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

llama : towards llama3 tokenization support (wip)

Browse source
This commit is contained in:
Georgi Gerganov 2024-04-26 14:55:03 +03:00
parent ed42711b90
commit 4907e41aa7
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
8 changed files with 298 additions and 121 deletions
Download patch file Download diff file Expand all files Collapse all files

11
tests/CMakeLists.txt
View file

@ -41,12 +41,13 @@ llama_test(test-quantize-perf.cpp)
llama_test(test-sampling.cpp)
llama_test(test-chat-template.cpp)
# TODO: tmp disabled LLaMA v3 and Deepseek tests
llama_test(test-tokenizer-0-llama.cpp NAME test-tokenizer-0-llama ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama.gguf)
#llama_test(test-tokenizer-0-llama-v3.cpp NAME test-tokenizer-0-llama-v3 ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama-v3.gguf)
llama_test(test-tokenizer-0-falcon.cpp NAME test-tokenizer-0-falcon ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-falcon.gguf)
llama_test(test-tokenizer-0-llama.cpp NAME test-tokenizer-0-llama ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama.gguf)
llama_test(test-tokenizer-0-falcon.cpp NAME test-tokenizer-0-falcon ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-falcon.gguf)
llama_test(test-tokenizer-0-deepseek-coder.cpp NAME test-tokenizer-0-deepseek-coder ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-deepseek-coder.gguf)
llama_test(test-tokenizer-0-deepseek-llm.cpp NAME test-tokenizer-0-deepseek-llm ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-deepseek-llm.gguf)
#llama_test(test-tokenizer-0-deepseek-coder.cpp NAME test-tokenizer-0-deepseek-coder ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-deepseek-coder.gguf)
#llama_test(test-tokenizer-0-deepseek-llm.cpp NAME test-tokenizer-0-deepseek-llm ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-deepseek-llm.gguf)
llama_test(test-tokenizer-1-llama.cpp NAME test-tokenizer-1-llama ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama.gguf)
llama_test(test-tokenizer-1-llama.cpp NAME test-tokenizer-1-baichuan ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-baichuan.gguf)

4
tests/test-tokenizer-0-deepseek-coder.cpp
View file

@ -89,8 +89,8 @@ int main(int argc, char **argv) {
}
}
if (llama_vocab_type(model) != LLAMA_VOCAB_TYPE_DEEPSEEKCODER) {
fprintf(stderr, "%s : error: vocab type is not DEEPSEEKCODER\n", __func__);
if (llama_vocab_type(model) != LLAMA_VOCAB_TYPE_BPE) {
fprintf(stderr, "%s : error: vocab type is not BPE\n", __func__);
llama_free_model(model);
llama_free(ctx);
return 2;

4
tests/test-tokenizer-0-deepseek-llm.cpp
View file

@ -89,8 +89,8 @@ int main(int argc, char **argv) {
}
}
if (llama_vocab_type(model) != LLAMA_VOCAB_TYPE_DEEPSEEKLLM) {
fprintf(stderr, "%s : error: vocab type is not DEEPSEEKLLM\n", __func__);
if (llama_vocab_type(model) != LLAMA_VOCAB_TYPE_BPE) {
fprintf(stderr, "%s : error: vocab type is not BPE\n", __func__);
llama_free_model(model);
llama_free(ctx);
return 2;

