Update app.py

app.py

@@ -1,78 +1,162 @@
 import gradio as gr
-from transformers import AutoModelForCausalLM, AutoTokenizer
-import torch
+from ctransformers import AutoModelForCausalLM
+
+model_name = "lmstudio-community/Devstral-Small-2505-GGUF"
+model_file = "devstral-small-2505.Q4_K_M.gguf"  # Выберем версию с квантизацией Q4_K_M для экономии памяти
 
-model_name = "mistralai/Devstral-Small-2505"
-tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1", use_fast=False)
 model = AutoModelForCausalLM.from_pretrained(
-    model_name, 
-    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
+    model_name,
+    model_file=model_file,
+    model_type="mistral",
+    gpu_layers=50,  # Используем GPU насколько возможно
+    context_length=4096  # Максимальный контекст
 )
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model = model.to(device)
-
-def code_completion(prompt, max_new_tokens=128, temperature=0.2):
-    if not prompt.strip():
-        return "Please enter some code to complete."
-    inputs = tokenizer(prompt, return_tensors="pt").to(device)
-    with torch.no_grad():
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=max_new_tokens,
-            temperature=temperature,
-            do_sample=True,
-            pad_token_id=tokenizer.eos_token_id
-        )
-    generated = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    return generated[len(prompt):]
+
+def generate_text(prompt, max_tokens=512, temperature=0.7, top_p=0.9):
+    # Форматируем запрос в стиле Mistral
+    formatted_prompt = f"<s>[INST] {prompt} [/INST]"
+    
+    # Генерируем ответ
+    response = model(
+        formatted_prompt,
+        max_new_tokens=max_tokens,
+        temperature=temperature,
+        top_p=top_p,
+        repetition_penalty=1.1,
+        stream=False
+    )
+    
+    # Удаляем исходный запрос из ответа
+    return response.replace(formatted_prompt, "").strip()
 
 custom_css = """
-body {background: #f7f8fa;}
-.gradio-container {background: #f7f8fa;}
-h1, h2, h3, h4, h5, h6 {font-family: 'Inter', sans-serif;}
-#main-title {
+:root {
+    --primary-color: #4F46E5;
+    --secondary-color: #6366F1;
+    --background-color: #F9FAFB;
+    --surface-color: #FFFFFF;
+    --text-color: #1F2937;
+    --border-radius: 10px;
+}
+
+body {
+    background-color: var(--background-color);
+}
+
+.container {
+    max-width: 900px;
+    margin: auto;
+    padding-top: 1.5rem;
+}
+
+.title {
     text-align: center;
-    font-weight: 800;
-    font-size: 2.3em;
-    margin-bottom: 0.2em;
-    letter-spacing: -1px;
-    color: #222;
+    color: var(--primary-color);
+    font-size: 2.2rem;
+    font-weight: 700;
+    margin-bottom: 0.5rem;
 }
-#subtitle {
+
+.subtitle {
     text-align: center;
-    color: #6c6f7a;
-    font-size: 1.1em;
-    margin-bottom: 2em;
+    color: var(--text-color);
+    opacity: 0.8;
+    margin-bottom: 2rem;
+}
+
+footer {display: none !important;}
+
+.gradio-container {
+    font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif;
+}
+
+.gr-button {
+    border-radius: var(--border-radius) !important;
+}
+
+.gr-button-primary {
+    background-color: var(--primary-color) !important;
+}
+
+.gr-input, .gr-textarea {
+    border-radius: var(--border-radius) !important;
+    border: 1px solid #E5E7EB !important;
+}
+
+.gr-box {
+    border-radius: var(--border-radius) !important;
+    box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1), 0 2px 4px -1px rgba(0, 0, 0, 0.06) !important;
+    background-color: var(--surface-color) !important;
+    padding: 1.5rem !important;
+}
+
+.advanced-options {
+    margin-top: 1rem;
+    padding: 1rem;
+    border-radius: var(--border-radius);
+    background: #F3F4F6;
+}
+
+.footer-text {
+    text-align: center;
+    margin-top: 1rem;
+    color: var(--text-color);
+    opacity: 0.7;
+    font-size: 0.9rem;
 }
-.gr-box {border-radius: 16px;}
 """
 
-with gr.Blocks(css=custom_css, theme=gr.themes.Soft()) as demo:
-    gr.Markdown(
-        """
-        <h1 id="main-title">Devstral Code Autocomplete</h1>
-        <div id="subtitle">Minimal, beautiful code completion powered by <b>Devstral</b></div>
-        """)
-    with gr.Row():
-        with gr.Column(scale=1):
+with gr.Blocks(css=custom_css) as demo:
+    with gr.Column(elem_classes="container"):
+        gr.Markdown("# Devstral Code Assistant", elem_classes="title")
+        gr.Markdown("Powered by Devstral-Small-2505 - Specialized for code generation", elem_classes="subtitle")
+        
+        with gr.Box():
             prompt = gr.Textbox(
-                label="Your code prompt",
-                lines=10,
-                placeholder="def quicksort(arr):\n    \"\"\"Sort the array using quicksort algorithm.\"\"\"\n    if len(arr) <= 1:\n        return arr\n    pivot = arr[len(arr) // 2]\n    ",
-                show_copy_button=True,
-                autofocus=True
+                placeholder="Write a function in Python to implement a binary search tree", 
+                label="Your Request",
+                lines=5
             )
+            
             with gr.Row():
-                max_tokens = gr.Slider(16, 256, value=128, step=8, label="Max new tokens")
-                temperature = gr.Slider(0.1, 1.0, value=0.2, step=0.05, label="Temperature")
-            btn = gr.Button("Generate Completion", elem_id="generate-btn")
-        with gr.Column(scale=1):
-            output = gr.Code(
-                label="Generated code",
-                language="python",
+                submit_btn = gr.Button("Generate Code", variant="primary", scale=2)
+                clear_btn = gr.Button("Clear", scale=1)
+            
+            with gr.Accordion("Advanced Settings", open=False):
+                with gr.Row():
+                    with gr.Column():
+                        max_tokens = gr.Slider(
+                            minimum=64, maximum=2048, value=512, step=64,
+                            label="Maximum Output Length"
+                        )
+                    with gr.Column():
+                        temperature = gr.Slider(
+                            minimum=0.1, maximum=1.0, value=0.7, step=0.1,
+                            label="Temperature (Creativity)"
+                        )
+                with gr.Row():
+                    with gr.Column():
+                        top_p = gr.Slider(
+                            minimum=0.1, maximum=1.0, value=0.9, step=0.05,
+                            label="Top-p (Nucleus Sampling)"
+                        )
+            
+            output = gr.Textbox(
+                label="Generated Code", 
                 lines=12,
-                interactive=False
+                show_copy_button=True
             )
-    btn.click(code_completion, inputs=[prompt, max_tokens, temperature], outputs=output)
+        
+        gr.Markdown(
+            "⚡ Optimized for code generation and technical tasks", 
+            elem_classes="footer-text"
+        )
+    
+    submit_btn.click(
+        generate_text, 
+        inputs=[prompt, max_tokens, temperature, top_p], 
+        outputs=output
+    )
+    clear_btn.click(lambda: "", None, prompt)
 
 demo.launch()