Spaces:
Build error
Build error
Update app.py
Browse files
app.py
CHANGED
|
@@ -11,142 +11,86 @@ from io import BytesIO
|
|
| 11 |
import docx
|
| 12 |
import ntpath
|
| 13 |
|
| 14 |
-
###############################################################################
|
| 15 |
-
# REPLACE WITH YOUR DEFAULT (OPEN-SOURCE) DEEPSEEK API KEY
|
| 16 |
-
###############################################################################
|
| 17 |
-
OPEN_SOURCE_DEEPSEEK_KEY = "YOUR_DEFAULT_API_KEY_HERE"
|
| 18 |
|
| 19 |
-
def chat_with_deepseek(
|
| 20 |
-
user_message, # The user’s message to the chatbot
|
| 21 |
-
history, # Chat history (handled by Gradio’s ChatInterface)
|
| 22 |
-
system_message, # The system message from the left panel
|
| 23 |
-
user_api_key, # The user-provided optional HF API key
|
| 24 |
-
max_tokens, # The “Max new tokens” slider
|
| 25 |
-
temperature, # The “Temperature” slider
|
| 26 |
-
top_p # The “Top-p” slider
|
| 27 |
-
):
|
| 28 |
-
"""
|
| 29 |
-
This function is called every time a user submits a new message.
|
| 30 |
-
It uses either the user’s HF API key if provided, or a default
|
| 31 |
-
open-source DeepSeek API key if empty.
|
| 32 |
-
"""
|
| 33 |
-
|
| 34 |
-
# Decide which key to use
|
| 35 |
-
final_api_key = user_api_key.strip() if user_api_key else OPEN_SOURCE_DEEPSEEK_KEY
|
| 36 |
|
| 37 |
-
|
| 38 |
-
|
|
|
|
| 39 |
|
| 40 |
-
|
| 41 |
-
#
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
repetition_penalty=1.1, # If you want to add a penalty
|
| 54 |
)
|
| 55 |
-
|
| 56 |
-
#
|
| 57 |
-
partial_response = ""
|
| 58 |
stream = client.text_generation(
|
| 59 |
-
|
| 60 |
stream=True,
|
| 61 |
details=True,
|
| 62 |
-
**
|
| 63 |
)
|
| 64 |
-
|
| 65 |
-
|
|
|
|
|
|
|
| 66 |
continue
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
yield partial_response
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
###############################################################################
|
| 73 |
-
# BUILD THE GRADIO INTERFACE
|
| 74 |
-
###############################################################################
|
| 75 |
|
| 76 |
with gr.Blocks(theme="soft") as demo:
|
| 77 |
-
#
|
| 78 |
-
gr.Markdown(
|
| 79 |
-
"""
|
| 80 |
-
<h1 style="text-align:center; margin-bottom: 5px;">
|
| 81 |
-
<b>balianone.com</b>
|
| 82 |
-
</h1>
|
| 83 |
-
<h3 style="text-align:center; margin-top: 0px;">
|
| 84 |
-
Chat with DeepSeek-R1
|
| 85 |
-
</h3>
|
| 86 |
-
""",
|
| 87 |
-
elem_id="title"
|
| 88 |
-
)
|
| 89 |
|
| 90 |
with gr.Row():
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
type="password",
|
| 100 |
-
placeholder="Leave blank to use open-source DeepSeek key"
|
| 101 |
-
)
|
| 102 |
-
|
| 103 |
-
gr.Markdown(
|
| 104 |
-
"If error occurs, you can set your own API key. You’ll get R1 premium access."
|
| 105 |
-
)
|
| 106 |
-
|
| 107 |
-
max_tokens = gr.Slider(
|
| 108 |
-
minimum=1,
|
| 109 |
-
maximum=4000,
|
| 110 |
-
step=1,
|
| 111 |
-
value=4000,
|
| 112 |
-
label="Max new tokens"
|
| 113 |
-
)
|
| 114 |
-
temperature = gr.Slider(
|
| 115 |
-
minimum=0.1,
|
| 116 |
-
maximum=4.0,
|
| 117 |
-
value=0.7,
|
| 118 |
-
step=0.1,
|
| 119 |
-
label="Temperature"
|
| 120 |
-
)
|
| 121 |
-
top_p = gr.Slider(
|
| 122 |
-
minimum=0.1,
|
| 123 |
-
maximum=1.0,
|
| 124 |
-
value=0.95,
|
| 125 |
-
step=0.01,
|
| 126 |
-
label="Top-p (nucleus sampling)"
|
| 127 |
-
)
|
| 128 |
-
|
| 129 |
-
# Right Column (Chat interface)
|
| 130 |
-
with gr.Column(scale=3):
|
| 131 |
-
chatbot = gr.ChatInterface(
|
| 132 |
-
fn=chat_with_deepseek,
|
| 133 |
-
# Additional inputs that feed into your chat function:
|
| 134 |
-
additional_inputs=[system_message, user_api_key, max_tokens, temperature, top_p],
|
| 135 |
-
|
| 136 |
-
# Use the newer messages format to avoid warnings
|
| 137 |
-
type="messages",
|
| 138 |
-
|
| 139 |
-
# You can customize how the chat area looks here as well:
|
| 140 |
-
chatbot_base_url=None, # Keep local
|
| 141 |
-
height=550,
|
| 142 |
-
title="DeepSeek-R1 Assistant"
|
| 143 |
-
)
|
| 144 |
-
|
| 145 |
-
# Launch the app
|
| 146 |
-
demo.launch()
|
| 147 |
-
|
| 148 |
-
|
| 149 |
|
|
|
|
| 150 |
|
| 151 |
|
| 152 |
|
|
|
|
| 11 |
import docx
|
| 12 |
import ntpath
|
| 13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
+
# Initialize clients
|
| 17 |
+
API_KEY = os.environ.get("HF_API_KEY")
|
| 18 |
+
client = InferenceClient(token=API_KEY)
|
| 19 |
|
| 20 |
+
def process_file(filepath):
