From 4f7e4b5e19aa5cf1c1e7e5f0a701c26a8538ede2 Mon Sep 17 00:00:00 2001 From: slaren Date: Wed, 9 Oct 2024 12:26:27 +0200 Subject: [PATCH] common : use common_ prefix for common library functions --- common/common.cpp | 82 +++++++++---------- common/common.h | 66 +++++++-------- common/sampling.cpp | 2 +- examples/batched-bench/batched-bench.cpp | 14 ++-- examples/batched/batched.cpp | 22 ++--- .../cvector-generator/cvector-generator.cpp | 10 +-- examples/embedding/embedding.cpp | 16 ++-- examples/eval-callback/eval-callback.cpp | 4 +- examples/export-lora/export-lora.cpp | 2 +- examples/gritlm/gritlm.cpp | 30 +++---- examples/imatrix/imatrix.cpp | 4 +- examples/infill/infill.cpp | 24 +++--- examples/llava/llava-cli.cpp | 20 ++--- examples/llava/minicpmv-cli.cpp | 8 +- examples/lookahead/lookahead.cpp | 26 +++--- examples/lookup/lookup-create.cpp | 4 +- examples/lookup/lookup-stats.cpp | 4 +- examples/lookup/lookup.cpp | 16 ++-- examples/main/main.cpp | 46 +++++------ examples/parallel/parallel.cpp | 18 ++-- examples/passkey/passkey.cpp | 22 ++--- examples/perplexity/perplexity.cpp | 36 ++++---- examples/retrieval/retrieval.cpp | 18 ++-- examples/save-load-state/save-load-state.cpp | 14 ++-- examples/server/server.cpp | 38 ++++----- examples/server/utils.hpp | 8 +- examples/simple/simple.cpp | 16 ++-- examples/speculative/speculative.cpp | 34 ++++---- examples/tokenize/tokenize.cpp | 4 +- tests/test-chat-template.cpp | 10 +-- tests/test-tokenizer-0.cpp | 14 ++-- tests/test-tokenizer-1-bpe.cpp | 10 +-- tests/test-tokenizer-1-spm.cpp | 10 +-- 33 files changed, 326 insertions(+), 326 deletions(-) diff --git a/common/common.cpp b/common/common.cpp index 29df16c95..64d6a5f8d 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -493,7 +493,7 @@ std::string string_from(const struct llama_context * ctx, const std::vector & lora_adapters) { +void common_lora_adapters_apply(struct llama_context * ctx, std::vector & lora_adapters) { llama_lora_adapter_clear(ctx); for (auto & la : lora_adapters) { if (la.scale != 0.0f) { @@ -970,7 +970,7 @@ void llama_lora_adapters_apply(struct llama_context * ctx, std::vector llama_tokenize( +std::vector common_tokenize( const struct llama_context * ctx, const std::string & text, bool add_special, bool parse_special) { - return llama_tokenize(llama_get_model(ctx), text, add_special, parse_special); + return common_tokenize(llama_get_model(ctx), text, add_special, parse_special); } -std::vector llama_tokenize( +std::vector common_tokenize( const struct llama_model * model, const std::string & text, bool add_special, @@ -1509,7 +1509,7 @@ std::vector llama_tokenize( return result; } -std::string llama_token_to_piece(const struct llama_context * ctx, llama_token token, bool special) { +std::string common_token_to_piece(const struct llama_context * ctx, llama_token token, bool special) { std::string piece; piece.resize(piece.capacity()); // using string internal cache, 15 bytes + '\n' const int n_chars = llama_token_to_piece(llama_get_model(ctx), token, &piece[0], piece.size(), 0, special); @@ -1525,7 +1525,7 @@ std::string llama_token_to_piece(const struct llama_context * ctx, llama_token t return piece; } -std::string llama_detokenize(llama_context * ctx, const std::vector & tokens, bool special) { +std::string common_detokenize(llama_context * ctx, const std::vector & tokens, bool special) { std::string text; text.resize(std::max(text.capacity(), tokens.size())); int32_t n_chars = llama_detokenize(llama_get_model(ctx), tokens.data(), (int32_t)tokens.size(), &text[0], (int32_t)text.size(), false, special); @@ -1551,9 +1551,9 @@ bool llama_chat_verify_template(const std::string & tmpl) { return res >= 0; } -std::string llama_chat_apply_template(const struct llama_model * model, +std::string common_chat_apply_template(const struct llama_model * model, const std::string & tmpl, - const std::vector & msgs, + const std::vector & msgs, bool add_ass) { int alloc_size = 0; bool fallback = false; // indicate if we must fallback to default chatml @@ -1595,42 +1595,42 @@ std::string llama_chat_apply_template(const struct llama_model * model, return formatted_chat; } -std::string llama_chat_format_single(const struct llama_model * model, +std::string common_chat_format_single(const struct llama_model * model, const std::string & tmpl, - const std::vector & past_msg, - const llama_chat_msg & new_msg, + const std::vector & past_msg, + const common_chat_msg & new_msg, bool add_ass) { std::ostringstream ss; - auto fmt_past_msg = past_msg.empty() ? "" : llama_chat_apply_template(model, tmpl, past_msg, false); - std::vector chat_new(past_msg); + auto fmt_past_msg = past_msg.empty() ? "" : common_chat_apply_template(model, tmpl, past_msg, false); + std::vector chat_new(past_msg); // if the past_msg ends with a newline, we must preserve it in the formatted version if (add_ass && !fmt_past_msg.empty() && fmt_past_msg.back() == '\n') { ss << "\n"; }; // format chat with new_msg chat_new.push_back(new_msg); - auto fmt_new_msg = llama_chat_apply_template(model, tmpl, chat_new, add_ass); + auto fmt_new_msg = common_chat_apply_template(model, tmpl, chat_new, add_ass); // get the diff part ss << fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size()); return ss.str(); } -std::string llama_chat_format_example(const struct llama_model * model, +std::string common_chat_format_example(const struct llama_model * model, const std::string & tmpl) { - std::vector msgs = { + std::vector msgs = { {"system", "You are a helpful assistant"}, {"user", "Hello"}, {"assistant", "Hi there"}, {"user", "How are you?"}, }; - return llama_chat_apply_template(model, tmpl, msgs, true); + return common_chat_apply_template(model, tmpl, msgs, true); } // // KV cache utils // -void llama_kv_cache_dump_view(const llama_kv_cache_view & view, int row_size) { +void common_kv_cache_dump_view(const llama_kv_cache_view & view, int row_size) { static const char slot_chars[] = ".123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz+"; printf("=== Dumping KV cache. total cells %d, max sequences per cell %d, populated cells %d, total tokens in cache %d, largest empty slot=%d @ %d", @@ -1653,7 +1653,7 @@ void llama_kv_cache_dump_view(const llama_kv_cache_view & view, int row_size) { printf("\n=== Done dumping\n"); } -void llama_kv_cache_dump_view_seqs(const llama_kv_cache_view & view, int row_size) { +void common_kv_cache_dump_view_seqs(const llama_kv_cache_view & view, int row_size) { static const char slot_chars[] = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"; printf("=== Dumping KV cache. total cells %d, max sequences per cell %d, populated cells %d, total tokens in cache %d, largest empty slot=%d @ %d\n", @@ -1705,7 +1705,7 @@ void llama_kv_cache_dump_view_seqs(const llama_kv_cache_view & view, int row_siz // Embedding utils // -void llama_embd_normalize(const float * inp, float * out, int n, int embd_norm) { +void common_embd_normalize(const float * inp, float * out, int n, int embd_norm) { double sum = 0.0; switch (embd_norm) { @@ -1739,7 +1739,7 @@ void llama_embd_normalize(const float * inp, float * out, int n, int embd_norm) } } -float llama_embd_similarity_cos(const float * embd1, const float * embd2, int n){ +float common_embd_similarity_cos(const float * embd1, const float * embd2, int n){ double sum = 0.0; double sum1 = 0.0; double sum2 = 0.0; @@ -1765,8 +1765,8 @@ float llama_embd_similarity_cos(const float * embd1, const float * embd2, int n) // Control vector utils // -static llama_control_vector_data llama_control_vector_load_one(const llama_control_vector_load_info & load_info) { - llama_control_vector_data result = { -1, {} }; +static common_control_vector_data llama_control_vector_load_one(const common_control_vector_load_info & load_info) { + common_control_vector_data result = { -1, {} }; ggml_context * ctx = nullptr; struct gguf_init_params meta_gguf_params = { @@ -1850,8 +1850,8 @@ static llama_control_vector_data llama_control_vector_load_one(const llama_contr return result; } -llama_control_vector_data llama_control_vector_load(const std::vector & load_infos) { - llama_control_vector_data result = { -1, {} }; +common_control_vector_data llama_control_vector_load(const std::vector & load_infos) { + common_control_vector_data result = { -1, {} }; for (const auto & info : load_infos) { auto cur = llama_control_vector_load_one(info); diff --git a/common/common.h b/common/common.h index 65add1f30..71453a629 100644 --- a/common/common.h +++ b/common/common.h @@ -24,12 +24,12 @@ #define DEFAULT_MODEL_PATH "models/7B/ggml-model-f16.gguf" -struct llama_lora_adapter_info { +struct common_lora_adapter_info { std::string path; float scale; }; -struct llama_lora_adapter_container : llama_lora_adapter_info { +struct common_lora_adapter_container : common_lora_adapter_info { struct llama_lora_adapter * adapter; }; @@ -39,7 +39,7 @@ extern char const * LLAMA_COMMIT; extern char const * LLAMA_COMPILER; extern char const * LLAMA_BUILD_TARGET; -struct llama_control_vector_load_info; +struct common_control_vector_load_info; // // CPU utils @@ -208,9 +208,9 @@ struct gpt_params { std::vector kv_overrides; bool lora_init_without_apply = false; // only load lora to memory, but do not apply it to ctx (user can manually apply lora later using llama_lora_adapter_apply) - std::vector lora_adapters; // lora adapter path with user defined scale + std::vector lora_adapters; // lora adapter path with user defined scale - std::vector control_vectors; // control vector with user defined scale + std::vector control_vectors; // control vector with user defined scale int32_t verbosity = 0; int32_t control_vector_layer_start = -1; // layer range for control vector @@ -404,29 +404,29 @@ std::string fs_get_cache_file(const std::string & filename); // Model utils // -struct llama_init_result { +struct common_init_result { struct llama_model * model = nullptr; struct llama_context * context = nullptr; - std::vector lora_adapters; + std::vector lora_adapters; }; -struct llama_init_result llama_init_from_gpt_params(gpt_params & params); +struct common_init_result llama_init_from_gpt_params(gpt_params & params); -struct llama_model_params llama_model_params_from_gpt_params (const gpt_params & params); -struct llama_context_params llama_context_params_from_gpt_params (const gpt_params & params); +struct llama_model_params common_model_params_from_gpt_params (const gpt_params & params); +struct llama_context_params common_context_params_from_gpt_params (const gpt_params & params); struct ggml_threadpool_params ggml_threadpool_params_from_cpu_params(const cpu_params & params); -struct llama_model * llama_load_model_from_url(const char * model_url, const char * path_model, const char * hf_token, const struct llama_model_params & params); -struct llama_model * llama_load_model_from_hf(const char * repo, const char * file, const char * path_model, const char * hf_token, const struct llama_model_params & params); +struct llama_model * common_load_model_from_url(const char * model_url, const char * path_model, const char * hf_token, const struct llama_model_params & params); +struct llama_model * common_load_model_from_hf(const char * repo, const char * file, const char * path_model, const char * hf_token, const struct llama_model_params & params); // clear LoRA adapters from context, then apply new list of adapters -void llama_lora_adapters_apply(struct llama_context * ctx, std::vector & lora_adapters); +void common_lora_adapters_apply(struct llama_context * ctx, std::vector & lora_adapters); // Batch utils -void llama_batch_clear(struct llama_batch & batch); +void common_batch_clear(struct llama_batch & batch); -void llama_batch_add( +void common_batch_add( struct llama_batch & batch, llama_token id, llama_pos pos, @@ -439,13 +439,13 @@ void llama_batch_add( // tokenizes a string into a vector of tokens // should work similar to Python's `tokenizer.encode` -std::vector llama_tokenize( +std::vector common_tokenize( const struct llama_context * ctx, const std::string & text, bool add_special, bool parse_special = false); -std::vector llama_tokenize( +std::vector common_tokenize( const struct llama_model * model, const std::string & text, bool add_special, @@ -453,7 +453,7 @@ std::vector llama_tokenize( // tokenizes a token into a piece, optionally renders special/control tokens // should work similar to Python's `tokenizer.id_to_piece` -std::string llama_token_to_piece( +std::string common_token_to_piece( const struct llama_context * ctx, llama_token token, bool special = true); @@ -461,7 +461,7 @@ std::string llama_token_to_piece( // detokenizes a vector of tokens into a string // should work similar to Python's `tokenizer.decode` // optionally renders special/control tokens -std::string llama_detokenize( +std::string common_detokenize( llama_context * ctx, const std::vector & tokens, bool special = true); @@ -471,7 +471,7 @@ std::string llama_detokenize( // // same with llama_chat_message, but uses std::string -struct llama_chat_msg { +struct common_chat_msg { std::string role; std::string content; }; @@ -482,20 +482,20 @@ bool llama_chat_verify_template(const std::string & tmpl); // CPP wrapper for llama_chat_apply_template // If the built-in template is not supported, we default to chatml // If the custom "tmpl" is not supported, we throw an error -std::string llama_chat_apply_template(const struct llama_model * model, +std::string common_chat_apply_template(const struct llama_model * model, const std::string & tmpl, - const std::vector & chat, + const std::vector & chat, bool add_ass); // Format single message, while taking into account the position of that message in chat history -std::string llama_chat_format_single(const struct llama_model * model, +std::string common_chat_format_single(const struct llama_model * model, const std::string & tmpl, - const std::vector & past_msg, - const llama_chat_msg & new_msg, + const std::vector & past_msg, + const common_chat_msg & new_msg, bool add_ass); // Returns an example of formatted chat -std::string llama_chat_format_example(const struct llama_model * model, +std::string common_chat_format_example(const struct llama_model * model, const std::string & tmpl); // @@ -503,31 +503,31 @@ std::string llama_chat_format_example(const struct llama_model * model, // // Dump the KV cache view with the number of sequences per cell. -void llama_kv_cache_dump_view(const llama_kv_cache_view & view, int row_size = 80); +void common_kv_cache_dump_view(const llama_kv_cache_view & view, int row_size = 80); // Dump the KV cache view showing individual sequences in each cell (long output). -void llama_kv_cache_dump_view_seqs(const llama_kv_cache_view & view, int row_size = 40); +void common_kv_cache_dump_view_seqs(const llama_kv_cache_view & view, int row_size = 40); // // Embedding utils // -void llama_embd_normalize(const float * inp, float * out, int n, int embd_norm = 2); +void common_embd_normalize(const float * inp, float * out, int n, int embd_norm = 2); -float llama_embd_similarity_cos(const float * embd1, const float * embd2, int n); +float common_embd_similarity_cos(const float * embd1, const float * embd2, int n); // // Control vector utils // -struct llama_control_vector_data { +struct common_control_vector_data { int n_embd; // stores data for layers [1, n_layer] where n_layer = data.size() / n_embd std::vector data; }; -struct llama_control_vector_load_info { +struct common_control_vector_load_info { float strength; std::string fname; @@ -535,7 +535,7 @@ struct llama_control_vector_load_info { // Load control vectors, scale each by strength, and add them together. // On error, returns {-1, empty} -llama_control_vector_data llama_control_vector_load(const std::vector & load_infos); +common_control_vector_data llama_control_vector_load(const std::vector & load_infos); // // Split utils diff --git a/common/sampling.cpp b/common/sampling.cpp index 3dc7f1120..ef4d24844 100644 --- a/common/sampling.cpp +++ b/common/sampling.cpp @@ -358,7 +358,7 @@ std::string gpt_sampler_prev_str(gpt_sampler * gsmpl, llama_context * ctx_main, GGML_ASSERT(id != LLAMA_TOKEN_NULL && "null token in the sampling history - should not happen"); - result += llama_token_to_piece(ctx_main, id); + result += common_token_to_piece(ctx_main, id); } return result; diff --git a/examples/batched-bench/batched-bench.cpp b/examples/batched-bench/batched-bench.cpp index 4a15941f1..ec58295d5 100644 --- a/examples/batched-bench/batched-bench.cpp +++ b/examples/batched-bench/batched-bench.cpp @@ -36,7 +36,7 @@ int main(int argc, char ** argv) { // initialize the model - llama_model_params model_params = llama_model_params_from_gpt_params(params); + llama_model_params model_params = common_model_params_from_gpt_params(params); llama_model * model = llama_load_model_from_file(params.model.c_str(), model_params); @@ -45,7 +45,7 @@ int main(int argc, char ** argv) { return 1; } - llama_context_params ctx_params = llama_context_params_from_gpt_params(params); + llama_context_params ctx_params = common_context_params_from_gpt_params(params); // ensure enough sequences are available ctx_params.n_seq_max = n_pl.empty() ? 1 : *std::max_element(n_pl.begin(), n_pl.end()); @@ -92,7 +92,7 @@ int main(int argc, char ** argv) { // warm up { for (int i = 0; i < 16; ++i) { - llama_batch_add(batch, 0, i, { 0 }, false); + common_batch_add(batch, 0, i, { 0 }, false); } if (!decode_helper(ctx, batch, ctx_params.n_batch)) { @@ -122,11 +122,11 @@ int main(int argc, char ** argv) { continue; } - llama_batch_clear(batch); + common_batch_clear(batch); for (int i = 0; i < pp; ++i) { for (int j = 0; j < (is_pp_shared ? 1 : pl); ++j) { - llama_batch_add(batch, 0, i, { j }, false); + common_batch_add(batch, 0, i, { j }, false); } } batch.logits[batch.n_tokens - 1] = true; @@ -151,10 +151,10 @@ int main(int argc, char ** argv) { const auto t_tg_start = ggml_time_us(); for (int i = 0; i < tg; ++i) { - llama_batch_clear(batch); + common_batch_clear(batch); for (int j = 0; j < pl; ++j) { - llama_batch_add(batch, 0, pp + i, { j }, true); + common_batch_add(batch, 0, pp + i, { j }, true); } if (!decode_helper(ctx, batch, ctx_params.n_batch)) { diff --git a/examples/batched/batched.cpp b/examples/batched/batched.cpp index 7887a43d6..da0d15e1d 100644 --- a/examples/batched/batched.cpp +++ b/examples/batched/batched.cpp @@ -39,7 +39,7 @@ int main(int argc, char ** argv) { // initialize the model - llama_model_params model_params = llama_model_params_from_gpt_params(params); + llama_model_params model_params = common_model_params_from_gpt_params(params); llama_model * model = llama_load_model_from_file(params.model.c_str(), model_params); @@ -51,13 +51,13 @@ int main(int argc, char ** argv) { // tokenize the prompt std::vector tokens_list; - tokens_list = ::llama_tokenize(model, params.prompt, true); + tokens_list = ::common_tokenize(model, params.prompt, true); const int n_kv_req = tokens_list.size() + (n_predict - tokens_list.size())*n_parallel; // initialize the context - llama_context_params ctx_params = llama_context_params_from_gpt_params(params); + llama_context_params ctx_params = common_context_params_from_gpt_params(params); ctx_params.n_ctx = n_kv_req; ctx_params.n_batch = std::max(n_predict, n_parallel); @@ -94,7 +94,7 @@ int main(int argc, char ** argv) { LOG("\n"); for (auto id : tokens_list) { - LOG("%s", llama_token_to_piece(ctx, id).c_str()); + LOG("%s", common_token_to_piece(ctx, id).c_str()); } // create a llama_batch @@ -108,7 +108,7 @@ int main(int argc, char ** argv) { // evaluate the initial prompt for (size_t i = 0; i < tokens_list.size(); ++i) { - llama_batch_add(batch, tokens_list[i], i, seq_ids, false); + common_batch_add(batch, tokens_list[i], i, seq_ids, false); } GGML_ASSERT(batch.n_tokens == (int) tokens_list.size()); @@ -123,8 +123,8 @@ int main(int argc, char ** argv) { decoder_start_token_id = llama_token_bos(model); } - llama_batch_clear(batch); - llama_batch_add(batch, decoder_start_token_id, 0, seq_ids, false); + common_batch_clear(batch); + common_batch_add(batch, decoder_start_token_id, 0, seq_ids, false); } // llama_decode will output logits only for the last token of the prompt @@ -161,7 +161,7 @@ int main(int argc, char ** argv) { while (n_cur <= n_predict) { // prepare the next batch - llama_batch_clear(batch); + common_batch_clear(batch); // sample the next token for each parallel sequence / stream for (int32_t i = 0; i < n_parallel; ++i) { @@ -185,15 +185,15 @@ int main(int argc, char ** argv) { // if there is only one stream, we print immediately to stdout if (n_parallel == 1) { - LOG("%s", llama_token_to_piece(ctx, new_token_id).c_str()); + LOG("%s", common_token_to_piece(ctx, new_token_id).c_str()); } - streams[i] += llama_token_to_piece(ctx, new_token_id); + streams[i] += common_token_to_piece(ctx, new_token_id); i_batch[i] = batch.n_tokens; // push this new token for next evaluation - llama_batch_add(batch, new_token_id, n_cur, { i }, true); + common_batch_add(batch, new_token_id, n_cur, { i }, true); n_decode += 1; } diff --git a/examples/cvector-generator/cvector-generator.cpp b/examples/cvector-generator/cvector-generator.cpp index 41bf4eb2a..cdaf8f390 100644 --- a/examples/cvector-generator/cvector-generator.cpp +++ b/examples/cvector-generator/cvector-generator.cpp @@ -31,7 +31,7 @@ template static std::string tokens_to_str(llama_context * ctx, Iter begin, Iter end) { std::string ret; for (; begin != end; ++begin) { - ret += llama_token_to_piece(ctx, *begin); + ret += common_token_to_piece(ctx, *begin); } return ret; @@ -272,8 +272,8 @@ struct tokenized_prompt { tokenized_prompt(llama_context * ctx, std::string pos, std::string neg) { const bool add_bos = llama_add_bos_token(llama_get_model(ctx)); - tokens_pos = ::llama_tokenize(ctx, pos, add_bos, true); - tokens_neg = ::llama_tokenize(ctx, neg, add_bos, true); + tokens_pos = ::common_tokenize(ctx, pos, add_bos, true); + tokens_neg = ::common_tokenize(ctx, neg, add_bos, true); max_seq_len = std::max(tokens_pos.size(), tokens_neg.size()); padding_seq(ctx, tokens_pos, max_seq_len); padding_seq(ctx, tokens_neg, max_seq_len); @@ -281,7 +281,7 @@ struct tokenized_prompt { void padding_seq(llama_context * ctx, std::vector & tokens, size_t len) { // TODO: customize padding token - std::vector pad_tokens = ::llama_tokenize(ctx, " ", false); + std::vector pad_tokens = ::common_tokenize(ctx, " ", false); llama_token pad_tok = pad_tokens.back(); while (tokens.size() < len) { tokens.push_back(pad_tok); @@ -413,7 +413,7 @@ int main(int argc, char ** argv) { llama_numa_init(params.numa); // load the model to get hparams - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; diff --git a/examples/embedding/embedding.cpp b/examples/embedding/embedding.cpp index 734926822..f5113ef85 100644 --- a/examples/embedding/embedding.cpp +++ b/examples/embedding/embedding.cpp @@ -28,7 +28,7 @@ static std::vector split_lines(const std::string & s, const std::st static void batch_add_seq(llama_batch & batch, const std::vector & tokens, llama_seq_id seq_id) { size_t n_tokens = tokens.size(); for (size_t i = 0; i < n_tokens; i++) { - llama_batch_add(batch, tokens[i], i, { seq_id }, true); + common_batch_add(batch, tokens[i], i, { seq_id }, true); } } @@ -74,7 +74,7 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu } float * out = output + embd_pos * n_embd; - llama_embd_normalize(embd, out, n_embd, embd_norm); + common_embd_normalize(embd, out, n_embd, embd_norm); } } @@ -95,7 +95,7 @@ int main(int argc, char ** argv) { llama_numa_init(params.numa); // load the model - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; @@ -135,7 +135,7 @@ int main(int argc, char ** argv) { // tokenize the prompts and trim std::vector> inputs; for (const auto & prompt : prompts) { - auto inp = ::llama_tokenize(ctx, prompt, true, true); + auto inp = ::common_tokenize(ctx, prompt, true, true); if (inp.size() > n_batch) { LOG_ERR("%s: number of tokens in input line (%lld) exceeds batch size (%lld), increase batch size and re-run\n", __func__, (long long int) inp.size(), (long long int) n_batch); @@ -159,7 +159,7 @@ int main(int argc, char ** argv) { LOG_INF("%s: prompt %d: '%s'\n", __func__, i, prompts[i].c_str()); LOG_INF("%s: number of tokens in prompt = %zu\n", __func__, inputs[i].size()); for (int j = 0; j < (int) inputs[i].size(); j++) { - LOG("%6d -> '%s'\n", inputs[i][j], llama_token_to_piece(ctx, inputs[i][j]).c_str()); + LOG("%6d -> '%s'\n", inputs[i][j], common_token_to_piece(ctx, inputs[i][j]).c_str()); } LOG("\n\n"); } @@ -199,7 +199,7 @@ int main(int argc, char ** argv) { batch_decode(ctx, batch, out, s, n_embd, params.embd_normalize); e += pooling_type == LLAMA_POOLING_TYPE_NONE ? batch.n_tokens : s; s = 0; - llama_batch_clear(batch); + common_batch_clear(batch); } // add to batch @@ -263,7 +263,7 @@ int main(int argc, char ** argv) { LOG("\n"); for (int i = 0; i < n_prompts; i++) { for (int j = 0; j < n_prompts; j++) { - float sim = llama_embd_similarity_cos(emb + i * n_embd, emb + j * n_embd, n_embd); + float sim = common_embd_similarity_cos(emb + i * n_embd, emb + j * n_embd, n_embd); LOG("%6.2f ", sim); } LOG("%1.10s", prompts[i].c_str()); @@ -296,7 +296,7 @@ int main(int argc, char ** argv) { for (int i = 0;;) { // at least two iteration (n_embd_count > 1) LOG(" ["); for (int j = 0;;) { // at least two iteration (n_embd_count > 1) - float sim = llama_embd_similarity_cos(emb + i * n_embd, emb + j * n_embd, n_embd); + float sim = common_embd_similarity_cos(emb + i * n_embd, emb + j * n_embd, n_embd); LOG("%6.2f", sim); j++; if (j < n_embd_count) LOG(", "); else break; diff --git a/examples/eval-callback/eval-callback.cpp b/examples/eval-callback/eval-callback.cpp index 6d629fe4e..ae08965c6 100644 --- a/examples/eval-callback/eval-callback.cpp +++ b/examples/eval-callback/eval-callback.cpp @@ -129,7 +129,7 @@ static bool ggml_debug(struct ggml_tensor * t, bool ask, void * user_data) { static bool run(llama_context * ctx, const gpt_params & params) { const bool add_bos = llama_add_bos_token(llama_get_model(ctx)); - std::vector tokens = ::llama_tokenize(ctx, params.prompt, add_bos); + std::vector tokens = ::common_tokenize(ctx, params.prompt, add_bos); if (llama_decode(ctx, llama_batch_get_one(tokens.data(), tokens.size(), 0, 0))) { LOG_ERR("%s : failed to eval\n", __func__); @@ -160,7 +160,7 @@ int main(int argc, char ** argv) { params.warmup = false; // init - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; diff --git a/examples/export-lora/export-lora.cpp b/examples/export-lora/export-lora.cpp index 644d46a62..f97a3c67b 100644 --- a/examples/export-lora/export-lora.cpp +++ b/examples/export-lora/export-lora.cpp @@ -128,7 +128,7 @@ struct lora_merge_ctx { lora_merge_ctx( std::string & base_fname, - std::vector & lora_files, + std::vector & lora_files, std::string & outfile, int n_threads) : base_model(base_fname, 0), n_threads(n_threads), fout(outfile, std::ios::binary) { fout.exceptions(std::ofstream::failbit); // fail fast on write errors diff --git a/examples/gritlm/gritlm.cpp b/examples/gritlm/gritlm.cpp index 20b99a4fd..4304019c0 100644 --- a/examples/gritlm/gritlm.cpp +++ b/examples/gritlm/gritlm.cpp @@ -15,11 +15,11 @@ static std::vector> encode(llama_context * ctx, const std::ve llama_batch batch = llama_batch_init(llama_n_batch(ctx), 0, 1); for (uint64_t i = 0; i < sentences.size(); i++) { - llama_batch_clear(batch); + common_batch_clear(batch); const std::string input_string = instruction + sentences[i]; - std::vector inputs = llama_tokenize(model, input_string, true, false); + std::vector inputs = common_tokenize(model, input_string, true, false); const int32_t n_toks = inputs.size(); @@ -28,7 +28,7 @@ static std::vector> encode(llama_context * ctx, const std::ve // inputs.push_back(llama_token_eos(model)); // we want to ignore instruction tokens for mean pooling - const int32_t n_inst = llama_tokenize(model, instruction, true, false).size(); + const int32_t n_inst = common_tokenize(model, instruction, true, false).size(); #ifdef GRIT_DEBUG // debug tokens - should be matching as referenced in the GritLM sample @@ -40,7 +40,7 @@ static std::vector> encode(llama_context * ctx, const std::ve // add input to batch (this increments n_tokens) for (int32_t j = 0; j < n_toks; j++) { - llama_batch_add(batch, inputs[j], j, { 0 }, j >= n_inst); + common_batch_add(batch, inputs[j], j, { 0 }, j >= n_inst); } // clear previous kv_cache values (irrelevant for embeddings) @@ -75,7 +75,7 @@ static std::vector> encode(llama_context * ctx, const std::ve } std::vector emb_norm(emb_unorm.size()); - llama_embd_normalize(emb_unorm.data(), emb_norm.data(), n_embd); + common_embd_normalize(emb_unorm.data(), emb_norm.data(), n_embd); result.push_back(emb_norm); #ifdef GRIT_DEBUG @@ -105,16 +105,16 @@ static std::string generate(llama_context * ctx, llama_sampler * smpl, const std llama_batch bat = llama_batch_init(llama_n_batch(ctx), 0, 1); - std::vector inputs = llama_tokenize(model, prompt, false, true); + std::vector inputs = common_tokenize(model, prompt, false, true); int32_t i_current_token = 0; while (true) { - llama_batch_clear(bat); + common_batch_clear(bat); { const int32_t n_inputs = inputs.size(); for (int32_t i = 0; i < n_inputs; i++) { - llama_batch_add(bat, inputs[i], i_current_token++, { 0 }, i == n_inputs - 1); + common_batch_add(bat, inputs[i], i_current_token++, { 0 }, i == n_inputs - 1); } } inputs.clear(); @@ -127,7 +127,7 @@ static std::string generate(llama_context * ctx, llama_sampler * smpl, const std break; } - std::string piece = llama_token_to_piece(ctx, token); + std::string piece = common_token_to_piece(ctx, token); if (stream) { std::printf("%s", piece.c_str()); std::fflush(stdout); @@ -160,8 +160,8 @@ int main(int argc, char * argv[]) { gpt_init(); - llama_model_params mparams = llama_model_params_from_gpt_params(params); - llama_context_params cparams = llama_context_params_from_gpt_params(params); + llama_model_params mparams = common_model_params_from_gpt_params(params); + llama_context_params cparams = common_context_params_from_gpt_params(params); llama_backend_init(); @@ -199,10 +199,10 @@ int main(int argc, char * argv[]) { const int n_embd = llama_n_embd(model); - const float cosine_sim_q0_d0 = llama_embd_similarity_cos(q_rep[0].data(), d_rep[0].data(), n_embd); - const float cosine_sim_q0_d1 = llama_embd_similarity_cos(q_rep[0].data(), d_rep[1].data(), n_embd); - const float cosine_sim_q1_d0 = llama_embd_similarity_cos(q_rep[1].data(), d_rep[0].data(), n_embd); - const float cosine_sim_q1_d1 = llama_embd_similarity_cos(q_rep[1].data(), d_rep[1].data(), n_embd); + const float cosine_sim_q0_d0 = common_embd_similarity_cos(q_rep[0].data(), d_rep[0].data(), n_embd); + const float cosine_sim_q0_d1 = common_embd_similarity_cos(q_rep[0].data(), d_rep[1].data(), n_embd); + const float cosine_sim_q1_d0 = common_embd_similarity_cos(q_rep[1].data(), d_rep[0].data(), n_embd); + const float cosine_sim_q1_d1 = common_embd_similarity_cos(q_rep[1].data(), d_rep[1].data(), n_embd); std::printf("Cosine similarity between \"%.50s\" and \"%.50s\" is: %.3f\n", queries[0].c_str(), documents[0].c_str(), cosine_sim_q0_d0); std::printf("Cosine similarity between \"%.50s\" and \"%.50s\" is: %.3f\n", queries[0].c_str(), documents[1].c_str(), cosine_sim_q0_d1); diff --git a/examples/imatrix/imatrix.cpp b/examples/imatrix/imatrix.cpp index c8e273529..3a6e93344 100644 --- a/examples/imatrix/imatrix.cpp +++ b/examples/imatrix/imatrix.cpp @@ -436,7 +436,7 @@ static bool compute_imatrix(llama_context * ctx, const gpt_params & params) { auto tim1 = std::chrono::high_resolution_clock::now(); LOG_INF("%s: tokenizing the input ..