Skip to content

Tokenizer SPM fixes for phi-3 and llama-spm (bugfix) #7425

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
May 21, 2024
Merged

Tokenizer SPM fixes for phi-3 and llama-spm (bugfix) #7425

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
c78a53a
Style
May 21, 2024
3d490e8
Update brute force test: add_special
May 21, 2024
9b21dc3
Enable rtrim when pre-inserting BOS
May 21, 2024
f2340b4
Default values for add_bos_token and add_eos_token
May 21, 2024
0594bfd
Revert "server : fix test regexes"
ggerganov May 21, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions convert-hf-to-gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1749,7 +1749,7 @@ def set_vocab(self):
token_id = int(token_id)
token = foken_data["content"].encode("utf-8")
if toktypes[token_id] != SentencePieceTokenTypes.UNKNOWN:
assert(tokens[token_id] == token)
assert tokens[token_id] == token
tokens[token_id] = token
scores[token_id] = -1000.0
toktypes[token_id] = SentencePieceTokenTypes.USER_DEFINED
Expand All @@ -1765,7 +1765,7 @@ def set_vocab(self):
token_id = int(foken_data["id"])
token = foken_data["content"].encode("utf-8")
if toktypes[token_id] != SentencePieceTokenTypes.UNKNOWN:
assert(tokens[token_id] == token)
assert tokens[token_id] == token
tokens[token_id] = token
scores[token_id] = -1000.0
toktypes[token_id] = SentencePieceTokenTypes.USER_DEFINED
Expand Down
10 changes: 5 additions & 5 deletions examples/server/tests/features/server.feature
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,8 +37,8 @@ Feature: llama.cpp server

Examples: Prompts
| prompt | n_predict | re_content | n_prompt | n_predicted | truncated |
| I believe the meaning of life is | 8 | (read\|going\|pretty)+ | 18 | 8 | not |
| Write a joke about AI from a very long prompt which will not be truncated | 256 | (princesses\|everyone\|kids\|Anna\|forest)+ | 45 | 64 | not |
| I believe the meaning of life is | 8 | (read\|going)+ | 18 | 8 | not |
| Write a joke about AI from a very long prompt which will not be truncated | 256 | (princesses\|everyone\|kids\|Anna\|forest)+ | 46 | 64 | not |

Scenario: Completion prompt truncated
Given a prompt:
Expand Down Expand Up @@ -67,8 +67,8 @@ Feature: llama.cpp server

Examples: Prompts
| model | system_prompt | user_prompt | max_tokens | re_content | n_prompt | n_predicted | enable_streaming | truncated |
| llama-2 | Book | What is the best book | 8 | (Here\|what)+ | 76 | 8 | disabled | not |
| codellama70b | You are a coding assistant. | Write the fibonacci function in c++. | 128 | (thanks\|happy\|bird\|fireplace)+ | -1 | 64 | enabled | |
| llama-2 | Book | What is the best book | 8 | (Here\|what)+ | 77 | 8 | disabled | not |
| codellama70b | You are a coding assistant. | Write the fibonacci function in c++. | 128 | (thanks\|happy\|bird\|Annabyear)+ | -1 | 64 | enabled | |


Scenario Outline: OAI Compatibility w/ response format
Expand All @@ -84,7 +84,7 @@ Feature: llama.cpp server
| response_format | n_predicted | re_content |
| {"type": "json_object", "schema": {"const": "42"}} | 5 | "42" |
| {"type": "json_object", "schema": {"items": [{"type": "integer"}]}} | 10 | \[ -300 \] |
| {"type": "json_object"} | 10 | \{ " Saragine. |
| {"type": "json_object"} | 10 | \{ " Jacky. |


