Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -10,16 +10,17 @@ import ollama
|
|
| 10 |
# List of available models for selection.
|
| 11 |
# IMPORTANT: These names must correspond to models that have been either
|
| 12 |
|
|
|
|
| 13 |
# Model from run.sh
|
| 14 |
-
|
| 15 |
-
'hf.co/bartowski/Qwen_Qwen3-4B-Instruct-2507-GGUF:Q4_K_M',
|
| 16 |
-
#'hf.co/bartowski/Qwen_Qwen3-4B-Thinking-2507-GGUF:Q4_K_M',
|
| 17 |
-
'smollm2:360m-instruct-q5_K_M',
|
| 18 |
-
'hf.co/bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M', # OK speed with CPU
|
| 19 |
-
#'gemma3n:e2b-it-q4_K_M',
|
| 20 |
-
'granite3.3:2b',
|
| 21 |
-
'hf.co/bartowski/tencent_Hunyuan-4B-Instruct-GGUF:Q4_K_M'
|
| 22 |
-
|
| 23 |
|
| 24 |
|
| 25 |
# Default System Prompt
|
|
@@ -36,9 +37,9 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue", secondary_hue="neutra
|
|
| 36 |
|
| 37 |
# Model Selection
|
| 38 |
with gr.Row():
|
| 39 |
-
|
| 40 |
-
choices=
|
| 41 |
-
value=
|
| 42 |
label="Select Model",
|
| 43 |
info="Choose the LLM model to chat with.",
|
| 44 |
interactive=True
|
|
@@ -75,7 +76,10 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue", secondary_hue="neutra
|
|
| 75 |
|
| 76 |
# --- New: System Prompt Options ---
|
| 77 |
SYSTEM_PROMPT_OPTIONS = {
|
| 78 |
-
"Smart & Accurate (
|
|
|
|
|
|
|
|
|
|
| 79 |
"Friendly & Conversational": """Respond in a warm, friendly, and engaging tone. Use natural language and offer helpful suggestions. Keep responses concise but personable.""",
|
| 80 |
"Professional & Formal": """Maintain a formal and professional tone. Use precise language, avoid slang, and ensure responses are suitable for business or academic contexts.""",
|
| 81 |
"Elon Musk style": "You must chat in Elon Musk style!"
|
|
@@ -123,13 +127,15 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue", secondary_hue="neutra
|
|
| 123 |
|
| 124 |
# --- Core Chat Logic ---
|
| 125 |
# This function is the heart of the application.
|
| 126 |
-
def respond(history, system_prompt, stream_output,
|
| 127 |
"""
|
| 128 |
This is the single function that handles the entire chat process.
|
| 129 |
It takes the history, prepends the system prompt, calls the Ollama API,
|
| 130 |
and streams the response back to the chatbot.
|
| 131 |
"""
|
| 132 |
|
|
|
|
|
|
|
| 133 |
#Disable Qwen3 thinking
|
| 134 |
if "Qwen3".lower() in current_selected_model:
|
| 135 |
system_prompt = system_prompt+" /no_think"
|
|
@@ -176,9 +182,9 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue", secondary_hue="neutra
|
|
| 176 |
queue=False
|
| 177 |
).then(
|
| 178 |
respond,
|
| 179 |
-
inputs=[chatbot, system_prompt_textbox, stream_checkbox,
|
| 180 |
outputs=[chatbot]
|
| 181 |
)
