text-generation-webui/modules/exllama.py

106 lines
3.9 KiB
Python
Raw Normal View History

2023-06-17 01:35:38 +02:00
import sys
from pathlib import Path
from modules import shared
2023-06-17 01:35:38 +02:00
from modules.logging_colors import logger
try:
from exllama.generator import ExLlamaGenerator
from exllama.model import ExLlama, ExLlamaCache, ExLlamaConfig
from exllama.tokenizer import ExLlamaTokenizer
except:
logger.warning('Exllama module failed to load. Will attempt to load from repositories.')
try:
from modules.relative_imports import RelativeImport
with RelativeImport("repositories/exllama"):
from generator import ExLlamaGenerator
from model import ExLlama, ExLlamaCache, ExLlamaConfig
from tokenizer import ExLlamaTokenizer
except:
logger.error("Could not find repositories/exllama/. Make sure that exllama is cloned inside repositories/ and is up to date.")
raise
2023-06-17 01:35:38 +02:00
class ExllamaModel:
def __init__(self):
pass
@classmethod
def from_pretrained(self, path_to_model):
path_to_model = Path(f'{shared.args.model_dir}') / Path(path_to_model)
2023-06-17 01:35:38 +02:00
tokenizer_model_path = path_to_model / "tokenizer.model"
model_config_path = path_to_model / "config.json"
# Find the model checkpoint
model_path = None
for ext in ['.safetensors', '.pt', '.bin']:
found = list(path_to_model.glob(f"*{ext}"))
if len(found) > 0:
if len(found) > 1:
logger.warning(f'More than one {ext} model has been found. The last one will be selected. It could be wrong.')
model_path = found[-1]
break
config = ExLlamaConfig(str(model_config_path))
config.model_path = str(model_path)
2023-06-26 03:49:26 +02:00
config.max_seq_len = shared.args.max_seq_len
config.compress_pos_emb = shared.args.compress_pos_emb
if shared.args.gpu_split:
config.set_auto_map(shared.args.gpu_split)
config.gpu_peer_fix = True
2023-06-17 01:35:38 +02:00
model = ExLlama(config)
tokenizer = ExLlamaTokenizer(str(tokenizer_model_path))
cache = ExLlamaCache(model)
2023-06-17 23:00:10 +02:00
generator = ExLlamaGenerator(model, tokenizer, cache)
2023-06-17 01:35:38 +02:00
result = self()
result.config = config
result.model = model
result.cache = cache
result.tokenizer = tokenizer
result.generator = generator
2023-06-17 01:35:38 +02:00
return result, result
2023-06-18 00:02:08 +02:00
def generate_with_streaming(self, prompt, state):
2023-06-17 23:00:10 +02:00
self.generator.settings.temperature = state['temperature']
self.generator.settings.top_p = state['top_p']
self.generator.settings.top_k = state['top_k']
self.generator.settings.typical = state['typical_p']
self.generator.settings.token_repetition_penalty_max = state['repetition_penalty']
2023-06-17 01:35:38 +02:00
if state['ban_eos_token']:
2023-06-17 23:00:10 +02:00
self.generator.disallow_tokens([self.tokenizer.eos_token_id])
else:
self.generator.disallow_tokens(None)
self.generator.end_beam_search()
ids = self.generator.tokenizer.encode(prompt)
self.generator.gen_begin_reuse(ids)
initial_len = self.generator.sequence[0].shape[0]
has_leading_space = False
for i in range(state['max_new_tokens']):
2023-06-17 23:00:10 +02:00
token = self.generator.gen_single_token()
if i == 0 and self.generator.tokenizer.tokenizer.IdToPiece(int(token)).startswith(''):
has_leading_space = True
decoded_text = self.generator.tokenizer.decode(self.generator.sequence[0][initial_len:])
if has_leading_space:
decoded_text = ' ' + decoded_text
yield decoded_text
2023-06-17 23:00:10 +02:00
if token.item() == self.generator.tokenizer.eos_token_id or shared.stop_everything:
2023-06-17 01:35:38 +02:00
break
2023-06-18 00:02:08 +02:00
def generate(self, prompt, state):
output = ''
for output in self.generate_with_streaming(prompt, state):
pass
return output
2023-06-17 01:35:38 +02:00
def encode(self, string, **kwargs):
return self.tokenizer.encode(string)