\n", __func__); - std::vector tokens = ::llama_tokenize(ctx, params.prompt, true); + std::vector tokens = ::common_tokenize(ctx, params.prompt, true); auto tim2 = std::chrono::high_resolution_clock::now(); LOG_INF("%s: tokenization took %g ms\n",__func__,1e-3*std::chrono::duration_cast(tim2-tim1).count()); @@ -607,7 +607,7 @@ int main(int argc, char ** argv) { params.warmup = false; // init - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; diff --git a/examples/infill/infill.cpp b/examples/infill/infill.cpp index d52425ae6..0f9e10893 100644 --- a/examples/infill/infill.cpp +++ b/examples/infill/infill.cpp @@ -174,7 +174,7 @@ int main(int argc, char ** argv) { // load the model and apply lora adapter, if any LOG_INF("%s: load the model and apply lora adapter, if any\n", __func__); - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); model = llama_init.model; ctx = llama_init.context; @@ -202,8 +202,8 @@ int main(int argc, char ** argv) { std::vector embd_inp; std::vector embd_end; - std::vector inp_pfx = ::llama_tokenize(ctx, params.input_prefix, false); - std::vector inp_sfx = ::llama_tokenize(ctx, params.input_suffix, false); + std::vector inp_pfx = ::common_tokenize(ctx, params.input_prefix, false); + std::vector inp_sfx = ::common_tokenize(ctx, params.input_suffix, false); GGML_ASSERT(llama_token_prefix(model) >= 0); GGML_ASSERT(llama_token_suffix(model) >= 0); @@ -257,13 +257,13 @@ int main(int argc, char ** argv) { LOG_INF("%s: prompt: '%s'\n", __func__, params.prompt.c_str()); LOG_INF("%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size()); for (int i = 0; i < (int) embd_inp.size(); i++) { - LOG_INF("%6d -> '%s'\n", embd_inp[i], llama_token_to_piece(ctx, embd_inp[i]).c_str()); + LOG_INF("%6d -> '%s'\n", embd_inp[i], common_token_to_piece(ctx, embd_inp[i]).c_str()); } if (params.n_keep > 0) { LOG_INF("%s: static prompt based on n_keep: '", __func__); for (int i = 0; i < params.n_keep; i++) { - LOG_CNT("%s", llama_token_to_piece(ctx, embd_inp[i]).c_str()); + LOG_CNT("%s", common_token_to_piece(ctx, embd_inp[i]).c_str()); } LOG_CNT("'\n"); } @@ -446,7 +446,7 @@ int main(int argc, char ** argv) { // display text if (input_echo) { for (auto id : embd) { - const std::string token_str = llama_token_to_piece(ctx, id); + const std::string token_str = common_token_to_piece(ctx, id); LOG("%s", token_str.c_str()); if (embd.size() > 1) { @@ -468,7 +468,7 @@ int main(int argc, char ** argv) { if ((gpt_sampler_last(smpl) == llama_token_eot(model) || is_interacting) && params.interactive){ if (is_interacting && !params.interactive_first) { // print an eot token - LOG("%s", llama_token_to_piece(ctx, llama_token_eot(model)).c_str()); + LOG("%s", common_token_to_piece(ctx, llama_token_eot(model)).c_str()); } LOG("\n"); console::set_display(console::user_input); @@ -505,8 +505,8 @@ int main(int argc, char ** argv) { } // tokenize new prefix and suffix - std::vector inp_pfx = ::llama_tokenize(ctx, params.input_prefix, false); - std::vector inp_sfx = ::llama_tokenize(ctx, params.input_suffix, false); + std::vector inp_pfx = ::common_tokenize(ctx, params.input_prefix, false); + std::vector inp_sfx = ::common_tokenize(ctx, params.input_suffix, false); inp_pfx.insert(inp_pfx.begin(), llama_token_prefix(model)); inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model)); @@ -579,7 +579,7 @@ int main(int argc, char ** argv) { const size_t original_size = embd_inp.size(); - const auto line_inp = ::llama_tokenize(ctx, buffer, false); + const auto line_inp = ::common_tokenize(ctx, buffer, false); LOG_DBG("input tokens: %s\n", string_from(ctx, line_inp).c_str()); embd_inp.insert(embd_inp.end(), line_inp.begin(), line_inp.end()); @@ -587,7 +587,7 @@ int main(int argc, char ** argv) { for (size_t i = original_size; i < embd_inp.size(); ++i) { const llama_token token = embd_inp[i]; output_tokens.push_back(token); - output_ss << llama_token_to_piece(ctx, token); + output_ss << common_token_to_piece(ctx, token); } n_remain -= line_inp.size(); @@ -620,7 +620,7 @@ int main(int argc, char ** argv) { } } if (!params.interactive && n_remain <= 0) { - LOG("%s", llama_token_to_piece(ctx, llama_token_eot(model)).c_str()); + LOG("%s", common_token_to_piece(ctx, llama_token_eot(model)).c_str()); } LOG("\n"); diff --git a/examples/llava/llava-cli.cpp b/examples/llava/llava-cli.cpp index 8f437863f..13d6daa93 100644 --- a/examples/llava/llava-cli.cpp +++ b/examples/llava/llava-cli.cpp @@ -37,7 +37,7 @@ static bool eval_id(struct llama_context * ctx_llama, int id, int * n_past) { static bool eval_string(struct llama_context * ctx_llama, const char* str, int n_batch, int * n_past, bool add_bos){ std::string str2 = str; - std::vector embd_inp = ::llama_tokenize(ctx_llama, str2, add_bos, true); + std::vector embd_inp = ::common_tokenize(ctx_llama, str2, add_bos, true); eval_tokens(ctx_llama, embd_inp, n_batch, n_past); return true; } @@ -51,7 +51,7 @@ static const char * sample(struct gpt_sampler * smpl, if (llama_token_is_eog(llama_get_model(ctx_llama), id)) { ret = ""; } else { - ret = llama_token_to_piece(ctx_llama, id); + ret = common_token_to_piece(ctx_llama, id); } eval_id(ctx_llama, id, n_past); return ret.c_str(); @@ -159,16 +159,16 @@ static void process_prompt(struct llava_context * ctx_llava, struct llava_image_ user_prompt = prompt.substr(image_pos + std::string("").length()); LOG_INF("system_prompt: %s\n", system_prompt.c_str()); if (params->verbose_prompt) { - auto tmp = ::llama_tokenize(ctx_llava->ctx_llama, system_prompt, true, true); + auto tmp = ::common_tokenize(ctx_llava->ctx_llama, system_prompt, true, true); for (int i = 0; i < (int) tmp.size(); i++) { - LOG_INF("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx_llava->ctx_llama, tmp[i]).c_str()); + LOG_INF("%6d -> '%s'\n", tmp[i], common_token_to_piece(ctx_llava->ctx_llama, tmp[i]).c_str()); } } LOG_INF("user_prompt: %s\n", user_prompt.c_str()); if (params->verbose_prompt) { - auto tmp = ::llama_tokenize(ctx_llava->ctx_llama, user_prompt, true, true); + auto tmp = ::common_tokenize(ctx_llava->ctx_llama, user_prompt, true, true); for (int i = 0; i < (int) tmp.size(); i++) { - LOG_INF("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx_llava->ctx_llama, tmp[i]).c_str()); + LOG_INF("%6d -> '%s'\n", tmp[i], common_token_to_piece(ctx_llava->ctx_llama, tmp[i]).c_str()); } } } else { @@ -176,9 +176,9 @@ static void process_prompt(struct llava_context * ctx_llava, struct llava_image_ system_prompt = "A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.\nUSER:"; user_prompt = prompt + "\nASSISTANT:"; if (params->verbose_prompt) { - auto tmp = ::llama_tokenize(ctx_llava->ctx_llama, user_prompt, true, true); + auto tmp = ::common_tokenize(ctx_llava->ctx_llama, user_prompt, true, true); for (int i = 0; i < (int) tmp.size(); i++) { - LOG_INF("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx_llava->ctx_llama, tmp[i]).c_str()); + LOG_INF("%6d -> '%s'\n", tmp[i], common_token_to_piece(ctx_llava->ctx_llama, tmp[i]).c_str()); } } } @@ -219,7 +219,7 @@ static struct llama_model * llava_init(gpt_params * params) { llama_backend_init(); llama_numa_init(params->numa); - llama_model_params model_params = llama_model_params_from_gpt_params(*params); + llama_model_params model_params = common_model_params_from_gpt_params(*params); llama_model * model = llama_load_model_from_file(params->model.c_str(), model_params); if (model == NULL) { @@ -240,7 +240,7 @@ static struct llava_context * llava_init_context(gpt_params * params, llama_mode auto ctx_clip = clip_model_load(clip_path, /*verbosity=*/ 1); - llama_context_params ctx_params = llama_context_params_from_gpt_params(*params); + llama_context_params ctx_params = common_context_params_from_gpt_params(*params); ctx_params.n_ctx = params->n_ctx < 2048 ? 2048 : params->n_ctx; // we need a longer context size to process image embeddings llama_context * ctx_llama = llama_new_context_with_model(model, ctx_params); diff --git a/examples/llava/minicpmv-cli.cpp b/examples/llava/minicpmv-cli.cpp index c5156c35b..4408f3d1b 100644 --- a/examples/llava/minicpmv-cli.cpp +++ b/examples/llava/minicpmv-cli.cpp @@ -29,7 +29,7 @@ static struct llama_model * llava_init(gpt_params * params) { llama_backend_init(); llama_numa_init(params->numa); - llama_model_params model_params = llama_model_params_from_gpt_params(*params); + llama_model_params model_params = common_model_params_from_gpt_params(*params); llama_model * model = llama_load_model_from_file(params->model.c_str(), model_params); if (model == NULL) { @@ -45,7 +45,7 @@ static struct llava_context * llava_init_context(gpt_params * params, llama_mode prompt = "describe the image in detail."; } - llama_context_params ctx_params = llama_context_params_from_gpt_params(*params); + llama_context_params ctx_params = common_context_params_from_gpt_params(*params); if (params->n_ctx < 2048) { // warn user here, "Image processing requires at least 2048 context, setting context to 2048" LOG_WRN("%s: Image processing requires at least 2048 context, setting context to 2048\n" , __func__); @@ -114,7 +114,7 @@ static bool eval_id(struct llama_context * ctx_llama, int id, int * n_past) { static bool eval_string(struct llama_context * ctx_llama, const char* str, int n_batch, int * n_past, bool add_bos){ std::string str2 = str; - std::vector embd_inp = ::llama_tokenize(ctx_llama, str2, add_bos, true); + std::vector embd_inp = ::common_tokenize(ctx_llama, str2, add_bos, true); return eval_tokens(ctx_llama, embd_inp, n_batch, n_past); } @@ -171,7 +171,7 @@ static const char * sample(struct gpt_sampler * smpl, if (llama_token_is_eog(llama_get_model(ctx_llama), id)) { ret = ""; } else { - ret = llama_token_to_piece(ctx_llama, id); + ret = common_token_to_piece(ctx_llama, id); } eval_id(ctx_llama, id, n_past); return ret.c_str(); diff --git a/examples/lookahead/lookahead.cpp b/examples/lookahead/lookahead.cpp index 49870b4a4..f35628c47 100644 --- a/examples/lookahead/lookahead.cpp +++ b/examples/lookahead/lookahead.cpp @@ -56,7 +56,7 @@ int main(int argc, char ** argv) { llama_numa_init(params.numa); // load the target model - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; @@ -65,7 +65,7 @@ int main(int argc, char ** argv) { std::vector inp; std::vector all; - inp = ::llama_tokenize(ctx, params.prompt, true, true); + inp = ::common_tokenize(ctx, params.prompt, true, true); all = inp; const int max_context_size = llama_n_ctx(ctx); @@ -79,7 +79,7 @@ int main(int argc, char ** argv) { LOG("\n\n"); for (auto id : inp) { - LOG("%s", llama_token_to_piece(ctx, id).c_str()); + LOG("%s", common_token_to_piece(ctx, id).c_str()); } fflush(stderr); @@ -161,7 +161,7 @@ int main(int argc, char ** argv) { gpt_sampler_accept(smpl, id, true); { - const std::string token_str = llama_token_to_piece(ctx, id); + const std::string token_str = common_token_to_piece(ctx, id); LOG("%s", token_str.c_str()); fflush(stdout); @@ -172,7 +172,7 @@ int main(int argc, char ** argv) { // debug if (dump_kv_cache) { llama_kv_cache_view_update(ctx, &kvc_view); - llama_kv_cache_dump_view_seqs(kvc_view, 40); + common_kv_cache_dump_view_seqs(kvc_view, 40); } // build the mask from https://lmsys.org/blog/2023-11-21-lookahead-decoding/ @@ -201,10 +201,10 @@ int main(int argc, char ** argv) { // V V V V V V // id { - llama_batch_clear(batch); + common_batch_clear(batch); // current token - first token of the first level - llama_batch_add(batch, id, n_past, seq_id_all, true); + common_batch_add(batch, id, n_past, seq_id_all, true); // verification n-grams - queue this before the lookahead tokens for less KV cache fragmentation { @@ -229,7 +229,7 @@ int main(int argc, char ** argv) { ngrams_cur[g].tokens [j + 1] = t; ngrams_cur[g].i_batch[j + 1] = batch.n_tokens; - llama_batch_add(batch, t, n_past + j + 1, { W + 1 + g }, true); + common_batch_add(batch, t, n_past + j + 1, { W + 1 + g }, true); } } } @@ -241,13 +241,13 @@ int main(int argc, char ** argv) { seq_id_look[j] = i + j + 1; } - llama_batch_add(batch, tokens_j[0][i], n_past + i, seq_id_look, false); + common_batch_add(batch, tokens_j[0][i], n_past + i, seq_id_look, false); } // fill the rest of the levels for (int j = 1; j < N - 1; j++) { for (int i = 0; i < W; i++) { - llama_batch_add(batch, tokens_j[j][i], n_past + j + i, { i + 1 }, j == N - 2); + common_batch_add(batch, tokens_j[j][i], n_past + j + i, { i + 1 }, j == N - 2); } } } @@ -287,7 +287,7 @@ int main(int argc, char ** argv) { // print { - const std::string token_str = llama_token_to_piece(ctx, id); + const std::string token_str = common_token_to_piece(ctx, id); if (v == 0) { LOG("%s", token_str.c_str()); @@ -327,7 +327,7 @@ int main(int argc, char ** argv) { // print known n-grams starting with token id (debug) if (0 && v == 0) { if (ngrams_observed.cnt[id] > 0) { - LOG("\n - %d n-grams starting with '%s'\n", ngrams_observed.cnt[id], llama_token_to_piece(ctx, id).c_str()); + LOG("\n - %d n-grams starting with '%s'\n", ngrams_observed.cnt[id], common_token_to_piece(ctx, id).c_str()); } for (int i = 0; i < ngrams_observed.cnt[id]; i++) { @@ -336,7 +336,7 @@ int main(int argc, char ** argv) { const int idx = id*(N - 1)*G + i*(N - 1); for (int j = 0; j < N - 1; j++) { - const std::string token_str = llama_token_to_piece(ctx, ngrams_observed.tokens[idx + j]); + const std::string token_str = common_token_to_piece(ctx, ngrams_observed.tokens[idx + j]); LOG("%s", token_str.c_str()); } diff --git a/examples/lookup/lookup-create.cpp b/examples/lookup/lookup-create.cpp index 33287c02c..2232be369 100644 --- a/examples/lookup/lookup-create.cpp +++ b/examples/lookup/lookup-create.cpp @@ -23,7 +23,7 @@ int main(int argc, char ** argv){ llama_numa_init(params.numa); // load the model - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; @@ -31,7 +31,7 @@ int main(int argc, char ** argv){ // tokenize the prompt std::vector inp; - inp = ::llama_tokenize(ctx, params.prompt, true, true); + inp = ::common_tokenize(ctx, params.prompt, true, true); fprintf(stderr, "%s: tokenization done\n", __func__); diff --git a/examples/lookup/lookup-stats.cpp b/examples/lookup/lookup-stats.cpp index 6d1e1ceb9..27755b944 100644 --- a/examples/lookup/lookup-stats.cpp +++ b/examples/lookup/lookup-stats.cpp @@ -28,14 +28,14 @@ int main(int argc, char ** argv){ llama_numa_init(params.numa); // load the model - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; // tokenize the prompt std::vector inp; - inp = ::llama_tokenize(ctx, params.prompt, true, true); + inp = ::common_tokenize(ctx, params.prompt, true, true); llama_ngram_cache ngram_cache_context; llama_ngram_cache ngram_cache_dynamic; diff --git a/examples/lookup/lookup.cpp b/examples/lookup/lookup.cpp index 2ccd0e6c1..5f4ff7853 100644 --- a/examples/lookup/lookup.cpp +++ b/examples/lookup/lookup.cpp @@ -31,14 +31,14 @@ int main(int argc, char ** argv){ llama_numa_init(params.numa); // load the model - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; // tokenize the prompt std::vector inp; - inp = ::llama_tokenize(ctx, params.prompt, true, true); + inp = ::common_tokenize(ctx, params.prompt, true, true); llama_ngram_cache ngram_cache_context; llama_ngram_cache ngram_cache_dynamic; @@ -80,7 +80,7 @@ int main(int argc, char ** argv){ LOG("\n\n"); for (auto id : inp) { - LOG("%s", llama_token_to_piece(ctx, id).c_str()); + LOG("%s", common_token_to_piece(ctx, id).c_str()); } fflush(stderr); @@ -117,7 +117,7 @@ int main(int argc, char ** argv){ // debug if (dump_kv_cache) { llama_kv_cache_view_update(ctx, &kvc_view); - llama_kv_cache_dump_view_seqs(kvc_view, 40); + common_kv_cache_dump_view_seqs(kvc_view, 40); } // print current draft sequence @@ -130,7 +130,7 @@ int main(int argc, char ** argv){ gpt_sampler_accept(smpl, id, true); - const std::string token_str = llama_token_to_piece(ctx, id); + const std::string token_str = common_token_to_piece(ctx, id); if (!params.use_color) { LOG("%s", token_str.c_str()); @@ -192,8 +192,8 @@ int main(int argc, char ** argv){ // clean the cache of draft tokens that weren't accepted llama_kv_cache_seq_rm(ctx, 0, n_past, -1); - llama_batch_clear(batch_tgt); - llama_batch_add(batch_tgt, draft[0], n_past, { 0 }, true); + common_batch_clear(batch_tgt); + common_batch_add(batch_tgt, draft[0], n_past, { 0 }, true); // Draft already contains a single token sampled from the model: GGML_ASSERT(draft.size() == 1); @@ -203,7 +203,7 @@ int main(int argc, char ** argv){ llama_ngram_cache_draft(inp, draft, n_draft, LLAMA_NGRAM_MIN, LLAMA_NGRAM_MAX, ngram_cache_context, ngram_cache_dynamic, ngram_cache_static); for (size_t i = 1; i < draft.size(); ++i) { - llama_batch_add(batch_tgt, draft[i], n_past + i, { 0 }, true); + common_batch_add(batch_tgt, draft[i], n_past + i, { 0 }, true); } t_draft_us += ggml_time_us() - t_start_draft_us; diff --git a/examples/main/main.cpp b/examples/main/main.cpp index 6bbb1e13e..cfd340f6c 100644 --- a/examples/main/main.cpp +++ b/examples/main/main.cpp @@ -127,9 +127,9 @@ static void sigint_handler(int signo) { } #endif -static std::string chat_add_and_format(struct llama_model * model, std::vector & chat_msgs, const std::string & role, const std::string & content) { - llama_chat_msg new_msg{role, content}; - auto formatted = llama_chat_format_single(model, g_params->chat_template, chat_msgs, new_msg, role == "user"); +static std::string chat_add_and_format(struct llama_model * model, std::vector & chat_msgs, const std::string & role, const std::string & content) { + common_chat_msg new_msg{role, content}; + auto formatted = common_chat_format_single(model, g_params->chat_template, chat_msgs, new_msg, role == "user"); chat_msgs.push_back({role, content}); LOG_DBG("formatted: '%s'\n", formatted.c_str()); return formatted; @@ -189,7 +189,7 @@ int main(int argc, char ** argv) { llama_context * ctx = nullptr; gpt_sampler * smpl = nullptr; - std::vector chat_msgs; + std::vector chat_msgs; g_model = &model; g_ctx = &ctx; @@ -197,7 +197,7 @@ int main(int argc, char ** argv) { // load the model and apply lora adapter, if any LOG_INF("%s: load the model and apply lora adapter, if any\n", __func__); - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); model = llama_init.model; ctx = llama_init.context; @@ -246,7 +246,7 @@ int main(int argc, char ** argv) { // print chat template example in conversation mode if (params.conversation) { if (params.enable_chat_template) { - LOG_INF("%s: chat template example:\n%s\n", __func__, llama_chat_format_example(model, params.chat_template).c_str()); + LOG_INF("%s: chat template example:\n%s\n", __func__, common_chat_format_example(model, params.chat_template).c_str()); } else { LOG_INF("%s: in-suffix/prefix is specified, chat template will be disabled\n", __func__); } @@ -296,7 +296,7 @@ int main(int argc, char ** argv) { : params.prompt; if (params.interactive_first || !params.prompt.empty() || session_tokens.empty()) { LOG_DBG("tokenize the prompt\n"); - embd_inp = ::llama_tokenize(ctx, prompt, true, true); + embd_inp = ::common_tokenize(ctx, prompt, true, true); } else { LOG_DBG("use session tokens\n"); embd_inp = session_tokens; @@ -379,13 +379,13 @@ int main(int argc, char ** argv) { LOG_INF("%s: prompt: '%s'\n", __func__, params.prompt.c_str()); LOG_INF("%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size()); for (int i = 0; i < (int) embd_inp.size(); i++) { - LOG_INF("%6d -> '%s'\n", embd_inp[i], llama_token_to_piece(ctx, embd_inp[i]).c_str()); + LOG_INF("%6d -> '%s'\n", embd_inp[i], common_token_to_piece(ctx, embd_inp[i]).c_str()); } if (params.n_keep > add_bos) { LOG_INF("%s: static prompt based on n_keep: '", __func__); for (int i = 0; i < params.n_keep; i++) { - LOG_CNT("%s", llama_token_to_piece(ctx, embd_inp[i]).c_str()); + LOG_CNT("%s", common_token_to_piece(ctx, embd_inp[i]).c_str()); } LOG_CNT("'\n"); } @@ -415,9 +415,9 @@ int main(int argc, char ** argv) { for (const auto & antiprompt : params.antiprompt) { LOG_INF("Reverse prompt: '%s'\n", antiprompt.c_str()); if (params.verbose_prompt) { - auto tmp = ::llama_tokenize(ctx, antiprompt, false, true); + auto tmp = ::common_tokenize(ctx, antiprompt, false, true); for (int i = 0; i < (int) tmp.size(); i++) { - LOG_INF("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx, tmp[i]).c_str()); + LOG_INF("%6d -> '%s'\n", tmp[i], common_token_to_piece(ctx, tmp[i]).c_str()); } } } @@ -430,9 +430,9 @@ int main(int argc, char ** argv) { if (!params.input_prefix.empty()) { LOG_INF("Input prefix: '%s'\n", params.input_prefix.c_str()); if (params.verbose_prompt) { - auto tmp = ::llama_tokenize(ctx, params.input_prefix, true, true); + auto tmp = ::common_tokenize(ctx, params.input_prefix, true, true); for (int i = 0; i < (int) tmp.size(); i++) { - LOG_INF("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx, tmp[i]).c_str()); + LOG_INF("%6d -> '%s'\n", tmp[i], common_token_to_piece(ctx, tmp[i]).c_str()); } } } @@ -440,9 +440,9 @@ int main(int argc, char ** argv) { if (!params.input_suffix.empty()) { LOG_INF("Input suffix: '%s'\n", params.input_suffix.c_str()); if (params.verbose_prompt) { - auto tmp = ::llama_tokenize(ctx, params.input_suffix, false, true); + auto tmp = ::common_tokenize(ctx, params.input_suffix, false, true); for (int i = 0; i < (int) tmp.size(); i++) { - LOG_INF("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx, tmp[i]).c_str()); + LOG_INF("%6d -> '%s'\n", tmp[i], common_token_to_piece(ctx, tmp[i]).c_str()); } } } @@ -521,7 +521,7 @@ int main(int argc, char ** argv) { antiprompt_ids.reserve(params.antiprompt.size()); for (const std::string & antiprompt : params.antiprompt) { - antiprompt_ids.emplace_back(::llama_tokenize(ctx, antiprompt, false, true)); + antiprompt_ids.emplace_back(::common_tokenize(ctx, antiprompt, false, true)); } if (llama_model_has_encoder(model)) { @@ -714,7 +714,7 @@ int main(int argc, char ** argv) { // display text if (input_echo && display) { for (auto id : embd) { - const std::string token_str = llama_token_to_piece(ctx, id, params.special); + const std::string token_str = common_token_to_piece(ctx, id, params.special); // Console/Stream Output LOG("%s", token_str.c_str()); @@ -788,7 +788,7 @@ int main(int argc, char ** argv) { if (params.interactive) { if (!params.antiprompt.empty()) { // tokenize and inject first reverse prompt - const auto first_antiprompt = ::llama_tokenize(ctx, params.antiprompt.front(), false, true); + const auto first_antiprompt = ::common_tokenize(ctx, params.antiprompt.front(), false, true); embd_inp.insert(embd_inp.end(), first_antiprompt.begin(), first_antiprompt.end()); is_antiprompt = true; } @@ -804,7 +804,7 @@ int main(int argc, char ** argv) { // if current token is not EOG, we add it to current assistant message if (params.conversation) { const auto id = gpt_sampler_last(smpl); - assistant_ss << llama_token_to_piece(ctx, id, false); + assistant_ss << common_token_to_piece(ctx, id, false); } if (n_past > 0 && is_interacting) { @@ -862,9 +862,9 @@ int main(int argc, char ** argv) { ? chat_add_and_format(model, chat_msgs, "user", std::move(buffer)) : std::move(buffer); // TODO: one inconvenient of current chat template implementation is that we can't distinguish between user input and special tokens (prefix/postfix) - const auto line_pfx = ::llama_tokenize(ctx, params.input_prefix, false, true); - const auto line_inp = ::llama_tokenize(ctx, user_inp, false, format_chat); - const auto line_sfx = ::llama_tokenize(ctx, params.input_suffix, false, true); + const auto line_pfx = ::common_tokenize(ctx, params.input_prefix, false, true); + const auto line_inp = ::common_tokenize(ctx, user_inp, false, format_chat); + const auto line_sfx = ::common_tokenize(ctx, params.input_suffix, false, true); LOG_DBG("input tokens: %s\n", string_from(ctx, line_inp).c_str()); @@ -882,7 +882,7 @@ int main(int argc, char ** argv) { for (size_t i = original_size; i < embd_inp.size(); ++i) { const llama_token token = embd_inp[i]; output_tokens.push_back(token); - output_ss << llama_token_to_piece(ctx, token); + output_ss << common_token_to_piece(ctx, token); } // reset assistant message diff --git a/examples/parallel/parallel.cpp b/examples/parallel/parallel.cpp index 81e2f7ed7..0e720807f 100644 --- a/examples/parallel/parallel.cpp +++ b/examples/parallel/parallel.cpp @@ -130,7 +130,7 @@ int main(int argc, char ** argv) { llama_numa_init(params.numa); // load the target model - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; @@ -164,7 +164,7 @@ int main(int argc, char ** argv) { } std::vector tokens_system; - tokens_system = ::llama_tokenize(ctx, k_system, true); + tokens_system = ::common_tokenize(ctx, k_system, true); const int32_t n_tokens_system = tokens_system.size(); llama_seq_id g_seq_id = 0; @@ -189,7 +189,7 @@ int main(int argc, char ** argv) { LOG_INF("%s: Evaluating the system prompt ...\n", __func__); for (int32_t i = 0; i < n_tokens_system; ++i) { - llama_batch_add(batch, tokens_system[i], i, { 0 }, false); + common_batch_add(batch, tokens_system[i], i, { 0 }, false); } if (llama_decode(ctx, batch) != 0) { @@ -210,10 +210,10 @@ int main(int argc, char ** argv) { while (true) { if (dump_kv_cache) { llama_kv_cache_view_update(ctx, &kvc_view); - llama_kv_cache_dump_view_seqs(kvc_view, 40); + common_kv_cache_dump_view_seqs(kvc_view, 40); } - llama_batch_clear(batch); + common_batch_clear(batch); // decode any currently ongoing sequences for (auto & client : clients) { @@ -223,7 +223,7 @@ int main(int argc, char ** argv) { client.i_batch = batch.n_tokens; - llama_batch_add(batch, client.sampled, n_tokens_system + client.n_prompt + client.n_decoded, { client.id + 1 }, true); + common_batch_add(batch, client.sampled, n_tokens_system + client.n_prompt + client.n_decoded, { client.id + 1 }, true); client.n_decoded += 1; } @@ -256,10 +256,10 @@ int main(int argc, char ** argv) { // do not prepend BOS because we have a system prompt! std::vector tokens_prompt; - tokens_prompt = ::llama_tokenize(ctx, client.prompt, false); + tokens_prompt = ::common_tokenize(ctx, client.prompt, false); for (size_t i = 0; i < tokens_prompt.size(); ++i) { - llama_batch_add(batch, tokens_prompt[i], i + n_tokens_system, { client.id + 1 }, false); + common_batch_add(batch, tokens_prompt[i], i + n_tokens_system, { client.id + 1 }, false); } // extract the logits only for the last token @@ -350,7 +350,7 @@ int main(int argc, char ** argv) { client.t_start_gen = ggml_time_us(); } - const std::string token_str = llama_token_to_piece(ctx, id); + const std::string token_str = common_token_to_piece(ctx, id); client.response += token_str; client.sampled = id; diff --git a/examples/passkey/passkey.cpp b/examples/passkey/passkey.cpp index 7ef8d14f3..028094eb9 100644 --- a/examples/passkey/passkey.cpp +++ b/examples/passkey/passkey.cpp @@ -61,7 +61,7 @@ int main(int argc, char ** argv) { // initialize the model - llama_model_params model_params = llama_model_params_from_gpt_params(params); + llama_model_params model_params = common_model_params_from_gpt_params(params); llama_model * model = llama_load_model_from_file(params.model.c_str(), model_params); @@ -72,7 +72,7 @@ int main(int argc, char ** argv) { // initialize the context - llama_context_params ctx_params = llama_context_params_from_gpt_params(params); + llama_context_params ctx_params = common_context_params_from_gpt_params(params); ctx_params.n_ctx = llama_n_ctx_train(model)*n_grp + n_keep; @@ -92,10 +92,10 @@ int main(int argc, char ** argv) { // tokenize the prompt std::vector tokens_list; - tokens_list = ::llama_tokenize(ctx, params.prompt, true); + tokens_list = ::common_tokenize(ctx, params.prompt, true); // tokenize the prefix and use it as a sink - const int n_tokens_prefix = ::llama_tokenize(ctx, prompt_prefix, true).size(); + const int n_tokens_prefix = ::common_tokenize(ctx, prompt_prefix, true).size(); const int n_tokens_all = tokens_list.size(); @@ -137,10 +137,10 @@ int main(int argc, char ** argv) { n_past = llama_kv_cache_seq_pos_max(ctx, 0) + 1; } - llama_batch_clear(batch); + common_batch_clear(batch); for (int j = 0; j < n_batch && i + j < n_tokens_all; j++) { - llama_batch_add(batch, tokens_list[i + j], n_past++, { 0 }, false); + common_batch_add(batch, tokens_list[i + j], n_past++, { 0 }, false); } if (i + n_batch >= n_tokens_all) { @@ -171,10 +171,10 @@ int main(int argc, char ** argv) { n_past = llama_kv_cache_seq_pos_max(ctx, 0) + 1; - llama_batch_clear(batch); + common_batch_clear(batch); for (int j = 0; j < n_batch && i + j < n_tokens_all; j++) { - llama_batch_add(batch, tokens_list[i + j], n_past++, { 0 }, false); + common_batch_add(batch, tokens_list[i + j], n_past++, { 0 }, false); } if (i + n_batch >= n_tokens_all) { @@ -229,15 +229,15 @@ int main(int argc, char ** argv) { break; } - LOG("%s", llama_token_to_piece(ctx, new_token_id).c_str()); + LOG("%s", common_token_to_piece(ctx, new_token_id).c_str()); n_decode += 1; // prepare the next batch - llama_batch_clear(batch); + common_batch_clear(batch); // push this new token for next evaluation - llama_batch_add(batch, new_token_id, n_past++, { 0 }, true); + common_batch_add(batch, new_token_id, n_past++, { 0 }, true); } n_cur += 1; diff --git a/examples/perplexity/perplexity.cpp b/examples/perplexity/perplexity.cpp index 87347135e..d72e533fe 100644 --- a/examples/perplexity/perplexity.cpp +++ b/examples/perplexity/perplexity.cpp @@ -348,7 +348,7 @@ static results_perplexity perplexity_v2(llama_context * ctx, const gpt_params & LOG_INF("%s: tokenizing the input ..\n", __func__); - std::vector tokens = ::llama_tokenize(ctx, params.prompt, true); + std::vector tokens = ::common_tokenize(ctx, params.prompt, true); const int n_ctx = llama_n_ctx(ctx); @@ -500,7 +500,7 @@ static results_perplexity perplexity(llama_context * ctx, const gpt_params & par auto tim1 = std::chrono::high_resolution_clock::now(); LOG_INF("%s: tokenizing the input ..\n", __func__); - std::vector tokens = ::llama_tokenize(ctx, params.prompt, true); + std::vector tokens = ::common_tokenize(ctx, params.prompt, true); auto tim2 = std::chrono::high_resolution_clock::now(); LOG_INF("%s: tokenization took %g ms\n",__func__,1e-3*std::chrono::duration_cast(tim2-tim1).count()); @@ -844,7 +844,7 @@ static void hellaswag_score(llama_context * ctx, const gpt_params & params) { hs_cur.gold_ending_idx = std::stoi( prompt_lines[idx*6+1] ); for (size_t j = 0; j < 4; j++) { hs_cur.ending[j] = prompt_lines[idx*6+2+j]; - hs_cur.seq_tokens[j] = ::llama_tokenize(ctx, hs_cur.context + " " + hs_cur.ending[j], true); + hs_cur.seq_tokens[j] = ::common_tokenize(ctx, hs_cur.context + " " + hs_cur.ending[j], true); } // determine the common prefix of the endings @@ -900,7 +900,7 @@ static void hellaswag_score(llama_context * ctx, const gpt_params & params) { size_t i1 = i0; size_t i_logits = 0; // this tells us how many logits were needed before this point in the batch - llama_batch_clear(batch); + common_batch_clear(batch); // batch as much tasks as possible into the available context // each task has 4 unique sequence ids - one for each ending @@ -916,7 +916,7 @@ static void hellaswag_score(llama_context * ctx, const gpt_params & params) { } for (size_t i = 0; i < hs_cur.common_prefix; ++i) { - llama_batch_add(batch, hs_cur.seq_tokens[0][i], i, { s0 + 0, s0 + 1, s0 + 2, s0 + 3 }, false); + common_batch_add(batch, hs_cur.seq_tokens[0][i], i, { s0 + 0, s0 + 1, s0 + 2, s0 + 3 }, false); } batch.logits[batch.n_tokens - 1] = true; // we need logits for the last token of the common prefix n_logits += 1; @@ -926,7 +926,7 @@ static void hellaswag_score(llama_context * ctx, const gpt_params & params) { // TODO: don't evaluate the last token of each sequence for (size_t i = hs_cur.common_prefix; i < seq_tokens_size; ++i) { const bool needs_logits = i < seq_tokens_size - 1; - llama_batch_add(batch, hs_cur.seq_tokens[s][i], i, { s0 + s }, needs_logits); + common_batch_add(batch, hs_cur.seq_tokens[s][i], i, { s0 + s }, needs_logits); n_logits += needs_logits; } } @@ -1136,8 +1136,8 @@ static void winogrande_score(llama_context * ctx, const gpt_params & params) { LOG_INF("%s : tokenizing selected tasks\n", __func__); for (auto & task : data) { - task.seq_tokens[0] = ::llama_tokenize(ctx, task.first + task.choices[0] + task.second, true); - task.seq_tokens[1] = ::llama_tokenize(ctx, task.first + task.choices[1] + task.second, true); + task.seq_tokens[0] = ::common_tokenize(ctx, task.first + task.choices[0] + task.second, true); + task.seq_tokens[1] = ::common_tokenize(ctx, task.first + task.choices[1] + task.second, true); task.common_prefix = 0; for (size_t k = 0; k < task.seq_tokens[0].size(); k++) { @@ -1152,8 +1152,8 @@ static void winogrande_score(llama_context * ctx, const gpt_params & params) { task.seq_tokens[0].size() - task.common_prefix + task.seq_tokens[1].size() - task.common_prefix; - task.n_base1 = ::llama_tokenize(ctx, task.first + task.choices[0], true).size(); - task.n_base2 = ::llama_tokenize(ctx, task.first + task.choices[1], true).size(); + task.n_base1 = ::common_tokenize(ctx, task.first + task.choices[0], true).size(); + task.n_base2 = ::common_tokenize(ctx, task.first + task.choices[1], true).size(); } LOG_INF("%s : calculating winogrande score over selected tasks.