClient(model))
+    except Exception as e:
+        print(f"Error loading model {model}: {str(e)}")
+
+# Cargar conjuntos de datos desde un archivo de texto
+def load_datasets_from_file(filename):
+    with open(filename, 'r') as file:
+        datasets = [line.strip() for line in file.readlines() if line.strip()]
+    return datasets
+
+# Cargar conjuntos de datos desde un archivo de texto (datasets.txt)
+datasets = load_datasets_from_file("datasets.txt")
+
+# Crear clientes de repositorio para cada conjunto de datos
+repository_clients = []
+for dataset in datasets:
+    try:
+        repository_clients.append(HFRepository(dataset))
+    except Exception as e:
+        print(f"Error loading dataset {dataset}: {str(e)}")
+
+VERBOSE = False
+
+def format_prompt(message, history, cust_p):
+    prompt = ""
+    if history:
+        for user_prompt, bot_response in history:
+            prompt += f"<start_of_turn>user{user_prompt}<end_of_turn>"
+            prompt += f"<start_of_turn>model{bot_response}<end_of_turn>"
+            if VERBOSE:
+                print(prompt)
+    prompt += cust_p.replace("USER_INPUT", message)
+    return prompt
+
+def chat_inf(system_prompt, prompt, history, memory, client_choice, seed, temp, tokens, top_p, rep_p, chat_mem, cust_p):
+    if not clients or not repository_clients:
+        yield [("Error", "No models or datasets available")], memory
+    else:
+        try:
+            client = clients[int(client_choice) - 1]
+            hist_len = 0
+            if not history:
+                history = []
+            if not memory:
+                memory = []
+            if memory:
+                for ea in memory[0 - chat_mem:]:
+                    hist_len += len(str(ea))
+            in_len = len(system_prompt + prompt) + hist_len
+
+            if (in_len + tokens) > 500000:
+                history.append((prompt, "Wait, that's too many tokens, please reduce the 'Chat Memory' value, or reduce the 'Max new tokens' value"))
+                yield history, memory
+            else:
+                generate_kwargs = dict(
+                    temperature=temp,
+                    max_new_tokens=tokens,
+                    top_p=top_p,
+                    repetition_penalty=rep_p,
+                    do_sample=True,
+                    seed=seed,
+                )
+                if system_prompt:
+                    formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", memory[0 - chat_mem:], cust_p)
+                else:
+                    formatted_prompt = format_prompt(prompt, memory[0 - chat_mem:], cust_p)
+                stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
+                output = ""
+                for response in stream:
+                    output += response.token.text
+                    yield [(prompt, output)], memory
+                history.append((prompt, output))
+                memory.append((prompt, output))
+        except Exception as e:
+            yield [(prompt, f"Error: {str(e)}")], memory
+
+def get_screenshot(chat: list, height=5000, width=600, chatblock=[], theme="light", wait=3000, header=True):
+    tog = 0
+    if chatblock:
+        tog = 3
+    result = ss_client.predict(str(chat), height, width, chatblock, header, theme, wait, api_name="/run_script")
+    out = f'https://xilixmeaty40-testing.hf.space/file={result[tog]}'
+    return out
+
+def clear_fn():
+    return None, None, None, None
+
+rand_val = random.randint(1, 1111111111111111)
+
+def check_rand(inp, val):
+    if inp:
+        return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=random.randint(1, 1111111111111111))
+    else:
+        return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=int(val))
+
+with gr.Blocks() as app:
+    memory = gr.State()
+    gr.HTML("""<center><h1 style='font-size:xx-large;'>Google Gemma Models</h1><br><h3>running on Huggingface Inference Client</h3><br><h7>EXPERIMENTAL""")
+    chat_b = gr.Chatbot(height=500)
+    with gr.Group():
+        with gr.Row():
+            with gr.Column(scale=3):
+                inp = gr.Textbox(label="Prompt")
+                sys_inp = gr.Textbox(label="System Prompt (optional)")
+                with gr.Accordion("Prompt Format", open=False):
+                    custom_prompt = gr.Textbox(label="Modify Prompt Format", info="For testing purposes. 'USER_INPUT' is where 'SYSTEM_PROMPT, PROMPT' will be placed", lines=3, value="<start_of_turn>userUSER_INPUT<end_of_turn><start_of_turn>model")
+                with gr.Row():
+                    with gr.Column(scale=2):
+                        btn = gr.Button("Chat")
+                    with gr.Column(scale=1):
+                        with gr.Group():
+                            stop_btn = gr.Button("Stop")
+                            clear_btn = gr.Button("Clear")
+                client_choice = gr.Dropdown(label="Models", type='index', choices=[c for c in models], value=models[0], interactive=True)
+            with gr.Column(scale=1):
+                with gr.Group():
+                    rand = gr.Checkbox(label="Random Seed", value=True)
+                    seed = gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, step=1, value=rand_val)
+                    tokens = gr.Slider(label="Max new tokens", value=1600, minimum=0, maximum=500000, step=64, interactive=True, visible=True, info="The maximum number of tokens")
+                    temp = gr.Slider(label="Temperature", step=0.01, minimum=0.01, maximum=1.0, value=0.49)
+                    top_p = gr.Slider(label="Top-P", step=0.01, minimum=0.01, maximum=1.0, value=0.49)
+                    rep_p = gr.Slider(label="Repetition Penalty", step=0.01, minimum=0.1, maximum=2.0, value=0.99)
+                    chat_mem = gr.Number(label="Chat Memory", info="Number of previous chats to retain", value=4)
+        with gr.Accordion(label="Screenshot", open=False):
+            with gr.Row():
+                with gr.Column(scale=3):
+                    im_btn = gr.Button("Screenshot")
+                    img = gr.Image(type='filepath')
+                with gr.Column(scale=1):
+                    with gr.Row():
+                        im_height = gr.Number(label="Height", value=5000)
+                        im_width = gr.Number(label="Width", value=500)
+                    wait_time = gr.Number(label="Wait Time", value=3000)
+                    theme = gr.Radio(label="Theme", choices=["light", "dark"], value="light")
+                    chatblock = gr.Dropdown(label="Chatblocks", info="Choose specific blocks of chat", choices=[c for c in range(1, 40)], multiselect=True)
+
+    client_choice.change(load_models, client_choice, [chat_b])
+    app.load(load_models, client_choice, [chat_b])
+
+    im_go = im_btn.click(get_screenshot, [chat_b, im_height, im_width, chatblock, theme, wait_time], img)
+
+    chat_sub = inp.submit(check_rand, [rand, seed], seed).then(chat_inf, [sys_inp, inp, chat_b, memory, client_choice, seed, temp, tokens, top_p, rep_p, chat_mem, custom_prompt], [chat_b, memory])
+    go = btn.click(check_rand, [rand, seed], seed).then(chat_inf, [sys_inp, inp, chat_b, memory, client_choice, seed, temp, tokens, top_p, rep_p, chat_mem, custom_prompt], [chat_b, memory])
+
+    stop_btn.click(None, None, None, cancels=[go, im_go, chat_sub])
+    clear_btn.click(clear_fn, None, [inp, sys_inp, chat_b, memory])
+
+app.queue(default_concurrency_limit=10).launch()