Spaces:
Runtime error
Runtime error
File size: 2,526 Bytes
fe1089d b0721f8 fe1089d a597c76 fe1089d 2492536 b0721f8 c7e16d0 1f063be c7e16d0 b0721f8 fe1089d 2492536 b0721f8 fe1089d b0721f8 fe1089d c7e16d0 2492536 fe1089d 2492536 c7e16d0 fe1089d 2492536 fe1089d |
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 |
# GODEL model module for chat interaction and model instance control
# external imports
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, GenerationConfig
# internal imports
from utils import modelling as mdl
# global model and tokenizer instance (created on initial build)
TOKENIZER = AutoTokenizer.from_pretrained("microsoft/GODEL-v1_1-large-seq2seq")
MODEL = AutoModelForSeq2SeqLM.from_pretrained("microsoft/GODEL-v1_1-large-seq2seq")
# model config definition
CONFIG = GenerationConfig.from_pretrained("microsoft/GODEL-v1_1-large-seq2seq")
base_config_dict = {
"max_new_tokens": 64,
"min_length": 8,
"top_p": 0.9,
"do_sample": True,
}
CONFIG.update(**base_config_dict)
# function to (re) set config
def set_config(config_dict: dict):
# if config dict is not given, set to default
if config_dict == {}:
config_dict = base_config_dict
CONFIG.update(**config_dict)
# formatting class to formatting input for the model
# CREDIT: Adapted from official interference example on Huggingface
## see https://huggingface.co/microsoft/GODEL-v1_1-large-seq2seq
def format_prompt(message: str, history: list, system_prompt: str, knowledge: str = ""):
# user input prompt initialization
prompt = ""
# limits the prompt elements to the maximum token count
message, history, system_prompt, knowledge = mdl.prompt_limiter(
TOKENIZER, message, history, system_prompt, knowledge
)
# adds knowledge text if not empty
if knowledge != "":
knowledge = "[KNOWLEDGE] " + knowledge
# adds conversation history to the prompt
for conversation in history:
prompt += f"EOS {conversation[0]} EOS {conversation[1]}"
# adds the message to the prompt
prompt += f" {message}"
# combines the entire prompt
full_prompt = f"{system_prompt} [CONTEXT] {prompt} {knowledge}"
# returns the formatted prompt
return full_prompt
# response class calling the model and returning the model output message
# CREDIT: Copied from official interference example on Huggingface
## see https://huggingface.co/microsoft/GODEL-v1_1-large-seq2seq
def respond(prompt):
set_config({})
# tokenizing input string
input_ids = TOKENIZER(f"{prompt}", return_tensors="pt").input_ids
# generating using config and decoding output
outputs = MODEL.generate(input_ids, generation_config=CONFIG)
output = TOKENIZER.decode(outputs[0], skip_special_tokens=True)
# returns the model output string
return output
|