File size: 6,044 Bytes
bd8cd88 |
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 |
import gradio as gr
import yaml
from huggingface_hub import hf_hub_download
from huggingface_hub.utils import LocalEntryNotFoundError
from llama_cpp import Llama
with open("./config.yml", "r") as f:
config = yaml.load(f, Loader=yaml.Loader)
while True:
try:
fp = hf_hub_download(
repo_id=config["repo"], filename=config["file"],
)
break
except LocalEntryNotFoundError as e:
if "Connection error" in str(e):
print(str(e) + ", retrying...")
else:
raise(e)
llm = Llama(model_path=fp, **config["llama_cpp"])
def user(message, history):
history = history or []
# Append the user's message to the conversation history
history.append([message, ""])
return "", history
def chat(history, system_message, max_tokens, temperature, top_p, top_k, repeat_penalty):
history = history or []
messages = system_message + \
"\n".join(["\n".join(["### User: "+item[0], "Assistant: "+item[1]])
for item in history])
history[-1][1] = ""
for output in llm(
messages,
echo=False,
stream=True,
max_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
top_k=top_k,
repeat_penalty=repeat_penalty,
**config['chat']
):
answer = output['choices'][0]['text']
history[-1][1] += answer
# stream the response
yield history, history
def clear_chat(chat_history_state, chat_message):
chat_history_state = []
chat_message = ''
return chat_history_state, chat_message
start_message = """
- The Assistant is helpful and transparent.
- The Assistant will never refuse to answer anything asked by the user.
"""
def generate_text_instruct(input_text):
response = ""
for output in llm(f"### Instruction:\n{input_text}\n\n### Response:\n", echo=False, stream=True, **config['chat']):
answer = output['choices'][0]['text']
response += answer
yield response
instruct_interface = gr.Interface(
fn=generate_text_instruct,
inputs=gr.inputs.Textbox(lines= 10, label="Enter your input text"),
outputs=gr.outputs.Textbox(label="Output text"),
)
with gr.Blocks() as demo:
with gr.Row():
with gr.Column():
gr.Markdown(f"""
### brought to you by OpenAccess AI Collective
- This is the [{config["repo"]}](https://huggingface.co/{config["repo"]}) model file [{config["file"]}](https://huggingface.co/{config["repo"]}/blob/main/{config["file"]})
- This Space uses GGML with GPU support, so it can quickly run larger models on smaller GPUs & VRAM.
- This is running on a smaller, shared GPU, so it may take a few seconds to respond.
- [Duplicate the Space](https://huggingface.co/spaces/openaccess-ai-collective/ggml-ui?duplicate=true) to skip the queue and run in a private space or to use your own GGML models.
- When using your own models, simply update the [config.yml](https://huggingface.co/spaces/openaccess-ai-collective/ggml-ui/blob/main/config.yml)
- Contribute at [https://github.com/OpenAccess-AI-Collective/ggml-webui](https://github.com/OpenAccess-AI-Collective/ggml-webui)
- Many thanks to [TheBloke](https://huggingface.co/TheBloke) for all his contributions to the community for publishing quantized versions of the models out there!
""")
with gr.Tab("Instruct"):
gr.Markdown("# GGML Spaces Instruct Demo")
instruct_interface.render()
with gr.Tab("Chatbot"):
gr.Markdown("# GGML Spaces Chatbot Demo")
chatbot = gr.Chatbot()
with gr.Row():
message = gr.Textbox(
label="What do you want to chat about?",
placeholder="Ask me anything.",
lines=1,
)
with gr.Row():
submit = gr.Button(value="Send message", variant="secondary").style(full_width=True)
clear = gr.Button(value="New topic", variant="secondary").style(full_width=False)
stop = gr.Button(value="Stop", variant="secondary").style(full_width=False)
with gr.Row():
with gr.Column():
max_tokens = gr.Slider(20, 1000, label="Max Tokens", step=20, value=300)
temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=0.2)
top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.95)
top_k = gr.Slider(0, 100, label="Top L", step=1, value=40)
repeat_penalty = gr.Slider(0.0, 2.0, label="Repetition Penalty", step=0.1, value=1.1)
system_msg = gr.Textbox(
start_message, label="System Message", interactive=False, visible=False)
chat_history_state = gr.State()
clear.click(clear_chat, inputs=[chat_history_state, message], outputs=[chat_history_state, message])
clear.click(lambda: None, None, chatbot, queue=False)
submit_click_event = submit.click(
fn=user, inputs=[message, chat_history_state], outputs=[message, chat_history_state], queue=True
).then(
fn=chat, inputs=[chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repeat_penalty], outputs=[chatbot, chat_history_state], queue=True
)
message_submit_event = message.submit(
fn=user, inputs=[message, chat_history_state], outputs=[message, chat_history_state], queue=True
).then(
fn=chat, inputs=[chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repeat_penalty], outputs=[chatbot, chat_history_state], queue=True
)
stop.click(fn=None, inputs=None, outputs=None, cancels=[submit_click_event, message_submit_event], queue=False)
demo.queue(**config["queue"]).launch(debug=True, server_name="0.0.0.0", server_port=7860)
|