Skip to content

Add test for MPT tokenization #3728

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 4 commits into from
Oct 22, 2023
Merged

Add test for MPT tokenization #3728

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
6a94ae6
Add test for MPT tokenization
goerch Oct 22, 2023
8ee7363
Revert code motion
goerch Oct 22, 2023
1dc1316
Remove unnecessary restriction in test case
goerch Oct 22, 2023
74204cc
Clarify logic in conversion
goerch Oct 22, 2023
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
15 changes: 11 additions & 4 deletions convert-mpt-hf-to-gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -128,15 +128,22 @@ def parse_args() -> argparse.Namespace:
# ref: https://github.com/cmp-nct/ggllm.cpp/blob/master/falcon_convert.py
tokenizer = AutoTokenizer.from_pretrained(dir_model)

added_vocab = tokenizer.get_added_vocab()
reverse_vocab = {id: encoded_tok for encoded_tok, id in tokenizer.vocab.items()}

for i in range(vocab_size):
tokens.append(reverse_vocab[i] if i in reverse_vocab else f"[PAD{i}]")
scores.append(0.0) # dummy
toktypes.append(gguf.TokenType.NORMAL)
if i not in reverse_vocab:
tokens.append(f"[PAD{i}]")
toktypes.append(gguf.TokenType.USER_DEFINED)
elif reverse_vocab[i] in added_vocab:
# NOTE: wouldn't we like to distinguish CONTROL tokens here?
tokens.append(reverse_vocab[i])
toktypes.append(gguf.TokenType.USER_DEFINED)
else:
tokens.append(reverse_vocab[i])
toktypes.append(gguf.TokenType.NORMAL)

gguf_writer.add_token_list(tokens)
gguf_writer.add_token_scores(scores)
gguf_writer.add_token_types(toktypes)

special_vocab = gguf.SpecialVocab(dir_model, load_merges = True)
Expand Down
17 changes: 9 additions & 8 deletions llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -975,14 +975,15 @@ static void llama_nop(struct ggml_tensor * tensor) { // don't offload by default
(void) tensor;
}

static std::string llama_token_to_str(const struct llama_context * ctx, llama_token token) {
static std::string llama_token_to_piece(const struct llama_context * ctx, llama_token token) {
std::vector<char> result(8, 0);
const int n_tokens = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size());
if (n_tokens < 0) {
result.resize(-n_tokens);
int check = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size());
GGML_ASSERT(check == -n_tokens);
} else {
}
else {
result.resize(n_tokens);
}

Expand Down Expand Up @@ -1202,10 +1203,10 @@ struct llama_vocab {
id special_eot_id = 32010;

int find_bpe_rank(std::string token_left, std::string token_right) const {
replace_all(token_left, " ", "\u0120");
replace_all(token_left, "\n", "\u010A");
replace_all(token_right, " ", "\u0120");
replace_all(token_right, "\n", "\u010A");
GGML_ASSERT(token_left.find(" ") == std::string::npos);
GGML_ASSERT(token_left.find("\n") == std::string::npos);
GGML_ASSERT(token_right.find(" ") == std::string::npos);
GGML_ASSERT(token_right.find("\n") == std::string::npos);

auto it = bpe_ranks.find(std::make_pair(token_left, token_right));
if (it == bpe_ranks.end()) {
Expand Down Expand Up @@ -7480,7 +7481,7 @@ void llama_sample_grammar(struct llama_context * ctx, llama_token_data_array * c

for (size_t i = 0; i < candidates->size; ++i) {
const llama_token id = candidates->data[i].id;
const std::string piece = llama_token_to_str(ctx, id);
const std::string piece = llama_token_to_piece(ctx, id);
if (id == eos) {
if (!allow_eos) {
candidates->data[i].logit = -INFINITY;
Expand Down Expand Up @@ -7692,7 +7693,7 @@ void llama_grammar_accept_token(struct llama_context * ctx, struct llama_grammar
GGML_ASSERT(false);
}

const std::string piece = llama_token_to_str(ctx, token);
const std::string piece = llama_token_to_piece(ctx, token);

// Note terminating 0 in decoded string
const auto decoded = decode_utf8(piece.c_str(), grammar->partial_utf8);
Expand Down
Binary file added models/ggml-vocab-mpt.gguf
View file
Open in desktop
Binary file not shown.
1 change: 1 addition & 0 deletions tests/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,7 @@ llama_test_executable (test-tokenizer-1-llama test-tokenizer-1-llama.cpp ${CMAKE
llama_build_executable(test-tokenizer-1-bpe.cpp)
llama_test_executable (test-tokenizer-1-falcon test-tokenizer-1-bpe.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-falcon.gguf)
llama_test_executable(test-tokenizer-1-aquila test-tokenizer-1-bpe.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-aquila.gguf)
llama_test_executable(test-tokenizer-1-mpt test-tokenizer-1-bpe.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-mpt.gguf)
llama_build_and_test_executable(test-grammar-parser.cpp)
llama_build_and_test_executable(test-llama-grammar.cpp)
llama_build_and_test_executable(test-grad0.cpp) # SLOW
Expand Down