vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update bruteforce test

Browse source
This commit is contained in:
jaime-m-p 2024-06-25 17:36:44 +02:00
parent 107923cdd2
commit 68220feaf8
1 changed files with 7 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

11
tests/test-tokenizer-random.py
View file

@ -235,6 +235,8 @@ def generator_custom_text_edge_cases() -> Iterator[str]:
'å', # mpt 'å', # mpt
'\U000ac517', # utf-8 encode error, falcon '\U000ac517', # utf-8 encode error, falcon
'\U000522f4', # utf-8 encode error, starcoder '\U000522f4', # utf-8 encode error, starcoder
"<s><s><unk><s>a<s>b<s>c<unk>d<unk></s>",
"<s> <s> <unk><s>a<s>b<s>c<unk>d<unk></s>",
] ]
@ -334,7 +336,7 @@ def generator_unicodes() -> Iterator[str]:
return False return False
# if cpt == 0x2029: # deepseek-llm # if cpt == 0x2029: # deepseek-llm
# return False # return False
if unicodedata.category(chr(cpt)) in ( "Cn", "Cs", "Co" ): # undefined, surrogates, private if unicodedata.category(chr(cpt)) in ("Cn", "Cs", "Co"): # undefined, surrogates, private
return False return False
return True return True
@ -426,6 +428,7 @@ def compare_tokenizers(tokenizer1: TokenizerGroundtruth, tokenizer2: TokenizerLl
t_start = time.perf_counter() t_start = time.perf_counter()
encode_errors = 0 encode_errors = 0
decode_errors = 0 decode_errors = 0
MAX_ERRORS = 10
logger.info("%s: %s" % (generator.__name__, "ini")) logger.info("%s: %s" % (generator.__name__, "ini"))
for text in generator: for text in generator:
@ -444,7 +447,7 @@ def compare_tokenizers(tokenizer1: TokenizerGroundtruth, tokenizer2: TokenizerLl
t_encode2 += t2 - t1 t_encode2 += t2 - t1
t_decode1 += t3 - t2 t_decode1 += t3 - t2
t_decode2 += t4 - t3 t_decode2 += t4 - t3
if ids1 != ids2: if encode_errors < MAX_ERRORS and ids1 != ids2:
i = find_first_mismatch(ids1, ids2) i = find_first_mismatch(ids1, ids2)
ids1 = list(ids1)[max(0, i - 2) : i + 5 + 1] ids1 = list(ids1)[max(0, i - 2) : i + 5 + 1]
ids2 = list(ids2)[max(0, i - 2) : i + 5 + 1] ids2 = list(ids2)[max(0, i - 2) : i + 5 + 1]
@ -452,7 +455,7 @@ def compare_tokenizers(tokenizer1: TokenizerGroundtruth, tokenizer2: TokenizerLl
logger.error(" Result: " + str(ids2)) logger.error(" Result: " + str(ids2))
encode_errors += 1 encode_errors += 1
logger.error(f" {encode_errors=}") logger.error(f" {encode_errors=}")
if not check_detokenizer(text, text1, text2): if decode_errors < MAX_ERRORS and not check_detokenizer(text, text1, text2):
i = find_first_mismatch(text1, text2) i = find_first_mismatch(text1, text2)
text1 = list(text1[max(0, i - 2) : i + 5 + 1]) text1 = list(text1[max(0, i - 2) : i + 5 + 1])
text2 = list(text2[max(0, i - 2) : i + 5 + 1]) text2 = list(text2[max(0, i - 2) : i + 5 + 1])
@ -460,7 +463,7 @@ def compare_tokenizers(tokenizer1: TokenizerGroundtruth, tokenizer2: TokenizerLl
logger.error(" Result: " + " ".join(hex(ord(x)) for x in text2)) logger.error(" Result: " + " ".join(hex(ord(x)) for x in text2))
decode_errors += 1 decode_errors += 1
logger.error(f" {decode_errors=}") logger.error(f" {decode_errors=}")
if encode_errors >= 10 or decode_errors >= 10: if encode_errors >= MAX_ERRORS and decode_errors >= MAX_ERRORS:
logger.error(f" EXIT: {encode_errors=} {decode_errors=}") logger.error(f" EXIT: {encode_errors=} {decode_errors=}")
# raise Exception() # raise Exception()
break break