Scenario: Tokenize / Detokenize
Expand Down
4 changes: 2 additions & 2 deletions examples/server/tests/features/slotsave.feature
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ Feature: llama.cpp server slot management
# Since we have cache, this should only process the last tokens
Given a user prompt "What is the capital of Germany?"
And a completion request with no api error
Then 24 tokens are predicted matching (Thank|special|Lily)
Then 24 tokens are predicted matching (Thank|special)
And 7 prompt tokens are processed
# Loading the original cache into slot 0,
# we should only be processing 1 prompt token and get the same output
Expand All @@ -41,7 +41,7 @@ Feature: llama.cpp server slot management
Given a user prompt "What is the capital of Germany?"
And using slot id 1
And a completion request with no api error
Then 24 tokens are predicted matching (Thank|special|Lily)
Then 24 tokens are predicted matching (Thank|special)
And 1 prompt tokens are processed

Scenario: Erase Slot
Expand Down
9 changes: 5 additions & 4 deletions llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12498,15 +12498,16 @@ static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab &
// tokenizer.encode('', add_special_tokens=True) returns [1]
// tokenizer.encode('', add_special_tokens=False) returns []

static const bool rtrim = true; //TODO: as param
bool is_prev_special = false;
bool special_token_rtrim = false;

if (add_special && vocab.special_add_bos != 0) {
GGML_ASSERT(vocab.special_bos_id != -1);
output.push_back(vocab.special_bos_id);
is_prev_special = true;
}

static const bool rtrim = true; //TODO: as param
bool is_prev_special = false;
bool special_token_rtrim = false;

for (const auto & fragment : fragment_buffer) {
if (fragment.type == FRAGMENT_BUFFER_VARIANT_TYPE_RAW_TEXT) {
// without adding this leading whitespace, we do not get the same results as the original tokenizer
Expand Down
22 changes: 13 additions & 9 deletions tests/test-tokenizer-random.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -154,19 +154,22 @@ def generator_custom_text_edge_cases() -> Iterator[str]:
'\uFEFF//', # unicode_ranges_control, 0xFEFF (BOM)
'Cửa Việt', # llama-3, ignore_merges = true
'<s>a', # Phi-3 fail
'<unk><|endoftext|><s>' # Phi-3 fail
'<unk><|endoftext|><s>', # Phi-3 fail
'a\na', # TODO: Bert fail
]


def generator_random_special_tokens(special_tokens:list[str], iterations=100) -> Iterator[str]:
special_tokens = set(special_tokens)
def generator_random_special_tokens(tokenizer, iterations=100) -> Iterator[str]:
special_tokens = set(tokenizer.all_special_tokens)
special_tokens.update([" ", "\n", "\t", "-", "!", "one", "1", "<s>", "</s>"])
special_tokens = list(sorted(special_tokens))
rand = random.Random()
for m in range(iterations):
rand.seed(m)
words = rand.choices(special_tokens, k=500)
if tokenizer.add_bos_token: # skip spam warning of double BOS
while words and words[0] == tokenizer.bos_token:
words.pop(0)
yield "".join(words)


Expand Down Expand Up @@ -290,18 +293,19 @@ def main(argv: list[str] = None):
model = LibLlamaModel(LibLlama(), args.vocab_file, mparams=dict(vocab_only=True), cparams=dict(n_ctx=4096))
tokenizer = AutoTokenizer.from_pretrained(args.dir_tokenizer)

def func_tokenize2(text: str):
return tokenizer.encode(text, add_special_tokens=False)

parse_special = all(len(func_tokenize2(t)) == 1 for t in tokenizer.all_special_tokens)
tokenizer.add_bos_token = getattr(tokenizer, "add_bos_token", True)
tokenizer.add_eos_token = getattr(tokenizer, "add_eos_token", False)

def func_tokenize1(text: str):
return model.tokenize(text, add_special=False, parse_special=parse_special)
return model.tokenize(text, add_special=True, parse_special=True)

def func_tokenize2(text: str):
return tokenizer.encode(text, add_special_tokens=True)

vocab = list(sorted(tokenizer.batch_decode(list(tokenizer.get_vocab().values()), skip_special_tokens=True)))
test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_custom_text())
test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_custom_text_edge_cases())
test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_special_tokens(tokenizer.all_special_tokens, 10_000))
test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_special_tokens(tokenizer, 10_000))
test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_vocab_words(vocab))
test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_chars(10_000))
test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_vocab_chars(vocab, 10_000))
Expand Down
Loading