Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-22 08:07:56 +01:00
Code Issues Packages Projects Releases Wiki Activity

Fix partial unicode characters issue (#4837)

Browse Source
This commit is contained in:
Yiximail 2023-12-08 20:50:53 +08:00 committed by GitHub
parent 2c5a1e67f9
commit 1c74b3ab45
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 33 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
modules/exllama.py
View File

@ -165,10 +165,19 @@ class ExllamaModel:
if has_leading_space: if has_leading_space:
decoded_text = ' ' + decoded_text decoded_text = ' ' + decoded_text
yield decoded_text # Check the partial unicode character
if chr(0xfffd) in decoded_text:
is_last = i == max_new_tokens - 1
is_stopping = token.item() == self.generator.tokenizer.eos_token_id or shared.stop_everything
# If we are not at the end of the generation, we skip this token
if not (is_last or is_stopping):
continue
if token.item() == self.generator.tokenizer.eos_token_id or shared.stop_everything: if token.item() == self.generator.tokenizer.eos_token_id or shared.stop_everything:
break break
yield decoded_text
# Case 2: CFG # Case 2: CFG
# Copied from https://github.com/turboderp/exllama/blob/master/example_cfg.py # Copied from https://github.com/turboderp/exllama/blob/master/example_cfg.py
else: else:
@ -205,6 +214,14 @@ class ExllamaModel:
if has_leading_space: if has_leading_space:
decoded_text = ' ' + decoded_text decoded_text = ' ' + decoded_text
# Check the partial unicode character
if chr(0xfffd) in decoded_text:
is_last = i == max_new_tokens - 1
is_stopping = token.item() == self.tokenizer.eos_token_id or shared.stop_everything
# If we are not at the end of the generation, we skip this token
if not (is_last or is_stopping):
continue
yield decoded_text yield decoded_text
if token.item() == self.tokenizer.eos_token_id or shared.stop_everything: if token.item() == self.tokenizer.eos_token_id or shared.stop_everything:
break break

10
modules/exllamav2.py
View File

@ -138,11 +138,19 @@ class Exllamav2Model:
if has_leading_space: if has_leading_space:
decoded_text = ' ' + decoded_text decoded_text = ' ' + decoded_text
yield decoded_text # Check the partial unicode character
if chr(0xfffd) in decoded_text:
is_last = i == max_new_tokens - 1
is_stopping = token.item() == self.tokenizer.eos_token_id or shared.stop_everything
# If we are not at the end of the generation, we skip this token
if not (is_last or is_stopping):
continue
if token.item() == self.tokenizer.eos_token_id or shared.stop_everything: if token.item() == self.tokenizer.eos_token_id or shared.stop_everything:
break break
yield decoded_text
def generate(self, prompt, state): def generate(self, prompt, state):
output = '' output = ''
for output in self.generate_with_streaming(prompt, state): for output in self.generate_with_streaming(prompt, state):

7
modules/text_generation.py
View File

@ -362,7 +362,12 @@ def generate_reply_HF(question, original_question, seed, state, stopping_strings
if output[-1] in eos_token_ids: if output[-1] in eos_token_ids:
break break
cumulative_reply += get_reply_from_output_ids(output, state, starting_from=starting_from) new_content = get_reply_from_output_ids(output, state, starting_from=starting_from)
# check the partial unicode character
if chr(0xfffd) in new_content:
continue
cumulative_reply += new_content
starting_from = len(output) starting_from = len(output)
yield cumulative_reply yield cumulative_reply