Spaces:
Sleeping
Sleeping
File size: 3,860 Bytes
9b69685 34d125a 9b69685 1d80a44 34d125a 9b69685 34d125a 9b69685 34d125a 9b69685 34d125a 9b69685 34d125a 9b69685 34d125a 9b69685 34d125a 2bc3349 9b69685 |
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 |
import gradio as gr
import outlines
import transformers
import torch
"""
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
"""
pipe = transformers.pipeline("text-generation", "HuggingFaceTB/SmolLM-1.7B-Instruct", torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32)
outlines_tokenizer = outlines.models.TransformerTokenizer(pipe.tokenizer)
### TODO 1: use outliunes with a transformer model made directly
### TODO 2: use a cfg
def string_to_acrostic_grammar(s, dash_initial=True):
# this will convert a string to a CFG grammar
chars = filter(str.isalpha, s.upper())
grammar_rules = [('"- " ' if dash_initial else '') + f'"{char}" /[^-\\r\\n]+/ "\\n"' for char in chars]
return "?start: " + " ".join(grammar_rules)
def is_this_prompt_a_list(prompt):
return False
# ask the model if the prompt is a list, by constraining the generation to yes or no about a question whether the prompt is a list
question = f'This is a prompt that you have been asked to answer:\n\n```\n{prompt}\n```\n\nIs this prompt asking for a list of items, instead of a story? Begin your answer with "Yes" if asking for a list, otherwise "No", and then give an explanation of why.'
grammar = '?start: ("Yes" | "No")'
cfg_logits_processor = outlines.processors.CFGLogitsProcessor(grammar, outlines_tokenizer)
output = pipe([{"role": "user", "content": question}], logits_processor=transformers.LogitsProcessorList([cfg_logits_processor]), max_new_tokens=10,)
# output = pipe([{"role": "system", "content": "You are a helpful assistant who answers in one-word answers."}, {"role": "user", "content": question}], max_new_tokens=10,)
response = output[0]['generated_text'][-1]['content']
print("is this prompt a list?", response)
return response == "Yes"
def respond(
message,
history: list[tuple[str, str]],
system_message,
acrostic,
max_tokens,
temperature,
top_p,
):
print({"message": message, "history": history, "system_message": system_message, "acrostic": acrostic, "max_tokens": max_tokens, "temperature": temperature, "top_p": top_p})
# grammar = "\n".join(['?start: item item item','?item: "- " /[^-\\r\\n]+/ "\\n"'])
grammar = string_to_acrostic_grammar(acrostic, dash_initial=is_this_prompt_a_list(message))
two_items_logits_processor = outlines.processors.CFGLogitsProcessor( grammar , outlines_tokenizer )
output = pipe([{"role": "user", "content": message}], logits_processor=transformers.LogitsProcessorList([two_items_logits_processor]), max_new_tokens=max_tokens,)
print(output)
response = output[0]['generated_text'][-1]['content']
# messages = [{"role": "system", "content": system_message}]
# for val in history:
# if val[0]:
# messages.append({"role": "user", "content": val[0]})
# if val[1]:
# messages.append({"role": "assistant", "content": val[1]})
# messages.append({"role": "user", "content": message})
yield response
"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
demo = gr.ChatInterface(
respond,
additional_inputs=[
gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
gr.Textbox(value="I love you", label="acrostic"),
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Maximum new 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 (nucleus sampling)",
),
],
)
if __name__ == "__main__":
demo.launch()
|