Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,158 +1,173 @@
|
|
| 1 |
import os
|
| 2 |
-
import torch
|
| 3 |
-
import gradio as gr
|
| 4 |
from pathlib import Path
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
from transformers import AutoConfig, AutoTokenizer
|
| 6 |
from optimum.intel.openvino import OVModelForCausalLM
|
| 7 |
-
from typing import List, Tuple
|
| 8 |
-
from threading import Event, Thread
|
| 9 |
-
from gradio_helper import make_demo # Your helper function for Gradio demo
|
| 10 |
-
from llm_config import SUPPORTED_LLM_MODELS # Model configuration
|
| 11 |
-
from notebook_utils import device_widget # Device selection utility
|
| 12 |
import openvino as ov
|
| 13 |
import openvino.properties as props
|
| 14 |
import openvino.properties.hint as hints
|
| 15 |
import openvino.properties.streams as streams
|
| 16 |
-
import
|
| 17 |
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
|
| 26 |
-
|
| 27 |
-
int4_model_dir = Path(model_id) / "INT4_compressed_weights"
|
| 28 |
-
if (int4_model_dir / "openvino_model.xml").exists():
|
| 29 |
-
return int4_model_dir
|
| 30 |
-
remote_code = model_configuration.get("remote_code", False)
|
| 31 |
-
export_command_base = f"optimum-cli export openvino --model {model_configuration['model_id']} --task text-generation-with-past --weight-format int4"
|
| 32 |
-
int4_compression_args = f" --group-size {model_compression_params['group_size']} --ratio {model_compression_params['ratio']}"
|
| 33 |
-
if model_compression_params["sym"]:
|
| 34 |
-
int4_compression_args += " --sym"
|
| 35 |
-
if enable_awq:
|
| 36 |
-
int4_compression_args += " --awq --dataset wikitext2 --num-samples 128"
|
| 37 |
-
export_command_base += int4_compression_args
|
| 38 |
-
if remote_code:
|
| 39 |
-
export_command_base += " --trust-remote-code"
|
| 40 |
-
export_command = export_command_base + f" {str(int4_model_dir)}"
|
| 41 |
|
| 42 |
-
|
| 43 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
return int4_model_dir
|
| 45 |
|
| 46 |
-
#
|
| 47 |
-
def load_model(
|
|
|
|
|
|
|
|
|
|
| 48 |
ov_config = {hints.performance_mode(): hints.PerformanceMode.LATENCY, streams.num(): "1", props.cache_dir(): ""}
|
| 49 |
core = ov.Core()
|
| 50 |
-
|
|
|
|
|
|
|
|
|
|
| 51 |
tok = AutoTokenizer.from_pretrained(model_dir, trust_remote_code=True)
|
| 52 |
-
|
| 53 |
ov_model = OVModelForCausalLM.from_pretrained(
|
| 54 |
model_dir,
|
| 55 |
-
device=device,
|
| 56 |
ov_config=ov_config,
|
| 57 |
config=AutoConfig.from_pretrained(model_dir, trust_remote_code=True),
|
| 58 |
-
trust_remote_code=True
|
| 59 |
)
|
|
|
|
|
|
|
| 60 |
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
def bot(history, temperature, top_p, top_k, repetition_penalty, conversation_id):
|
| 65 |
-
input_ids = convert_history_to_token(history)
|
| 66 |
-
if input_ids.shape[1] > 2000:
|
| 67 |
-
history = [history[-1]] # Limit input size
|
| 68 |
-
input_ids = convert_history_to_token(history)
|
| 69 |
|
| 70 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 72 |
generate_kwargs = dict(
|
| 73 |
input_ids=input_ids,
|
| 74 |
max_new_tokens=256,
|
| 75 |
temperature=temperature,
|
| 76 |
-
do_sample=temperature > 0.0,
|
| 77 |
top_p=top_p,
|