|
| 182 |
|
| 183 |
# Launch the Gradio interface
|
| 184 |
-
demo.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
| 10 |
# List of available models for selection.
|
| 11 |
# IMPORTANT: These names must correspond to models that have been either
|
| 12 |
|
| 13 |
+
|
| 14 |
# Model from run.sh
|
| 15 |
+
MODEL_ID_MAP = {
|
| 16 |
+
"Qwen3-4B-Instruct-2507": 'hf.co/bartowski/Qwen_Qwen3-4B-Instruct-2507-GGUF:Q4_K_M',
|
| 17 |
+
#"Qwen3-4B-Thinking-2507": 'hf.co/bartowski/Qwen_Qwen3-4B-Thinking-2507-GGUF:Q4_K_M',
|
| 18 |
+
"SmolLM2-360M": 'smollm2:360m-instruct-q5_K_M',
|
| 19 |
+
"Llama3.2-3B-Instruct": 'hf.co/bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M', # OK speed with CPU
|
| 20 |
+
#"Gemma3n-e2b-it": 'gemma3n:e2b-it-q4_K_M',
|
| 21 |
+
"Granite3.3-2B": 'granite3.3:2b',
|
| 22 |
+
"Hunyuan-4B-Instruct": 'hf.co/bartowski/tencent_Hunyuan-4B-Instruct-GGUF:Q4_K_M'
|
| 23 |
+
}
|
| 24 |
|
| 25 |
|
| 26 |
# Default System Prompt
|
|
|
|
| 37 |
|
| 38 |
# Model Selection
|
| 39 |
with gr.Row():
|
| 40 |
+
selected_model_label = gr.Radio(
|
| 41 |
+
choices=list(MODEL_ID_MAP.keys()),
|
| 42 |
+
value=list(MODEL_ID_MAP.keys())[0], # Default to first display name
|
| 43 |
label="Select Model",
|
| 44 |
info="Choose the LLM model to chat with.",
|
| 45 |
interactive=True
|
|
|
|
| 76 |
|
| 77 |
# --- New: System Prompt Options ---
|
| 78 |
SYSTEM_PROMPT_OPTIONS = {
|
| 79 |
+
"Smart & Accurate (Auto TC/EN)": DEFAULT_SYSTEM_PROMPT,
|
| 80 |
+
"繁體中文回答":"無論如何,必須使用標準繁體中文回答. Answer everything in simple, smart, relevant and accurate style. No chatty!",
|
| 81 |
+
"简体中文回答":"无论如何,必须使用标准简体中文回答. Answer everything in simple, smart, relevant and accurate style. No chatty!",
|
| 82 |
+
"English Caht":"You must reply by English. Answer everything in simple, smart, relevant and accurate style. No chatty!",
|
| 83 |
"Friendly & Conversational": """Respond in a warm, friendly, and engaging tone. Use natural language and offer helpful suggestions. Keep responses concise but personable.""",
|
| 84 |
"Professional & Formal": """Maintain a formal and professional tone. Use precise language, avoid slang, and ensure responses are suitable for business or academic contexts.""",
|
| 85 |
"Elon Musk style": "You must chat in Elon Musk style!"
|
|
|
|
| 127 |
|
| 128 |
# --- Core Chat Logic ---
|
| 129 |
# This function is the heart of the application.
|
| 130 |
+
def respond(history, system_prompt, stream_output, selected_model_name, selected_prompt_key, use_custom_prompt): # Added selected_model_name
|
| 131 |
"""
|
| 132 |
This is the single function that handles the entire chat process.
|
| 133 |
It takes the history, prepends the system prompt, calls the Ollama API,
|
| 134 |
and streams the response back to the chatbot.
|
| 135 |
"""
|
| 136 |
|
| 137 |
+
current_selected_model = MODEL_ID_MAP[selected_model_name]
|
| 138 |
+
|
| 139 |
#Disable Qwen3 thinking
|
| 140 |
if "Qwen3".lower() in current_selected_model:
|
| 141 |
system_prompt = system_prompt+" /no_think"
|
|
|
|
| 182 |
queue=False
|
| 183 |
).then(
|
| 184 |
respond,
|
| 185 |
+
inputs=[chatbot, system_prompt_textbox, stream_checkbox, selected_model_label, system_prompt_selector, use_custom_prompt_checkbox], # Pass new inputs
|
| 186 |
outputs=[chatbot]
|
| 187 |
)
|
| 188 |
|
| 189 |
# Launch the Gradio interface
|
| 190 |
+
demo.launch(server_name="0.0.0.0", server_port=7860)
|