Spaces:
Running
on
Zero
Running
on
Zero
File size: 6,993 Bytes
0826ff1 18b894b 0826ff1 18b894b 0826ff1 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 |
# Defaults
import spaces
import logging
import gradio as gr
from huggingface_hub import hf_hub_download
from llama_cpp import Llama
# Locals
from llama_cpp_agent.providers import LlamaCppPythonProvider
from content import css, PLACEHOLDER
from utils import CitingSources
# Agents
from llama_cpp_agent import LlamaCppAgent
from llama_cpp_agent.chat_history import BasicChatHistory
from llama_cpp_agent.chat_history.messages import Roles
from llama_cpp_agent.llm_output_settings import (
LlmStructuredOutputSettings,
LlmStructuredOutputType,
)
# Tools
from llama_cpp_agent.tools import WebSearchTool
from llama_cpp_agent.prompt_templates import web_search_system_prompt, research_system_prompt
examples = [
["Latest uplifting news"],
["Latest news site:bloomberg.com"],
["Where I can find best hotel in Galapagos, Ecuador intitle:hotel"],
["filetype:pdf intitle:python"]
]
hf_hub_download(
repo_id="bartowski/Mistral-7B-Instruct-v0.3-GGUF",
filename="Mistral-7B-Instruct-v0.3-Q6_K.gguf",
local_dir="./models"
)
hf_hub_download(
repo_id="bartowski/Meta-Llama-3-8B-Instruct-GGUF",
filename="Meta-Llama-3-8B-Instruct-Q6_K.gguf",
local_dir="./models"
)
def get_context_by_model(model_name):
model_context_limits = {
"Mistral-7B-Instruct-v0.3-Q6_K.gguf": 32768,
"Meta-Llama-3-8B-Instruct-Q6_K.gguf": 8192
}
return model_context_limits.get(model_name, None)
def get_messages_formatter_type(model_name):
from llama_cpp_agent import MessagesFormatterType
if "Meta" in model_name or "aya" in model_name:
return MessagesFormatterType.LLAMA_3
elif "Mistral" in model_name:
return MessagesFormatterType.MISTRAL
elif "Einstein-v6-7B" in model_name or "dolphin" in model_name:
return MessagesFormatterType.CHATML
elif "Phi" in model_name:
return MessagesFormatterType.PHI_3
else:
return MessagesFormatterType.CHATML
## Run inference
@spaces.GPU(duration=120)
def respond(
message,
history: list[tuple[str, str]],
system_message,
max_tokens,
temperature,
top_p,
top_k,
repetition_penalty,
model,
):
chat_template = get_messages_formatter_type(model)
llm = Llama(
model_path=f"models/{model}",
flash_attn=True,
n_threads=40,
n_gpu_layers=81,
n_batch=1024,
n_ctx=get_context_by_model(model),
)
provider = LlamaCppPythonProvider(llm)
logging.info(f"Loaded chat examples: {chat_template}")
search_tool = WebSearchTool(
llm_provider=provider,
message_formatter_type=chat_template,
max_tokens_search_results=12000,
max_tokens_per_summary=2048,
)
web_search_agent = LlamaCppAgent(
provider,
system_prompt=web_search_system_prompt,
predefined_messages_formatter_type=chat_template,
debug_output=True,
)
answer_agent = LlamaCppAgent(
provider,
system_prompt=research_system_prompt,
predefined_messages_formatter_type=chat_template,
debug_output=True,
)
settings = provider.get_provider_default_settings()
settings.stream = False
settings.temperature = temperature
settings.top_k = top_k
settings.top_p = top_p
settings.max_tokens = max_tokens
settings.repeat_penalty = repetition_penalty
output_settings = LlmStructuredOutputSettings.from_functions(
[search_tool.get_tool()]
)
messages = BasicChatHistory()
for msn in history:
user = {"role": Roles.user, "content": msn[0]}
assistant = {"role": Roles.assistant, "content": msn[1]}
messages.add_message(user)
messages.add_message(assistant)
result = web_search_agent.get_chat_response(
message,
llm_sampling_settings=settings,
structured_output_settings=output_settings,
add_message_to_chat_history=False,
add_response_to_chat_history=False,
print_output=False,
)
outputs = ""
settings.stream = True
response_text = answer_agent.get_chat_response(
f"Write a detailed and complete research document that fulfills the following user request: '{message}', based on the information from the web below.\n\n" + result[0]["return_value"],
role=Roles.tool,
llm_sampling_settings=settings,
chat_history=messages,
returns_streaming_generator=True,
print_output=False,
)
for text in response_text:
outputs += text
yield outputs
output_settings = LlmStructuredOutputSettings.from_pydantic_models(
[CitingSources], LlmStructuredOutputType.object_instance
)
citing_sources = answer_agent.get_chat_response(
"Cite the sources you used in your response.",
role=Roles.tool,
llm_sampling_settings=settings,
chat_history=messages,
returns_streaming_generator=False,
structured_output_settings=output_settings,
print_output=False,
)
outputs += "\n\nSources:\n"
outputs += "\n".join(citing_sources.sources)
yield outputs
# Begin Gradio UI
main = gr.ChatInterface(
respond,
additional_inputs=[
gr.Textbox(
value=web_search_system_prompt,
label="System message",
interactive=True,
),
gr.Slider(minimum=1, maximum=4096, value=2048, step=1, label="Max tokens"),
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.95,
step=0.05,
label="Top-p",
),
gr.Slider(
minimum=0,
maximum=100,
value=40,
step=1,
label="Top-k",
),
gr.Slider(
minimum=0.0,
maximum=2.0,
value=1.1,
step=0.1,
label="Repetition penalty",
),
],
theme=gr.themes.Soft(
primary_hue="violet",
secondary_hue="violet",
neutral_hue="gray",
font=[gr.themes.GoogleFont("Exo"), "ui-sans-serif", "system-ui", "sans-serif"],
).set(
body_background_fill_dark="#111111",
block_background_fill_dark="#111111",
block_border_width="1px",
block_title_background_fill_dark="#1e1c26",
input_background_fill_dark="#292733",
button_secondary_background_fill_dark="#24212b",
border_color_primary_dark="#343140",
background_fill_secondary_dark="#111111",
color_accent_soft_dark="transparent",
),
css=css,
retry_btn="Retry",
undo_btn="Undo",
clear_btn="Clear",
submit_btn="Send",
examples=(examples),
analytics_enabled=False,
description="Llama-cpp-agent: Chat Web Search Agent",
chatbot=gr.Chatbot(scale=1, placeholder=PLACEHOLDER),
)
if __name__ == "__main__":
main.launch()
|