| 78 |
top_k=top_k,
|
| 79 |
repetition_penalty=repetition_penalty,
|
| 80 |
-
streamer=streamer
|
| 81 |
)
|
|
|
|
|
|
|
|
|
|
| 82 |
|
| 83 |
-
# Function to generate response in a separate thread
|
| 84 |
def generate_and_signal_complete():
|
| 85 |
ov_model.generate(**generate_kwargs)
|
| 86 |
-
|
| 87 |
|
| 88 |
t1 = Thread(target=generate_and_signal_complete)
|
| 89 |
t1.start()
|
| 90 |
|
| 91 |
-
#
|
| 92 |
partial_text = ""
|
| 93 |
for new_text in streamer:
|
| 94 |
-
partial_text
|
| 95 |
history[-1][1] = partial_text
|
| 96 |
yield history
|
| 97 |
|
| 98 |
-
#
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
# Create the Gradio chatbot interface
|
| 127 |
-
chatbot = gr.Chatbot()
|
| 128 |
-
|
| 129 |
-
# Parameters for bot generation
|
| 130 |
-
temperature = gr.Slider(minimum=0, maximum=1, step=0.1, label="Temperature", value=0.7)
|
| 131 |
-
top_p = gr.Slider(minimum=0, maximum=1, step=0.1, label="Top-p", value=0.9)
|
| 132 |
-
top_k = gr.Slider(minimum=0, maximum=50, step=1, label="Top-k", value=50)
|
| 133 |
-
repetition_penalty = gr.Slider(minimum=0, maximum=2, step=0.1, label="Repetition Penalty", value=1.0)
|
| 134 |
-
|
| 135 |
-
with gr.Blocks() as demo:
|
| 136 |
-
# Create the Gradio components and add them to the Blocks context
|
| 137 |
-
model_id_selector.change(update_model_ids, inputs=model_language, outputs=model_id_selector)
|
| 138 |
-
load_button = gr.Button("Load Model")
|
| 139 |
-
load_button.click(load_model_on_select, inputs=[model_language, model_id, enable_awq], outputs=[gr.Textbox(label="Model Status")])
|
| 140 |
-
|
| 141 |
-
# Set up the chatbot UI with all the required components
|
| 142 |
-
gr.Row([model_id_selector, enable_awq]) # Arrange the dropdowns and checkbox in a row
|
| 143 |
-
gr.Row([load_button]) # Add the button below the inputs
|
| 144 |
-
gr.Row([chatbot]) # Add the chatbot output
|
| 145 |
-
|
| 146 |
-
# Parameters for generation
|
| 147 |
-
gr.Row([temperature, top_p, top_k, repetition_penalty]) # Add sliders in a row
|
| 148 |
-
|
| 149 |
-
# Define bot function and run the interface
|
| 150 |
-
demo.queue() # This is used to queue inputs and outputs, handling concurrent generation calls
|
| 151 |
-
demo.launch(debug=True, share=True) # For public access
|
| 152 |
-
|
| 153 |
-
return demo
|
| 154 |
-
|
| 155 |
-
# Run the Gradio app
|
| 156 |
if __name__ == "__main__":
|
| 157 |
-
|
| 158 |
-
app.launch(debug=True, share=True) # share=True for public access
|
|
|
|
| 1 |
import os
|
|
|
|
|
|
|
| 2 |
from pathlib import Path
|
| 3 |
+
import requests
|
| 4 |
+
import shutil
|
| 5 |
+
import torch
|
| 6 |
+
from threading import Event, Thread
|
| 7 |
from transformers import AutoConfig, AutoTokenizer
|
| 8 |
from optimum.intel.openvino import OVModelForCausalLM
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
import openvino as ov
|
| 10 |
import openvino.properties as props
|
| 11 |
import openvino.properties.hint as hints
|
| 12 |
import openvino.properties.streams as streams
|
| 13 |
+
import gradio as gr
|
| 14 |
|
| 15 |
+
from llm_config import SUPPORTED_LLM_MODELS
|
| 16 |
+
from notebook_utils import device_widget
|
| 17 |
+
|
| 18 |
+
# Initialize model language options
|
| 19 |
