Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-23 00:18:20 +01:00
Code Issues Packages Projects Releases Wiki Activity

fix random seeds to actually randomize (#1004 from mcmonkey4eva/seed-fix)

Browse Source
This commit is contained in:
oobabooga 2023-04-10 10:56:12 -03:00 committed by GitHub
commit 843f672227
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 11 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
modules/text_generation.py
View File

@ -1,3 +1,4 @@
import random
import re import re
import time import time
import traceback import traceback
@ -97,10 +98,13 @@ def formatted_outputs(reply, model_name):
def set_manual_seed(seed): def set_manual_seed(seed):
if seed != -1: seed = int(seed)
if seed == -1:
seed = random.randint(1, 2**31)
torch.manual_seed(seed) torch.manual_seed(seed)
if torch.cuda.is_available(): if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed) torch.cuda.manual_seed_all(seed)
return seed
def stop_everything_event(): def stop_everything_event():
@ -109,7 +113,7 @@ def stop_everything_event():
def generate_reply(question, generate_state, eos_token=None, stopping_strings=[]): def generate_reply(question, generate_state, eos_token=None, stopping_strings=[]):
clear_torch_cache() clear_torch_cache()
set_manual_seed(generate_state['seed']) seed = set_manual_seed(generate_state['seed'])
shared.stop_everything = False shared.stop_everything = False
generate_params = {} generate_params = {}
t0 = time.time() t0 = time.time()
@ -151,7 +155,7 @@ def generate_reply(question, generate_state, eos_token=None, stopping_strings=[]
t1 = time.time() t1 = time.time()
original_tokens = len(encode(original_question)[0]) original_tokens = len(encode(original_question)[0])
new_tokens = len(encode(output)[0]) - original_tokens new_tokens = len(encode(output)[0]) - original_tokens
print(f'Output generated in {(t1-t0):.2f} seconds ({new_tokens/(t1-t0):.2f} tokens/s, {new_tokens} tokens, context {original_tokens})') print(f'Output generated in {(t1-t0):.2f} seconds ({new_tokens/(t1-t0):.2f} tokens/s, {new_tokens} tokens, context {original_tokens}, seed {seed})')
return return
input_ids = encode(question, generate_state['max_new_tokens']) input_ids = encode(question, generate_state['max_new_tokens'])
@ -272,5 +276,5 @@ def generate_reply(question, generate_state, eos_token=None, stopping_strings=[]
t1 = time.time() t1 = time.time()
original_tokens = len(original_input_ids[0]) original_tokens = len(original_input_ids[0])
new_tokens = len(output) - original_tokens new_tokens = len(output) - original_tokens
print(f'Output generated in {(t1-t0):.2f} seconds ({new_tokens/(t1-t0):.2f} tokens/s, {new_tokens} tokens, context {original_tokens})') print(f'Output generated in {(t1-t0):.2f} seconds ({new_tokens/(t1-t0):.2f} tokens/s, {new_tokens} tokens, context {original_tokens}, seed {seed})')
return return