Se modifico apra version 5

README.md

@@ -16,4 +16,4 @@ short_description: SipánGPT based on Llama-3.2-1B
 ---
 
 SipánGPT 0.3 Llama 3.2
-Entrenado con 5400 conversaciones.
+Entrenado con 50000 conversaciones.

app.py

@@ -1,12 +1,8 @@
 import os
+import torch
 from threading import Thread
 from typing import Iterator
-
 import gradio as gr
-from gradio.themes.base import Base
-from gradio.themes.utils import colors, sizes, fonts
-import time
-import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 
 DESCRIPTION = """\
@@ -16,12 +12,7 @@ This is a demo of [`meta-llama/Llama-3.2-3B-Instruct`](https://huggingface.co/me
 For more details, please check [our post](https://huggingface.co/blog/llama32).
 """
 
-MAX_MAX_NEW_TOKENS = 2048
-DEFAULT_MAX_NEW_TOKENS = 1024
-MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
-
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
-
+# Model setup
 model_id = "ussipan/SipanGPT-0.3-Llama-3.2-1B-GGUF"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForCausalLM.from_pretrained(
@@ -31,82 +22,48 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 model.eval()
 
-# Main Gradio inference function
 def generate(
     message: str,
-    chat_history: list[tuple[str, str]],
+    chat_history: list,
     max_new_tokens: int = 1024,
     temperature: float = 0.6,
-    top_p: float = 0.9,
-    top_k: int = 50,
-    repetition_penalty: float = 1.2,
 ) -> Iterator[str]:
+    conversation = chat_history + [{"role": "user", "content": message}]
+
+    input_ids = tokenizer.apply_chat_template(
+        conversation,
+        add_generation_prompt=True,
+        return_tensors="pt"
+    ).to(model.device)
+
+    streamer = TextIteratorStreamer(
+        tokenizer,
+        timeout=20.0,
+        skip_prompt=True,
+        skip_special_tokens=True
+    )
 
-    conversation = [{k: v for k, v in d.items() if k != 'metadata'} for d in chat_history]
-    conversation.append({"role": "user", "content": message})
-
-    input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt")
-    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
-        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
-        gr.Warning(f"Se recortó la entrada de la conversación porque era más larga que {MAX_INPUT_TOKEN_LENGTH} tokens.")
-    input_ids = input_ids.to(model.device)
-
-    streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
-    generate_kwargs = dict(
-        {"input_ids": input_ids},
+    generation_kwargs = dict(
+        input_ids=input_ids,
         streamer=streamer,
         max_new_tokens=max_new_tokens,
-        do_sample=True,
-        top_p=top_p,
-        top_k=top_k,
         temperature=temperature,
-        num_beams=1,
-        repetition_penalty=repetition_penalty,
+        do_sample=True,
     )
-    t = Thread(target=model.generate, kwargs=generate_kwargs)
-    t.start()
+
+    thread = Thread(target=model.generate, kwargs=generation_kwargs)
+    thread.start()
 
     conversation.append({"role": "assistant", "content": ""})
-    outputs = []
+    output = []
+
     for text in streamer:
-        outputs.append(text)
-        bot_response = "".join(outputs)
-        conversation[-1]['content'] = bot_response
+        output.append(text)
+        conversation[-1]["content"] = "".join(output)
         yield "", conversation
 
-
-# Implementing Gradio 5 features and building a ChatInterface UI yourself
-PLACEHOLDER = """<div style="padding: 20px; text-align: center; display: flex; flex-direction: column; align-items: center;">
-   <img src="https://corladlalibertad.org.pe/wp-content/uploads/2024/01/USS.jpg" style="width: 80%; max-width: 550px; height: auto; opacity: 0.55; margin-bottom: 10px;">
-   <h1 style="font-size: 28px; margin: 0;">SipánGPT 0.3 Llama 3.2</h1>
-   <p style="font-size: 8px; margin: 5px 0 0; opacity: 0.65;">
-       <a href="https://huggingface.co/spaces/ysharma/Llama3-2_with_Gradio-5" target="_blank" style="color: inherit; text-decoration: none;">Forked from @ysharma</a>
-   </p>
-   <p style="font-size: 12px; margin: 5px 0 0; opacity: 0.9;">Este modelo es experimental, puede generar alucinaciones o respuestas incorrectas.</p>
-   <p style="font-size: 12px; margin: 5px 0 0; opacity: 0.9;">Entrenado con un dataset de 50k conversaciones.</p>
-   <p style="font-size: 12px; margin: 5px 0 0; opacity: 0.9;">
-       <a href="https://huggingface.co/datasets/ussipan/sipangpt" target="_blank" style="color: inherit; text-decoration: none;">Ver el dataset aquí</a>
-   </p>
-</div>"""
-
-def handle_retry(history, retry_data: gr.RetryData):
-    new_history = history[:retry_data.index]
-    previous_prompt = history[retry_data.index]['content']
-    yield from generate(previous_prompt, chat_history = new_history, max_new_tokens = 1024, temperature = 0.6, top_p = 0.9, top_k = 50, repetition_penalty = 1.2)
-
 def handle_like(data: gr.LikeData):
-    if data.liked:
-        print("Votaste positivamente esta respuesta: ", data.value)
-    else:
-        print("Votaste negativamente esta respuesta: ", data.value)
-
-def handle_undo(history, undo_data: gr.UndoData):
-    chatbot = history[:undo_data.index]
-    prompt = history[undo_data.index]['content']
-    return chatbot, prompt
-
-def chat_examples_fill(data: gr.SelectData):
-    yield from generate(data.value['text'], chat_history = [], max_new_tokens = 1024, temperature = 0.6, top_p = 0.9, top_k = 50, repetition_penalty = 1.2)
+    print(f"El mensaje {data.index} fue puntuado como {'bueno' if data.liked else 'malo'}.")
 
