File size: 8,390 Bytes
7a17876 7cd6401 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 |
---
license: apache-2.0
pipeline_tag: text-generation
library_name: transformers
---
# AMD-135m
## Introduction
AMD-Llama-135m is a language model trained on AMD MI250 GPUs. Based on LLaMA2 model architecture, this model can be smoothly loaded as LlamaForCausalLM with huggingface transformers. Furthermore, we use the same tokenizer as LLaMA2, enabling it to be a draft model of speculative decoding for LLaMA2 and CodeLlama.
```python
import sys
import os
from PyQt5.QtWidgets import (QApplication, QWidget, QLabel, QPushButton,
QLineEdit, QTextEdit, QVBoxLayout, QHBoxLayout,
QFileDialog, QProgressBar, QMessageBox, QMenu)
from PyQt5.QtCore import Qt, QThread, pyqtSignal
from llama_cpp import Llama
class Worker(QThread):
finished = pyqtSignal(str)
progress = pyqtSignal(int, int) # Pass total tokens as well
def __init__(self, model, messages, max_tokens):
super().__init__()
self.model = model
self.messages = messages
self.max_tokens = max_tokens
def run(self):
try:
response = self.model.create_chat_completion(
messages=self.messages,
max_tokens=self.max_tokens,
temperature=0.7,
stream=True
)
total_tokens = 0
full_response = ""
for chunk in response:
if "choices" in chunk:
content = chunk["choices"][0]["delta"].get("content", "")
full_response += content
total_tokens += 1 # Assume each chunk is 1 token (adjust if needed)
self.progress.emit(total_tokens, self.max_tokens)
self.finished.emit(full_response)
except Exception as e:
self.finished.emit(f"Error generating response: {str(e)}")
class ChatbotGUI(QWidget):
def __init__(self):
super().__init__()
self.setWindowTitle("Chatbot GUI")
self.resize(800, 600)
self.model = None
self.messages = [
{"role": "system", "content": "You are a helpful AI assistant."}
]
self.initUI()
def initUI(self):
# Model loading section
model_label = QLabel("Model: No model loaded")
load_button = QPushButton("Load GGUF Model")
load_button.clicked.connect(self.load_model)
model_layout = QHBoxLayout()
model_layout.addWidget(model_label)
model_layout.addWidget(load_button)
# Chat display
self.chat_display = QTextEdit()
self.chat_display.setReadOnly(True)
self.chat_display.setContextMenuPolicy(Qt.CustomContextMenu)
self.chat_display.customContextMenuRequested.connect(self.show_context_menu)
# User input
self.user_input = QLineEdit()
self.user_input.returnPressed.connect(self.send_message)
send_button = QPushButton("Send")
send_button.clicked.connect(self.send_message)
input_layout = QHBoxLayout()
input_layout.addWidget(self.user_input)
input_layout.addWidget(send_button)
# Progress bar
self.progress_bar = QProgressBar()
self.progress_bar.hide()
# Clear conversation button
clear_button = QPushButton("Clear Conversation")
clear_button.clicked.connect(self.clear_conversation)
# Main layout
main_layout = QVBoxLayout()
main_layout.addLayout(model_layout)
main_layout.addWidget(self.chat_display)
main_layout.addWidget(self.progress_bar)
main_layout.addLayout(input_layout)
main_layout.addWidget(clear_button)
self.setLayout(main_layout)
def load_model(self):
model_path, _ = QFileDialog.getOpenFileName(self, "Load GGUF Model", "", "GGUF Files (*.gguf)")
if model_path:
try:
self.model = Llama(model_path=model_path, n_ctx=2048, n_gpu_layers=-1)
model_name = os.path.basename(model_path)
self.layout().itemAt(0).itemAt(0).widget().setText(f"Model: {model_name}")
QMessageBox.information(self, "Success", "Model loaded successfully!")
except Exception as e:
error_message = f"Error loading model: {str(e)}"
QMessageBox.critical(self, "Error", error_message)
def send_message(self):
user_message = self.user_input.text()
if user_message and self.model:
self.messages.append({"role": "user", "content": user_message})
self.update_chat_display(f"You: {user_message}")
self.user_input.clear()
max_tokens = 500 # Set your desired max tokens here
self.progress_bar.show()
self.progress_bar.setRange(0, max_tokens)
self.progress_bar.setValue(0)
self.worker = Worker(self.model, self.messages, max_tokens)
self.worker.finished.connect(self.on_response_finished)
self.worker.progress.connect(self.on_response_progress)
self.worker.start()
def on_response_finished(self, assistant_message):
self.progress_bar.hide()
self.messages.append({"role": "assistant", "content": assistant_message})
self.update_chat_display(f"Assistant: {assistant_message}")
# Python Code Download (Check for triple backticks)
if assistant_message.startswith("```python") and assistant_message.endswith("```"):
self.offer_code_download(assistant_message)
def on_response_progress(self, current_tokens, total_tokens):
self.progress_bar.setValue(current_tokens)
def offer_code_download(self, code):
reply = QMessageBox.question(self, "Download Code",
"The assistant generated Python code. Do you want to download it?",
QMessageBox.Yes | QMessageBox.No)
if reply == QMessageBox.Yes:
file_path, _ = QFileDialog.getSaveFileName(self, "Save Python Code", "code.py", "Python Files (*.py)")
if file_path:
try:
with open(file_path, "w") as f:
f.write(code.strip("```python").strip("```"))
QMessageBox.information(self, "Success", "Code saved successfully!")
except Exception as e:
QMessageBox.critical(self, "Error", f"Error saving code: {str(e)}")
def update_chat_display(self, message):
self.chat_display.append(message + "\n")
self.chat_display.verticalScrollBar().setValue(self.chat_display.verticalScrollBar().maximum())
def clear_conversation(self):
self.messages = [
{"role": "system", "content": "You are a helpful AI assistant."}
]
self.chat_display.clear()
def show_context_menu(self, point):
menu = QMenu(self)
copy_action = menu.addAction("Copy")
copy_action.triggered.connect(self.copy_text)
menu.exec_(self.chat_display.mapToGlobal(point))
def copy_text(self):
cursor = self.chat_display.textCursor()
if cursor.hasSelection():
text = cursor.selectedText()
QApplication.clipboard().setText(text)
if __name__ == "__main__":
app = QApplication(sys.argv)
gui = ChatbotGUI()
gui.show()
sys.exit(app.exec_())
```
## Training and finetuning cost
It takes 6 days to pretrain AMD-Llama-135m on 4 MI250 nodes each of which has 4 MI250 GPUs (8 virtual GPU cards, 64G memory for each).
It takes 4 days to finetune AMD-Llama-135m-code on 4 MI250 GPUs.
It takes 11T disk space to store raw and processed SlimPajama, project gutenberg and Starcoder datasets.
#### License
Copyright (c) 2018-2024 Advanced Micro Devices, Inc. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. |