vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-12 18:09:42 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

Merge branch 'upstream' into concedo_experimental

Browse source 
# Conflicts:
#	.flake8
#	.github/workflows/bench.yml
#	.github/workflows/python-lint.yml
#	.pre-commit-config.yaml
#	Makefile
#	README.md
#	models/ggml-vocab-bert-bge.gguf.inp
#	models/ggml-vocab-bert-bge.gguf.out
#	models/ggml-vocab-deepseek-coder.gguf.inp
#	models/ggml-vocab-deepseek-coder.gguf.out
#	models/ggml-vocab-deepseek-llm.gguf.inp
#	models/ggml-vocab-deepseek-llm.gguf.out
#	models/ggml-vocab-falcon.gguf.inp
#	models/ggml-vocab-falcon.gguf.out
#	models/ggml-vocab-gpt-2.gguf.inp
#	models/ggml-vocab-gpt-2.gguf.out
#	models/ggml-vocab-llama-bpe.gguf.inp
#	models/ggml-vocab-llama-bpe.gguf.out
#	models/ggml-vocab-llama-spm.gguf.inp
#	models/ggml-vocab-llama-spm.gguf.out
#	models/ggml-vocab-mpt.gguf.inp
#	models/ggml-vocab-mpt.gguf.out
#	models/ggml-vocab-phi-3.gguf
#	models/ggml-vocab-phi-3.gguf.inp
#	models/ggml-vocab-phi-3.gguf.out
#	models/ggml-vocab-refact.gguf
#	models/ggml-vocab-starcoder.gguf.inp
#	models/ggml-vocab-starcoder.gguf.out
#	requirements/requirements-convert.txt
#	scripts/compare-llama-bench.py
#	scripts/run-with-preset.py
#	scripts/verify-checksum-models.py
#	tests/CMakeLists.txt
#	tests/test-tokenizer-0.cpp
This commit is contained in:
Concedo 2024-05-06 18:09:45 +08:00
commit 6c000cbe7a
40 changed files with 1593 additions and 936 deletions
Download patch file Download diff file Expand all files Collapse all files

27
llama.cpp
View file

@ -2389,7 +2389,7 @@ static bool llama_kv_cache_init(
cache.recurrent = model.arch == LLM_ARCH_MAMBA;
cache.v_trans = !cparams.flash_attn;
// TODO: support mixed reccurent Transformer architectues
// TODO: support mixed recurrent Transformer architectures
// NOTE: (!a || b) is a logical implication (a -> b)
GGML_ASSERT(!cache.recurrent || n_embd_k_gqa == hparams.n_embd_k_s());
GGML_ASSERT(!cache.recurrent || n_embd_v_gqa == hparams.n_embd_v_s());
@ -4437,6 +4437,12 @@ static void llm_load_vocab(
} else if (
tokenizer_pre == "gpt-2") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_GPT2;
} else if (
tokenizer_pre == "refact") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_REFACT;
} else if (
tokenizer_pre == "command-r") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_COMMAND_R;
} else {
throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str()));
}
@ -12022,7 +12028,7 @@ static bool llama_is_user_defined_token(const llama_vocab& vocab, llama_token id
static uint8_t llama_token_to_byte(const llama_vocab& vocab, llama_token id) {
GGML_ASSERT(llama_vocab_get_type(vocab) != LLAMA_VOCAB_TYPE_NONE);
GGML_ASSERT(llama_is_byte_token(vocab, id));
const auto& token_data = vocab.id_to_token.at(id);
const auto & token_data = vocab.id_to_token.at(id);
switch (llama_vocab_get_type(vocab)) {
case LLAMA_VOCAB_TYPE_SPM: {
auto buf = token_data.text.substr(3, 2);
@ -12503,14 +12509,13 @@ struct llm_tokenizer_bpe {
"\\s?\\p{L}+",
"\\s?\\p{P}+",
"[一-龥ࠀ-一가-퟿]+",
"\\p{N}+",
"\\p{N}",
});
break;
case LLAMA_VOCAB_PRE_TYPE_FALCON:
word_collection = unicode_regex_split(text, {
"[\\p{P}\\$\\+<=>\\^~\\|]+",
"'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)",
"\\p{N}+",
"[0-9][0-9][0-9]",
});
break;
@ -12526,6 +12531,13 @@ struct llm_tokenizer_bpe {
});
break;
case LLAMA_VOCAB_PRE_TYPE_STARCODER:
case LLAMA_VOCAB_PRE_TYPE_REFACT:
case LLAMA_VOCAB_PRE_TYPE_COMMAND_R:
word_collection = unicode_regex_split(text, {
"\\p{N}",
"'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)",
});
break;
case LLAMA_VOCAB_PRE_TYPE_GPT2:
word_collection = unicode_regex_split(text, {
"'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)",
@ -17782,9 +17794,10 @@ int32_t llama_tokenize(
static std::string llama_decode_text(const std::string & text) {
std::string decoded_text;
auto unicode_sequences = unicode_cpts_from_utf8(text);
for (auto & unicode_sequence : unicode_sequences) {
decoded_text += unicode_utf8_to_byte(unicode_cpt_to_utf8(unicode_sequence));
const auto cpts = unicode_cpts_from_utf8(text);
for (const auto cpt : cpts) {
decoded_text += unicode_utf8_to_byte(unicode_cpt_to_utf8(cpt));
}
return decoded_text;