text-generation-webui/modules/exllama.py

216 lines
8.5 KiB
Python
Raw Normal View History

2023-06-17 01:35:38 +02:00
from pathlib import Path
import torch
import torch.nn.functional as F
from torch import version as torch_version
2023-09-11 16:57:38 +02:00
from modules import shared
2023-06-17 01:35:38 +02:00
from modules.logging_colors import logger
from modules.models import clear_torch_cache
2023-07-07 18:09:23 +02:00
from modules.text_generation import get_max_prompt_length
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
2023-09-11 06:14:36 +02:00
if shared.args.alpha_value > 1 and shared.args.rope_freq_base == 0:
config.alpha_value = shared.args.alpha_value
config.calculate_rotary_embedding_base()
2023-09-11 06:14:36 +02:00
elif shared.args.rope_freq_base > 0:
config.rotary_embedding_base = shared.args.rope_freq_base
if torch_version.hip:
config.rmsnorm_no_half2 = True
config.rope_no_half2 = True
config.matmul_no_half2 = True
config.silu_no_half2 = 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):
# The cache batch size must be 2 for CFG and 1 otherwise
if state['guidance_scale'] == 1:
if self.cache.batch_size == 2:
del self.cache
clear_torch_cache()
self.cache = ExLlamaCache(self.model)
self.generator = ExLlamaGenerator(self.model, self.tokenizer, self.cache)
else:
if self.cache.batch_size == 1:
del self.cache
clear_torch_cache()
self.cache = ExLlamaCache(self.model, batch_size=2)
self.generator = ExLlamaGenerator(self.model, self.tokenizer, self.cache)
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']
2023-06-29 18:53:06 +02:00
self.generator.settings.token_repetition_penalty_sustain = -1 if state['repetition_penalty_range'] <= 0 else state['repetition_penalty_range']
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)
2023-09-15 23:27:27 +02:00
if state['custom_token_bans']:
to_ban = [int(x) for x in state['custom_token_bans'].split(',')]
if len(to_ban) > 0:
self.generator.disallow_tokens(to_ban)
2023-09-15 23:27:27 +02:00
# Case 1: no CFG
if state['guidance_scale'] == 1:
self.generator.end_beam_search()
# Tokenizing the input
2023-08-20 17:50:32 +02:00
ids = self.generator.tokenizer.encode(prompt, max_seq_len=self.model.config.max_seq_len)
if state['add_bos_token']:
ids = torch.cat(
[torch.tensor([[self.tokenizer.bos_token_id]]).to(ids.device),
ids], dim=1
).to(torch.int64)
ids = ids[:, -get_max_prompt_length(state):]
if state['auto_max_new_tokens']:
max_new_tokens = state['truncation_length'] - ids.shape[-1]
else:
max_new_tokens = state['max_new_tokens']
self.generator.gen_begin_reuse(ids)
initial_len = self.generator.sequence[0].shape[0]
has_leading_space = False
2023-07-07 18:09:23 +02:00
for i in range(max_new_tokens):
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
if token.item() == self.generator.tokenizer.eos_token_id or shared.stop_everything:
break
# Case 2: CFG
2023-08-06 22:42:43 +02:00
# Copied from https://github.com/turboderp/exllama/blob/master/example_cfg.py
else:
alpha = state['guidance_scale']
prompts = [prompt, state['negative_prompt'] or '']
ids, mask = self.tokenizer.encode(
prompts,
return_mask=True,
max_seq_len=self.model.config.max_seq_len,
add_bos=state['add_bos_token']
)
if state['auto_max_new_tokens']:
max_new_tokens = state['truncation_length'] - ids[0].shape[-1]
else:
max_new_tokens = state['max_new_tokens']
self.generator.gen_begin(ids, mask=mask)
initial_len = self.generator.sequence[0].shape[0]
has_leading_space = False
for i in range(max_new_tokens):
logits = self.model.forward(self.generator.sequence[:, -1:], self.cache, input_mask=mask)
self.generator.apply_rep_penalty(logits)
logits = F.log_softmax(logits, dim=-1)
logits_mixed = alpha * logits[0] + (1 - alpha) * logits[1]
token, _ = self.generator.sample_current(logits_mixed)
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
if token.item() == self.tokenizer.eos_token_id or shared.stop_everything:
break
batch_token = token.repeat(2, 1)
self.generator.gen_accept_token(batch_token)
2023-06-17 01:35:38 +02:00
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, max_seq_len=self.model.config.max_seq_len, add_bos=True)
def decode(self, ids, **kwargs):
if isinstance(ids, int):
ids = torch.tensor([[ids]])
elif isinstance(ids, torch.Tensor) and ids.numel() == 1:
ids = ids.view(1, -1)
return self.tokenizer.decode(ids)[0]
2023-09-17 15:42:32 +02:00
def get_logits(self, token_ids, **kwargs):
self.cache.current_seq_len = 0
self.model.forward(token_ids[:, :-1], self.cache, input_mask=None, preprocess_only=True)
return self.model.forward(token_ids[:, -1:], self.cache, **kwargs).float().cpu()