SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Update tokenization_code_llama_fast.py (#26576)

Browse Source 
* Update tokenization_code_llama_fast.py

* Update test_tokenization_code_llama.py

* Update test_tokenization_code_llama.py
This commit is contained in:
Tianqi Liu
2023-10-06 01:49:02 -07:00
committed by GitHub
parent af38c837ee
commit 65aabafe2f
2 changed files with 13 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/transformers/models/code_llama/tokenization_code_llama_fast.py
View File

@@ -278,7 +278,7 @@ class CodeLlamaTokenizerFast(PreTrainedTokenizerFast):
special_tokens = [(self.bos_token, self.bos_token_id)] if self.add_bos_token and add_special_tokens else []
if suffix_first:
# format as " <PRE> <SUF>{suf} <MID> {pre}"
pair += [self.prefix_token, self.suffix_token, "$A", self.middle_token, "$B"]
pair += [self.prefix_token, self.suffix_token, "$B", self.middle_token, "$A"]
special_tokens += [
(self.prefix_token, self.prefix_id),
(self.suffix_token, self.suffix_id),

12
tests/models/code_llama/test_tokenization_code_llama.py
View File

@@ -643,3 +643,15 @@ end
input_ids = tokenizer.encode(PROMPTS[0])
self.assertEqual(input_ids, tokenizer.encode(prefix, suffix=suffix))
self.assertEqual(tokenizer.encode(prefix, suffix=suffix), tokenizer_fast.encode(prefix, suffix=suffix))
# Adding suffix_first check for infilling tasks
suffix_first_formatted_prompt = tokenizer.tokenize(PROMPTS[0], suffix_first=True)
self.assertEqual(suffix_first_formatted_prompt, tokenizer_fast.tokenize(PROMPTS[0], suffix_first=True))
prefix, suffix = PROMPTS[0].split("<FILL_ME>")
self.assertEqual(suffix_first_formatted_prompt, tokenizer.tokenize(prefix, suffix, suffix_first=True))
self.assertEqual(suffix_first_formatted_prompt, tokenizer_fast.tokenize(prefix, suffix, suffix_first=True))
prefix, suffix = PROMPTS[0].split("<FILL_ME>")
suffix_first_input_ids = tokenizer.encode(PROMPTS[0], suffix_first=True)
self.assertEqual(suffix_first_input_ids, tokenizer.encode(prefix, suffix=suffix, suffix_first=True))
self.assertEqual(suffix_first_input_ids, tokenizer_fast.encode(prefix, suffix=suffix, suffix_first=True))