 class SipanGPTTheme(Base):
     def __init__(
@@ -206,38 +163,47 @@ class SipanGPTTheme(Base):
             slider_color_dark="*primary_600",
         )
 
-# Uso del tema
 theme = SipanGPTTheme()
 
-with gr.Blocks(theme=theme, fill_height=True) as demo:
-    with gr.Column(elem_id="container", scale=1):
-        chatbot = gr.Chatbot(
-            label="SipánGPT 0.3 Llama 3.2",
-            show_label=False,
-            type="messages",
-            scale=1,
-            suggestions = [
-                {"text": "Háblame del reglamento de estudiantes de la universidad"},
-                {"text": "Qué becas ofrece la universidad"},
-                {"text": "Hablame sobre el temario del examen de admisión para pregrado"},
-                {"text": "Cuando se fundó la universidad?"},
-                ],
-            placeholder = PLACEHOLDER,
-            )
-
-    msg = gr.Textbox(submit_btn=True, show_label=False)
-    with gr.Accordion('Additional inputs', open=False):
-        max_new_tokens = gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS, )
-        temperature = gr.Slider(label="Temperature",minimum=0.1, maximum=4.0, step=0.1, value=0.6,)
-        top_p = gr.Slider(label="Top-p (nucleus sampling)", minimum=0.05, maximum=1.0, step=0.05, value=0.9, )
-        top_k = gr.Slider(label="Top-k", minimum=1, maximum=1000, step=1, value=50, )
-        repetition_penalty = gr.Slider(label="Repetition penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.2, )
-
-    msg.submit(generate, [msg, chatbot, max_new_tokens, temperature, top_p, top_k, repetition_penalty], [msg, chatbot])
-    chatbot.retry(handle_retry, chatbot, [msg, chatbot])
-    chatbot.like(handle_like, None, None)
-    chatbot.undo(handle_undo, chatbot, [chatbot, msg])
-    chatbot.suggestion_select(chat_examples_fill, None, [msg, chatbot] )
+with gr.Blocks(theme=theme, fill_height=True)  as demo:
+    chatbot = gr.Chatbot(
+        label="SipánGPT 0.3 Llama 3.2",
+        examples=[{"text": "Que carreras existen en la uss?"}, {"text": "Quien es el decano de la facultad de ingenieria?"}, {"text": "Que maestrias tiene la universidad?"}],
+        value=[],
+        show_label=True,
+        type="messages",
+        bubble_full_width=False,
+        placeholder = PLACEHOLDER,
+    )
+
+    msg = gr.Textbox(
+        show_label=False,
+        placeholder="Escribe tu pregunta aquí...",
+        scale=4
+    )
+
+    with gr.Row():
+        submit = gr.Button("Enviar")
+        clear = gr.ClearButton([msg, chatbot])
+
+    with gr.Accordion("Parameters", open=False):
+        temperature = gr.Slider(
+            minimum=0.1,
+            maximum=2.0,
+            value=0.6,
+            step=0.1,
+            label="Temperatura",
+        )
+        max_new_tokens = gr.Slider(
+            minimum=1,
+            maximum=2048,
+            value=1024,
+            step=1,
+            label="Máximo de nuevos Tokens",
+        )
 
+    msg.submit(generate, [msg, chatbot, max_new_tokens, temperature], [msg, chatbot])
+    submit.click(generate, [msg, chatbot, max_new_tokens, temperature], [msg, chatbot])
+    chatbot.like(handle_like)
 
 demo.launch()