|
| 21 |
+
# Handle different file types and extract text.
|
| 22 |
+
if not filepath:
|
| 23 |
+
return ""
|
| 24 |
+
|
| 25 |
+
ext = os.path.splitext(filepath)[1].lower()
|
| 26 |
+
|
| 27 |
+
try:
|
| 28 |
+
if ext in ['.png', '.jpg', '.jpeg']:
|
| 29 |
+
image = Image.open(filepath)
|
| 30 |
+
text = pytesseract.image_to_string(image)
|
| 31 |
+
return f"IMAGE CONTENT:\n{text}"
|
| 32 |
+
elif ext == '.pdf':
|
| 33 |
+
pdf_reader = PyPDF2.PdfReader(filepath)
|
| 34 |
+
text = "\n".join(page.extract_text() for page in pdf_reader.pages)
|
| 35 |
+
return f"PDF CONTENT:\n{text}"
|
| 36 |
+
elif ext == '.docx':
|
| 37 |
+
doc = docx.Document(filepath)
|
| 38 |
+
text = "\n".join(para.text for para in doc.paragraphs)
|
| 39 |
+
return f"DOCUMENT CONTENT:\n{text}"
|
| 40 |
+
else:
|
| 41 |
+
return "Unsupported file type"
|
| 42 |
+
except Exception as e:
|
| 43 |
+
print(f"File processing error: {e}")
|
| 44 |
+
return "Error reading file"
|
| 45 |
+
|
| 46 |
+
def chat(message, history, filepath):
|
| 47 |
+
file_content = process_file(filepath) if filepath else ""
|
| 48 |
+
|
| 49 |
+
full_prompt = f"""
|
| 50 |
+
{file_content}
|
| 51 |
+
|
| 52 |
+
User Message: {message}
|
| 53 |
+
|
| 54 |
+
Please respond considering both the message and any attached documents:
|
| 55 |
+
"""
|
| 56 |
|
| 57 |
+
generate_kwargs = dict(
|
| 58 |
+
temperature=0.7,
|
| 59 |
+
max_new_tokens=2000,
|
| 60 |
+
top_p=0.95,
|
| 61 |
+
repetition_penalty=1.2,
|
|
|
|
| 62 |
)
|
| 63 |
+
|
| 64 |
+
# Generate response
|
|
|
|
| 65 |
stream = client.text_generation(
|
| 66 |
+
full_prompt,
|
| 67 |
stream=True,
|
| 68 |
details=True,
|
| 69 |
+
**generate_kwargs
|
| 70 |
)
|
| 71 |
+
|
| 72 |
+
partial_message = ""
|
| 73 |
+
for response in stream:
|
| 74 |
+
if response.token.special:
|
| 75 |
continue
|
| 76 |
+
partial_message += response.token.text
|
| 77 |
+
yield partial_message
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
with gr.Blocks(theme="soft") as demo:
|
| 80 |
+
gr.Markdown("# DeepSeek-R1 Assistant with File Support")
|
| 81 |
+
gr.Markdown("Upload images, PDFs, or docs and chat about them!")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 82 |
|
| 83 |
with gr.Row():
|
| 84 |
+
file_input = gr.File(label="Upload File (PDF/Image/Doc)", type="filepath")
|
| 85 |
+
|
| 86 |
+
chatbot = gr.ChatInterface(
|
| 87 |
+
fn=chat,
|
| 88 |
+
additional_inputs=[file_input],
|
| 89 |
+
type="messages",
|
| 90 |
+
examples=[]
|
| 91 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 92 |
|
| 93 |
+
demo.launch()
|
| 94 |
|
| 95 |
|
| 96 |
|