+
model_languages = list(SUPPORTED_LLM_MODELS)
|
| 20 |
+
|
| 21 |
+
# Gradio components for selecting model language and model ID
|
| 22 |
+
model_language = gr.Dropdown(
|
| 23 |
+
choices=model_languages,
|
| 24 |
+
value=model_languages[0],
|
| 25 |
+
label="Model Language"
|
| 26 |
+
)
|
| 27 |
+
|
| 28 |
+
# Gradio dropdown for selecting model ID based on language
|
| 29 |
+
def update_model_id(model_language_value):
|
| 30 |
+
model_ids = list(SUPPORTED_LLM_MODELS[model_language_value])
|
| 31 |
+
return model_ids[0], gr.update(choices=model_ids)
|
| 32 |
+
|
| 33 |
+
model_id = gr.Dropdown(
|
| 34 |
+
choices=[], # will be dynamically populated
|
| 35 |
+
label="Model",
|
| 36 |
+
value=None
|
| 37 |
+
)
|
| 38 |
+
|
| 39 |
+
model_language.change(update_model_id, inputs=model_language, outputs=[model_id])
|
| 40 |
+
|
| 41 |
+
# Gradio checkbox for preparing INT4 model
|
| 42 |
+
prepare_int4_model = gr.Checkbox(
|
| 43 |
+
value=True,
|
| 44 |
+
label="Prepare INT4 Model"
|
| 45 |
+
)
|
| 46 |
+
|
| 47 |
+
# Gradio checkbox for enabling AWQ (depends on INT4 checkbox)
|
| 48 |
+
enable_awq = gr.Checkbox(
|
| 49 |
+
value=False,
|
| 50 |
+
label="Enable AWQ",
|
| 51 |
+
visible=False
|
| 52 |
+
)
|
| 53 |
+
|
| 54 |
+
# Device selection widget (e.g., CPU or GPU)
|
| 55 |
+
device = device_widget("CPU", exclude=["NPU"])
|
| 56 |
+
|
| 57 |
+
# Model directory and setup based on selections
|
| 58 |
+
def get_model_path(model_language_value, model_id_value):
|
| 59 |
+
model_configuration = SUPPORTED_LLM_MODELS[model_language_value][model_id_value]
|
| 60 |
+
pt_model_id = model_configuration["model_id"]
|
| 61 |
+
pt_model_name = model_id_value.split("-")[0]
|
| 62 |
+
int4_model_dir = Path(model_id_value) / "INT4_compressed_weights"
|
| 63 |
+
return model_configuration, int4_model_dir, pt_model_name
|
| 64 |
+
|
| 65 |
+
# Function to download the model if not already present
|
| 66 |
+
def download_model_if_needed(model_language_value, model_id_value):
|
| 67 |
+
model_configuration, int4_model_dir, pt_model_name = get_model_path(model_language_value, model_id_value)
|
| 68 |
|
| 69 |
+
int4_weights = int4_model_dir / "openvino_model.bin"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
|
| 71 |
+
if not int4_weights.exists():
|
| 72 |
+
print(f"Downloading model {model_id_value}...")
|
| 73 |
+
# Add your download logic here (e.g., from a URL)
|
| 74 |
+
# Example:
|
| 75 |
+
# r = requests.get(model_configuration["model_url"])
|
| 76 |
+
# with open(int4_weights, "wb") as f:
|
| 77 |
+
# f.write(r.content)
|
| 78 |
+
|
| 79 |
return int4_model_dir
|
| 80 |
|
| 81 |
+
# Load the model
|
| 82 |
+
def load_model(model_language_value, model_id_value):
|
| 83 |
+
int4_model_dir = download_model_if_needed(model_language_value, model_id_value)
|
| 84 |
+
|
| 85 |
+
# Load the OpenVINO model
|
| 86 |
ov_config = {hints.performance_mode(): hints.PerformanceMode.LATENCY, streams.num(): "1", props.cache_dir(): ""}
|
| 87 |
core = ov.Core()
|
| 88 |
+
|
| 89 |
+
model_dir = int4_model_dir
|
| 90 |
+
model_configuration = SUPPORTED_LLM_MODELS[model_language_value][model_id_value]
|
| 91 |
+
|
| 92 |
tok = AutoTokenizer.from_pretrained(model_dir, trust_remote_code=True)