197
tests/test-tokenizer-0-llama-v3.cpp Normal file
View file

@ -0,0 +1,197 @@
#include "llama.h"
#include "common.h"
#include "console.h"
#include <cstdio>
#include <string>
#include <map>
#include <vector>
#include <fstream>
// generate using test-tokenizer-0-llama.py
static const std::map<std::string, std::vector<llama_token>> & k_tests() {
static std::map<std::string, std::vector<llama_token>> _k_tests = {
{ "" , { }, },
{ " " , { 220, }, },
{ " " , { 256, }, },
{ " " , { 262, }, },
{ "\t" , { 197, }, },
{ "\n" , { 198, }, },
{ "\t\n" , { 1602, }, },
{ "Hello world" , { 9906, 1917, }, },
{ " Hello world" , { 22691, 1917, }, },
{ "Hello World" , { 9906, 4435, }, },
{ " Hello World" , { 22691, 4435, }, },
{ " Hello World!" , { 22691, 4435, 0, }, },
{ "Hello, world!" , { 9906, 11, 1917, 0, }, },
{ " Hello, world!" , { 22691, 11, 1917, 0, }, },
{ " this is 🦙.cpp" , { 420, 374, 11410, 99, 247, 13, 11055, }, },
{ "w048 7tuijk dsdfhu" , { 86, 23904, 220, 22, 83, 2005, 42908, 11729, 3013, 17156, }, },
{ "нещо на Български" , { 79862, 102118, 13373, 64571, 34694, 3114, 112203, 80112, }, },
{ "កាន់តែពិសេសអាចខលចេញ" , { 21549, 222, 98629, 241, 45358, 233, 21549, 237, 45358, 224, 21549, 244, 21549, 115, 21549, 253, 45358, 223, 21549, 253, 21549, 95, 98629, 227, 21549, 223, 21549, 249, 21549, 227, 45358, 223, 21549, 231, }, },
{ "🚀 (normal) 😶‍🌫️ (multiple emojis concatenated) ✅ (only emoji that has its own token)", { 9468, 248, 222, 320, 8416, 8, 27623, 114, 102470, 9468, 234, 104, 31643, 320, 36773, 100166, 98634, 8, 26602, 227, 320, 3323, 43465, 430, 706, 1202, 1866, 4037, 8, }, },
{ "Hello" , { 9906, }, },
{ " Hello" , { 22691, }, },
{ " Hello" , { 220, 22691, }, },
{ " Hello" , { 256, 22691, }, },
{ " Hello" , { 262, 22691, }, },
{ " Hello\n Hello" , { 262, 22691, 198, 262, 22691, }, },
{ " (" , { 320, }, },
{ "\n =" , { 198, 284, }, },
{ "' era" , { 6, 11639, }, },
{ "Hello, y'all! How are you 😁 ?我想在apple工作1314151天～", { 9906, 11, 379, 65948, 0, 2650, 527, 499, 27623, 223, 949, 37046, 101067, 19000, 23182, 102301, 9263, 18136, 16, 36827, 21909, }, },
{ "3" , { 18, }, },
{ "33" , { 1644, }, },
{ "333" , { 8765, }, },
{ "3333" , { 8765, 18, }, },
{ "33333" , { 8765, 1644, }, },
{ "333333" , { 8765, 8765, }, },
{ "3333333" , { 8765, 8765, 18, }, },
{ "33333333" , { 8765, 8765, 1644, }, },
{ "333333333" , { 8765, 8765, 8765, }, },
};
return _k_tests;
}
int main(int argc, char **argv) {
if (argc < 2) {
fprintf(stderr, "Usage: %s vocab-file [text-file]\n", argv[0]);
return 1;
}
const std::string fname = argv[1];
std::string fname_text;
if (argc > 2) {
fname_text = argv[2];
}
fprintf(stderr, "%s : reading vocab from: '%s'\n", __func__, fname.c_str());
llama_model * model;
llama_context * ctx;
llama_backend_init();
// load the vocab
{
auto mparams = llama_model_default_params();
mparams.vocab_only = true;
model = llama_load_model_from_file(fname.c_str(), mparams);
if (model == NULL) {
fprintf(stderr, "%s: error: failed to load vocab '%s'\n", __func__, fname.c_str());
return 1;
}
auto cparams = llama_context_default_params();
ctx = llama_new_context_with_model(model, cparams);
if (ctx == NULL) {
fprintf(stderr, "%s: error: failed to load vocab '%s'\n", __func__, fname.c_str());
llama_free_model(model);
return 1;
}
}
if (llama_vocab_type(model) != LLAMA_VOCAB_TYPE_BPE) {
fprintf(stderr, "%s : error: vocab type is not BPE\n", __func__);
llama_free_model(model);
llama_free(ctx);
return 2;
}
#ifdef _WIN32
// We need this for unicode console support
console::init(false, false);
atexit([]() { console::cleanup(); });
#endif
bool success = true;
for (const auto & test_kv : k_tests()) {
const std::vector<llama_token> res = llama_tokenize(ctx, test_kv.first, false);
printf("\n");
printf("src: '%s'\n", test_kv.first.c_str());
printf("res: '%s'\n", llama_detokenize_bpe(ctx, res).c_str());
printf("tok: ");
for (const auto & tok : res) {
printf("%d ", tok);
}
printf("\n");
bool correct = res.size() == test_kv.second.size();
for (int i = 0; i < (int) res.size() && correct; ++i) {
if (test_kv.second[i] != res[i]) {
correct = false;
}
}
if (!correct) {
fprintf(stderr, "%s : failed test: '%s'\n", __func__, test_kv.first.c_str());
fprintf(stderr, "%s : detokenized to: '%s' instead of '%s'\n", __func__,
llama_detokenize_bpe(ctx, res).c_str(),
llama_detokenize_bpe(ctx, test_kv.second).c_str());
fprintf(stderr, "%s : expected tokens: ", __func__);
for (const auto & t : test_kv.second) {
fprintf(stderr, "%6d, ", t);
}
fprintf(stderr, "\n");
fprintf(stderr, "%s : got tokens: ", __func__);
for (const auto & t : res) {
fprintf(stderr, "%6d, ", t);
}
fprintf(stderr, "\n");
success = false;
}
}
if (!fname_text.empty()) {
fprintf(stderr, "%s : tokenizing: '%s'\n", __func__, fname_text.c_str());
std::string text;
{
std::ifstream ifs(fname_text);
if (!ifs) {
fprintf(stderr, "%s : error: could not open file '%s'\n", __func__, fname_text.c_str());
return 1;
}
text = std::string(std::istreambuf_iterator<char>(ifs), std::istreambuf_iterator<char>());
}
fprintf(stderr, "%s : text size: %zu\n", __func__, text.size());
const std::vector<llama_token> res = llama_tokenize(ctx, text, false);
fprintf(stderr, "%s : tokens: %zu\n", __func__, res.size());
{
const std::string fname_out = fname_text + ".tokcpp";
std::ofstream ofs(fname_out);
if (!ofs) {
fprintf(stderr, "%s : error: could not open file '%s'\n", __func__, fname_out.c_str());
return 1;
}
for (const auto & tok : res) {
ofs << tok << " '" << llama_detokenize_bpe(ctx, std::vector<int>{tok}) << "'" << std::endl;
}
}
fprintf(stderr, "%s : tokens written to '%s'\n", __func__, (fname_text + ".tokcpp").c_str());
}
llama_free_model(model);
llama_free(ctx);
llama_backend_free();
return success ? 0 : 3;
}

2
tests/test-tokenizer-0-llama.cpp
View file

@ -126,7 +126,7 @@ int main(int argc, char **argv) {
}
printf("\n");
bool correct = res_nobos.size() == test_kv.second.size() && res_bos.size() == res_nobos.size() + 1 && res_bos[0] == 1;
bool correct = res_nobos.size() == test_kv.second.size() && res_bos.size() == res_nobos.size() + 1 && res_bos[0] == llama_token_bos(model);
for (int i = 0; i < (int) res_nobos.size() && correct; ++i) {
if (test_kv.second[i] != res_bos[i + 1]) {