File size: 3,217 Bytes
ed817dc 44092ff ed817dc 44092ff 42198c1 44092ff b282946 44092ff |
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 |
import torch
import transformers
from torch import cuda
from accelerate import dispatch_model, infer_auto_device_map
from accelerate.utils import get_balanced_memory
from transformers import BitsAndBytesConfig, StoppingCriteria, StoppingCriteriaList
from typing import Dict, List, Any
class PreTrainedPipeline():
def __init__(self, path=""):
path = "oleksandrfluxon/mpt-7b-instruct-evaluate"
print("===> path", path)
device = f'cuda:{cuda.current_device()}' if cuda.is_available() else 'cpu'
print("===> device", device)
model = transformers.AutoModelForCausalLM.from_pretrained(
'oleksandrfluxon/mpt-7b-instruct-evaluate',
trust_remote_code=True,
load_in_8bit=True, # this requires the `bitsandbytes` library
max_seq_len=8192,
init_device=device
)
model.eval()
#model.to(device)
print(f"===> Model loaded on {device}")
tokenizer = transformers.AutoTokenizer.from_pretrained("mosaicml/mpt-7b")
# we create a list of stopping criteria
stop_token_ids = [
tokenizer.convert_tokens_to_ids(x) for x in [
['Human', ':'], ['AI', ':']
]
]
stop_token_ids = [torch.LongTensor(x).to(device) for x in stop_token_ids]
print("===> stop_token_ids", stop_token_ids)
# define custom stopping criteria object
class StopOnTokens(StoppingCriteria):
def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
for stop_ids in stop_token_ids:
if torch.eq(input_ids[0][-len(stop_ids):], stop_ids).all():
return True
return False
stopping_criteria = StoppingCriteriaList([StopOnTokens()])
self.pipeline = transformers.pipeline(
model=model, tokenizer=tokenizer,
return_full_text=True, # langchain expects the full text
task='text-generation',
# we pass model parameters here too
stopping_criteria=stopping_criteria, # without this model rambles during chat
temperature=0.1, # 'randomness' of outputs, 0.0 is the min and 1.0 the max
top_p=0.15, # select from top tokens whose probability add up to 15%
top_k=0, # select from top 0 tokens (because zero, relies on top_p)
max_new_tokens=1000, # mex number of tokens to generate in the output
repetition_penalty=1.1 # without this output begins repeating
)
print("===> init finished")
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
"""
data args:
inputs (:obj: `str`)
parameters (:obj: `str`)
Return:
A :obj:`str`: todo
"""
# get inputs
inputs = data.pop("inputs",data)
parameters = data.pop("parameters", {})
date = data.pop("date", None)
print("===> inputs", inputs)
print("===> parameters", parameters)
result = self.pipeline(inputs, **parameters)
print("===> result", result)
return result |