|
|
|
|
| 93 |
ov_model = OVModelForCausalLM.from_pretrained(
|
| 94 |
model_dir,
|
| 95 |
+
device=device.value,
|
| 96 |
ov_config=ov_config,
|
| 97 |
config=AutoConfig.from_pretrained(model_dir, trust_remote_code=True),
|
| 98 |
+
trust_remote_code=True
|
| 99 |
)
|
| 100 |
+
|
| 101 |
+
return tok, ov_model, model_configuration
|
| 102 |
|
| 103 |
+
# Gradio interface function for generating text responses
|
| 104 |
+
def generate_response(history, temperature, top_p, top_k, repetition_penalty, model_language_value, model_id_value):
|
| 105 |
+
tok, ov_model, model_configuration = load_model(model_language_value, model_id_value)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
|
| 107 |
+
# Convert history to tokens
|
| 108 |
+
def convert_history_to_token(history):
|
| 109 |
+
# (Your history conversion logic here)
|
| 110 |
+
# Use model_configuration to determine the exact format
|
| 111 |
+
input_tokens = tok(" ".join([msg[0] for msg in history]), return_tensors="pt").input_ids
|
| 112 |
+
return input_tokens
|
| 113 |
|
| 114 |
+
input_ids = convert_history_to_token(history)
|
| 115 |
+
streamer = gr.Textbox.update()
|
| 116 |
+
|
| 117 |
+
# Adjust generation kwargs
|
| 118 |
generate_kwargs = dict(
|
| 119 |
input_ids=input_ids,
|
| 120 |
max_new_tokens=256,
|
| 121 |
temperature=temperature,
|
|
|
|
| 122 |
top_p=top_p,
|
| 123 |
top_k=top_k,
|
| 124 |
repetition_penalty=repetition_penalty,
|
| 125 |
+
streamer=streamer
|
| 126 |
)
|
| 127 |
+
|
| 128 |
+
# Start streaming response
|
| 129 |
+
event = Event()
|
| 130 |
|
|
|
|
| 131 |
def generate_and_signal_complete():
|
| 132 |
ov_model.generate(**generate_kwargs)
|
| 133 |
+
event.set()
|
| 134 |
|
| 135 |
t1 = Thread(target=generate_and_signal_complete)
|
| 136 |
t1.start()
|
| 137 |
|
| 138 |
+
# Collect generated text
|
| 139 |
partial_text = ""
|
| 140 |
for new_text in streamer:
|
| 141 |
+
partial_text += new_text
|
| 142 |
history[-1][1] = partial_text
|
| 143 |
yield history
|
| 144 |
|
| 145 |
+
# Gradio UI components
|
| 146 |
+
temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.7, label="Temperature")
|
| 147 |
+
top_p = gr.Slider(minimum=0.0, maximum=1.0, value=0.9, label="Top P")
|
| 148 |
+
top_k = gr.Slider(minimum=0, maximum=50, value=50, label="Top K")
|
| 149 |
+
repetition_penalty = gr.Slider(minimum=1.0, maximum=2.0, value=1.1, label="Repetition Penalty")
|
| 150 |
+
|
| 151 |
+
# Conversation history input/output
|
| 152 |
+
history = gr.State([]) # store the conversation history
|
| 153 |
+
|
| 154 |
+
# Gradio Interface
|
| 155 |
+
iface = gr.Interface(
|
| 156 |
+
fn=generate_response,
|
| 157 |
+
inputs=[
|
| 158 |
+
history,
|
| 159 |
+
temperature,
|
| 160 |
+
top_p,
|
| 161 |
+
top_k,
|
| 162 |
+
repetition_penalty,
|
| 163 |
+
model_language,
|
| 164 |
+
model_id
|
| 165 |
+
],
|
| 166 |
+
outputs=[gr.Textbox(label="Conversation History")],
|
| 167 |
+
live=True,
|
| 168 |
+
title="OpenVINO Chatbot"
|
| 169 |
+
)
|
| 170 |
+
|
| 171 |
+
# Launch Gradio app
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 172 |
if __name__ == "__main__":
|
| 173 |
+
iface.launch(debug=True, share=True, server_name="0.0.0.0", server_port=7860)
|
|
|