\n", __func__); @@ -1184,7 +1184,7 @@ static void winogrande_score(llama_context * ctx, const gpt_params & params) { size_t i1 = i0; size_t i_logits = 0; - llama_batch_clear(batch); + common_batch_clear(batch); while (n_cur + (int) data[i1].required_tokens <= n_ctx) { int n_logits = 0; @@ -1194,7 +1194,7 @@ static void winogrande_score(llama_context * ctx, const gpt_params & params) { } for (size_t i = 0; i < data[i1].common_prefix; ++i) { - llama_batch_add(batch, data[i1].seq_tokens[0][i], i, { s0 + 0, s0 + 1 }, false); + common_batch_add(batch, data[i1].seq_tokens[0][i], i, { s0 + 0, s0 + 1 }, false); } batch.logits[batch.n_tokens - 1] = true; n_logits += 1; @@ -1202,7 +1202,7 @@ static void winogrande_score(llama_context * ctx, const gpt_params & params) { for (int s = 0; s < 2; ++s) { // TODO: end before the last token, no need to predict past the end of the sequences for (size_t i = data[i1].common_prefix; i < data[i1].seq_tokens[s].size(); ++i) { - llama_batch_add(batch, data[i1].seq_tokens[s][i], i, { s0 + s }, true); + common_batch_add(batch, data[i1].seq_tokens[s][i], i, { s0 + s }, true); n_logits += 1; } } @@ -1359,7 +1359,7 @@ static bool multiple_choice_prepare_one_task(llama_context * ctx, multiple_choic } return false; } - task.seq_tokens.emplace_back(::llama_tokenize(ctx, task.question + " " + answer, true)); + task.seq_tokens.emplace_back(::common_tokenize(ctx, task.question + " " + answer, true)); } auto min_len = task.seq_tokens.front().size(); for (auto& seq : task.seq_tokens) { @@ -1536,7 +1536,7 @@ static void multiple_choice_score(llama_context * ctx, const gpt_params & params size_t i1 = i0; size_t i_logits = 0; // this tells us how many logits were needed before this point in the batch - llama_batch_clear(batch); + common_batch_clear(batch); // batch as much tasks as possible into the available context // each task has 4 unique sequence ids - one for each ending @@ -1559,7 +1559,7 @@ static void multiple_choice_score(llama_context * ctx, const gpt_params & params for (size_t i = 0; i < cur_task.common_prefix; ++i) { //llama_batch_add(batch, cur_task.seq_tokens[0][i], i, { s0 + 0, s0 + 1, s0 + 2, s0 + 3}, false); - llama_batch_add(batch, cur_task.seq_tokens[0][i], i, batch_indeces, false); + common_batch_add(batch, cur_task.seq_tokens[0][i], i, batch_indeces, false); } batch.logits[batch.n_tokens - 1] = true; // we need logits for the last token of the common prefix n_logits += 1; @@ -1569,7 +1569,7 @@ static void multiple_choice_score(llama_context * ctx, const gpt_params & params // TODO: don't evaluate the last token of each sequence for (size_t i = cur_task.common_prefix; i < seq_tokens_size; ++i) { const bool needs_logits = i < seq_tokens_size - 1; - llama_batch_add(batch, cur_task.seq_tokens[s][i], i, { s0 + s }, needs_logits); + common_batch_add(batch, cur_task.seq_tokens[s][i], i, { s0 + s }, needs_logits); n_logits += needs_logits; } } @@ -2004,7 +2004,7 @@ int main(int argc, char ** argv) { llama_numa_init(params.numa); // load the model and apply lora adapter, if any - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; diff --git a/examples/retrieval/retrieval.cpp b/examples/retrieval/retrieval.cpp index 5971690f1..2aeb25ac8 100644 --- a/examples/retrieval/retrieval.cpp +++ b/examples/retrieval/retrieval.cpp @@ -77,7 +77,7 @@ static std::vector chunk_file(const std::string & filename, int chunk_siz static void batch_add_seq(llama_batch & batch, const std::vector & tokens, llama_seq_id seq_id) { size_t n_tokens = tokens.size(); for (size_t i = 0; i < n_tokens; i++) { - llama_batch_add(batch, tokens[i], i, { seq_id }, true); + common_batch_add(batch, tokens[i], i, { seq_id }, true); } } @@ -107,7 +107,7 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu } float * out = output + batch.seq_id[i][0] * n_embd; - llama_embd_normalize(embd, out, n_embd); + common_embd_normalize(embd, out, n_embd); } } @@ -149,7 +149,7 @@ int main(int argc, char ** argv) { llama_numa_init(params.numa); // load the model - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; @@ -185,7 +185,7 @@ int main(int argc, char ** argv) { // tokenize the prompts and trim for (auto & chunk : chunks) { - auto inp = ::llama_tokenize(ctx, chunk.textdata, true, false); + auto inp = ::common_tokenize(ctx, chunk.textdata, true, false); if (inp.size() > n_batch) { LOG_ERR("%s: chunk size (%lld) exceeds batch size (%lld), increase batch size and re-run\n", __func__, (long long int) inp.size(), (long long int) n_batch); @@ -204,7 +204,7 @@ int main(int argc, char ** argv) { LOG_INF("%s: prompt %d: '%s'\n", __func__, i, chunks[i].textdata.c_str()); LOG_INF("%s: number of tokens in prompt = %zu\n", __func__, chunks[i].tokens.size()); for (int j = 0; j < (int) chunks[i].tokens.size(); j++) { - LOG_INF("%6d -> '%s'\n", chunks[i].tokens[j], llama_token_to_piece(ctx, chunks[i].tokens[j]).c_str()); + LOG_INF("%6d -> '%s'\n", chunks[i].tokens[j], common_token_to_piece(ctx, chunks[i].tokens[j]).c_str()); } LOG_INF("\n\n"); } @@ -232,7 +232,7 @@ int main(int argc, char ** argv) { if (batch.n_tokens + n_toks > n_batch) { float * out = emb + p * n_embd; batch_decode(ctx, batch, out, s, n_embd); - llama_batch_clear(batch); + common_batch_clear(batch); p += s; s = 0; } @@ -260,20 +260,20 @@ int main(int argc, char ** argv) { while (true) { LOG("Enter query: "); std::getline(std::cin, query); - std::vector query_tokens = llama_tokenize(ctx, query, true); + std::vector query_tokens = common_tokenize(ctx, query, true); batch_add_seq(query_batch, query_tokens, 0); std::vector query_emb(n_embd, 0); batch_decode(ctx, query_batch, query_emb.data(), 1, n_embd); - llama_batch_clear(query_batch); + common_batch_clear(query_batch); // compute cosine similarities { std::vector> similarities; for (int i = 0; i < n_chunks; i++) { - float sim = llama_embd_similarity_cos(chunks[i].embedding.data(), query_emb.data(), n_embd); + float sim = common_embd_similarity_cos(chunks[i].embedding.data(), query_emb.data(), n_embd); similarities.push_back(std::make_pair(i, sim)); } diff --git a/examples/save-load-state/save-load-state.cpp b/examples/save-load-state/save-load-state.cpp index 0117d9357..2fdb79742 100644 --- a/examples/save-load-state/save-load-state.cpp +++ b/examples/save-load-state/save-load-state.cpp @@ -28,7 +28,7 @@ int main(int argc, char ** argv) { std::string result2; // init - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); llama_model * model = llama_init.model; llama_context * ctx = llama_init.context; @@ -46,7 +46,7 @@ int main(int argc, char ** argv) { llama_sampler_chain_add(smpl, llama_sampler_init_dist(params.sparams.seed)); // tokenize prompt - auto tokens = llama_tokenize(ctx, params.prompt, true); + auto tokens = common_tokenize(ctx, params.prompt, true); // evaluate prompt llama_decode(ctx, llama_batch_get_one(tokens.data(), tokens.size(), n_past, 0)); @@ -72,7 +72,7 @@ int main(int argc, char ** argv) { for (auto i = 0; i < params.n_predict; i++) { auto next_token = llama_sampler_sample(smpl, ctx, -1); - auto next_token_str = llama_token_to_piece(ctx, next_token); + auto next_token_str = common_token_to_piece(ctx, next_token); printf("%s", next_token_str.c_str()); result0 += next_token_str; @@ -92,7 +92,7 @@ int main(int argc, char ** argv) { llama_free(ctx); // make new context - auto * ctx2 = llama_new_context_with_model(model, llama_context_params_from_gpt_params(params)); + auto * ctx2 = llama_new_context_with_model(model, common_context_params_from_gpt_params(params)); llama_sampler * smpl2 = llama_sampler_chain_init(sparams); @@ -128,7 +128,7 @@ int main(int argc, char ** argv) { // second run for (auto i = 0; i < params.n_predict; i++) { auto next_token = llama_sampler_sample(smpl2, ctx2, -1); - auto next_token_str = llama_token_to_piece(ctx2, next_token); + auto next_token_str = common_token_to_piece(ctx2, next_token); printf("%s", next_token_str.c_str()); result1 += next_token_str; @@ -152,7 +152,7 @@ int main(int argc, char ** argv) { } // make new context - auto * ctx3 = llama_new_context_with_model(model, llama_context_params_from_gpt_params(params)); + auto * ctx3 = llama_new_context_with_model(model, common_context_params_from_gpt_params(params)); llama_sampler * smpl3 = llama_sampler_chain_init(sparams); @@ -216,7 +216,7 @@ int main(int argc, char ** argv) { // third run with seq 1 instead of 0 for (auto i = 0; i < params.n_predict; i++) { auto next_token = llama_sampler_sample(smpl3, ctx3, -1); - auto next_token_str = llama_token_to_piece(ctx3, next_token); + auto next_token_str = common_token_to_piece(ctx3, next_token); printf("%s", next_token_str.c_str()); result2 += next_token_str; diff --git a/examples/server/server.cpp b/examples/server/server.cpp index aedfca0d6..2380d402c 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -611,7 +611,7 @@ struct server_response { struct server_context { llama_model * model = nullptr; llama_context * ctx = nullptr; - std::vector loras; + std::vector loras; gpt_params params; @@ -668,7 +668,7 @@ struct server_context { // dedicate one sequence to the system prompt params.n_parallel += 1; - llama_init_result llama_init = llama_init_from_gpt_params(params); + common_init_result llama_init = llama_init_from_gpt_params(params); model = llama_init.model; ctx = llama_init.context; @@ -771,10 +771,10 @@ struct server_context { std::vector p; if (first) { - p = ::llama_tokenize(ctx, s, add_special, TMP_FORCE_SPECIAL); + p = ::common_tokenize(ctx, s, add_special, TMP_FORCE_SPECIAL); first = false; } else { - p = ::llama_tokenize(ctx, s, false, TMP_FORCE_SPECIAL); + p = ::common_tokenize(ctx, s, false, TMP_FORCE_SPECIAL); } prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end()); @@ -788,7 +788,7 @@ struct server_context { } } else { auto s = json_prompt.template get(); - prompt_tokens = ::llama_tokenize(ctx, s, add_special, TMP_FORCE_SPECIAL); + prompt_tokens = ::common_tokenize(ctx, s, add_special, TMP_FORCE_SPECIAL); } return prompt_tokens; @@ -999,7 +999,7 @@ struct server_context { slot.sparams.logit_bias.push_back({tok, bias}); } } else if (el[0].is_string()) { - auto toks = llama_tokenize(model, el[0].get(), false); + auto toks = common_tokenize(model, el[0].get(), false); for (auto tok : toks) { slot.sparams.logit_bias.push_back({tok, bias}); } @@ -1073,7 +1073,7 @@ struct server_context { system_tokens.clear(); if (!system_prompt.empty()) { - system_tokens = ::llama_tokenize(ctx, system_prompt, true); + system_tokens = ::common_tokenize(ctx, system_prompt, true); const int32_t n_batch = llama_n_batch(ctx); const int32_t n_tokens_prompt = system_tokens.size(); @@ -1081,10 +1081,10 @@ struct server_context { for (int32_t i = 0; i < n_tokens_prompt; i += n_batch) { const int32_t n_tokens = std::min(n_batch, n_tokens_prompt - i); - llama_batch_clear(batch); + common_batch_clear(batch); for (int32_t j = 0; j < n_tokens; ++j) { - llama_batch_add(batch, system_tokens[i + j], i + j, { 0 }, false); + common_batch_add(batch, system_tokens[i + j], i + j, { 0 }, false); } if (llama_decode(ctx, batch) != 0) { @@ -1113,7 +1113,7 @@ struct server_context { bool process_token(completion_token_output & result, server_slot & slot) { // remember which tokens were sampled - used for repetition penalties during sampling - const std::string token_str = llama_token_to_piece(ctx, result.tok, params.special); + const std::string token_str = common_token_to_piece(ctx, result.tok, params.special); slot.sampled = result.tok; // search stop word and delete it @@ -1297,7 +1297,7 @@ struct server_context { }; if (slot.sparams.n_probs > 0) { - const std::vector to_send_toks = llama_tokenize(ctx, tkn.text_to_send, false); + const std::vector to_send_toks = common_tokenize(ctx, tkn.text_to_send, false); const size_t probs_pos = std::min(slot.n_sent_token_probs, slot.generated_token_probs.size()); const size_t probs_stop_pos = std::min(slot.n_sent_token_probs + to_send_toks.size(), slot.generated_token_probs.size()); @@ -1347,7 +1347,7 @@ struct server_context { if (slot.sparams.n_probs > 0) { std::vector probs; if (!slot.params.stream && slot.stopped_word) { - const std::vector stop_word_toks = llama_tokenize(ctx, slot.stopping_word, false); + const std::vector stop_word_toks = common_tokenize(ctx, slot.stopping_word, false); size_t safe_offset = std::min(slot.generated_token_probs.size(), stop_word_toks.size()); probs = std::vector( @@ -1401,7 +1401,7 @@ struct server_context { continue; } - llama_embd_normalize(embd, embd_res.data(), n_embd); + common_embd_normalize(embd, embd_res.data(), n_embd); res.data = json { {"embedding", embd_res}, @@ -1835,7 +1835,7 @@ struct server_context { } break; case SERVER_TASK_TYPE_SET_LORA: { - llama_lora_adapters_apply(ctx, loras); + common_lora_adapters_apply(ctx, loras); server_task_result result; result.id = task.id; result.stop = true; @@ -1921,7 +1921,7 @@ struct server_context { } // start populating the batch for this iteration - llama_batch_clear(batch); + common_batch_clear(batch); // frist, add sampled tokens from any ongoing sequences for (auto & slot : slots) { @@ -1935,7 +1935,7 @@ struct server_context { // TODO: we always have to take into account the "system_tokens" // this is not great and needs to be improved somehow - llama_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id + 1 }, true); + common_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id + 1 }, true); slot.n_past += 1; @@ -2184,7 +2184,7 @@ struct server_context { } } - llama_batch_add(batch, prompt_tokens[slot.n_past], system_tokens.size() + slot_npast, { slot.id + 1 }, false); + common_batch_add(batch, prompt_tokens[slot.n_past], system_tokens.size() + slot_npast, { slot.id + 1 }, false); if (slot.params.cache_prompt) { slot.cache_tokens.push_back(prompt_tokens[slot.n_past]); @@ -3014,7 +3014,7 @@ int main(int argc, char ** argv) { if (with_pieces) { for (const auto& token : tokens) { - std::string piece = llama_token_to_piece(ctx_server.ctx, token); + std::string piece = common_token_to_piece(ctx_server.ctx, token); json piece_json; // Check if the piece is valid UTF-8 @@ -3357,7 +3357,7 @@ int main(int argc, char ** argv) { } // print sample chat example to make it clear which template is used - LOG_INF("%s: chat template, built_in: %d, chat_example: '%s'\n", __func__, params.chat_template.empty(), llama_chat_format_example(ctx_server.model, params.chat_template).c_str()); + LOG_INF("%s: chat template, built_in: %d, chat_example: '%s'\n", __func__, params.chat_template.empty(), common_chat_format_example(ctx_server.model, params.chat_template).c_str()); ctx_server.queue_tasks.on_new_task(std::bind( &server_context::process_single_task, &ctx_server, std::placeholders::_1)); diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp index 452606cca..ad99e9574 100644 --- a/examples/server/utils.hpp +++ b/examples/server/utils.hpp @@ -57,7 +57,7 @@ static T json_value(const json & body, const std::string & key, const T & defaul // Format given chat. If tmpl is empty, we take the template from model metadata inline std::string format_chat(const struct llama_model * model, const std::string & tmpl, const std::vector & messages) { - std::vector chat; + std::vector chat; for (size_t i = 0; i < messages.size(); ++i) { const auto & curr_msg = messages[i]; @@ -84,7 +84,7 @@ inline std::string format_chat(const struct llama_model * model, const std::stri chat.push_back({role, content}); } - const auto formatted_chat = llama_chat_apply_template(model, tmpl, chat, true); + const auto formatted_chat = common_chat_apply_template(model, tmpl, chat, true); LOG_DBG("formatted_chat: '%s'\n", formatted_chat.c_str()); return formatted_chat; @@ -246,7 +246,7 @@ template static std::string tokens_to_str(llama_context * ctx, Iter begin, Iter end) { std::string ret; for (; begin != end; ++begin) { - ret += llama_token_to_piece(ctx, *begin); + ret += common_token_to_piece(ctx, *begin); } return ret; @@ -254,7 +254,7 @@ static std::string tokens_to_str(llama_context * ctx, Iter begin, Iter end) { // format incomplete utf-8 multibyte character for output static std::string tokens_to_output_formatted_string(const llama_context * ctx, const llama_token token) { - std::string out = token == -1 ? "" : llama_token_to_piece(ctx, token); + std::string out = token == -1 ? "" : common_token_to_piece(ctx, token); // if the size is 1 and first bit is 1, meaning it's a partial character // (size > 1 meaning it's already a known token) diff --git a/examples/simple/simple.cpp b/examples/simple/simple.cpp index c2b7267c8..a4b3c03cd 100644 --- a/examples/simple/simple.cpp +++ b/examples/simple/simple.cpp @@ -33,7 +33,7 @@ int main(int argc, char ** argv) { // initialize the model - llama_model_params model_params = llama_model_params_from_gpt_params(params); + llama_model_params model_params = common_model_params_from_gpt_params(params); llama_model * model = llama_load_model_from_file(params.model.c_str(), model_params); @@ -44,7 +44,7 @@ int main(int argc, char ** argv) { // initialize the context - llama_context_params ctx_params = llama_context_params_from_gpt_params(params); + llama_context_params ctx_params = common_context_params_from_gpt_params(params); llama_context * ctx = llama_new_context_with_model(model, ctx_params); @@ -64,7 +64,7 @@ int main(int argc, char ** argv) { // tokenize the prompt std::vector tokens_list; - tokens_list = ::llama_tokenize(ctx, params.prompt, true); + tokens_list = ::common_tokenize(ctx, params.prompt, true); const int n_ctx = llama_n_ctx(ctx); const int n_kv_req = tokens_list.size() + (n_predict - tokens_list.size()); @@ -84,7 +84,7 @@ int main(int argc, char ** argv) { LOG("\n"); for (auto id : tokens_list) { - LOG("%s", llama_token_to_piece(ctx, id).c_str()); + LOG("%s", common_token_to_piece(ctx, id).c_str()); } // create a llama_batch with size 512 @@ -94,7 +94,7 @@ int main(int argc, char ** argv) { // evaluate the initial prompt for (size_t i = 0; i < tokens_list.size(); i++) { - llama_batch_add(batch, tokens_list[i], i, { 0 }, false); + common_batch_add(batch, tokens_list[i], i, { 0 }, false); } // llama_decode will output logits only for the last token of the prompt @@ -124,14 +124,14 @@ int main(int argc, char ** argv) { break; } - LOG("%s", llama_token_to_piece(ctx, new_token_id).c_str()); + LOG("%s", common_token_to_piece(ctx, new_token_id).c_str()); fflush(stdout); // prepare the next batch - llama_batch_clear(batch); + common_batch_clear(batch); // push this new token for next evaluation - llama_batch_add(batch, new_token_id, n_cur, { 0 }, true); + common_batch_add(batch, new_token_id, n_cur, { 0 }, true); n_decode += 1; } diff --git a/examples/speculative/speculative.cpp b/examples/speculative/speculative.cpp index adf6255e1..815dab2a7 100644 --- a/examples/speculative/speculative.cpp +++ b/examples/speculative/speculative.cpp @@ -66,7 +66,7 @@ int main(int argc, char ** argv) { llama_context * ctx_dft = NULL; // load the target model - llama_init_result llama_init_tgt = llama_init_from_gpt_params(params); + common_init_result llama_init_tgt = llama_init_from_gpt_params(params); model_tgt = llama_init_tgt.model; ctx_tgt = llama_init_tgt.context; @@ -78,7 +78,7 @@ int main(int argc, char ** argv) { } params.cpuparams_batch.n_threads = params.draft_cpuparams_batch.n_threads; - llama_init_result llama_init_dft = llama_init_from_gpt_params(params); + common_init_result llama_init_dft = llama_init_from_gpt_params(params); model_dft = llama_init_dft.model; ctx_dft = llama_init_dft.context; @@ -124,8 +124,8 @@ int main(int argc, char ** argv) { if (std::strcmp(token_text_tgt, token_text_dft) != 0) { LOG_ERR("%s: draft model vocab must match target model to use speculation but ", __func__); LOG_ERR("token %d content differs - target '%s', draft '%s'\n", i, - llama_token_to_piece(ctx_tgt, i).c_str(), - llama_token_to_piece(ctx_dft, i).c_str()); + common_token_to_piece(ctx_tgt, i).c_str(), + common_token_to_piece(ctx_dft, i).c_str()); return 1; } } @@ -134,7 +134,7 @@ int main(int argc, char ** argv) { // Tokenize the prompt std::vector inp; - inp = ::llama_tokenize(ctx_tgt, params.prompt, true, true); + inp = ::common_tokenize(ctx_tgt, params.prompt, true, true); const int max_context_size = llama_n_ctx(ctx_tgt); const int max_tokens_list_size = max_context_size - 4; @@ -147,7 +147,7 @@ int main(int argc, char ** argv) { LOG("\n\n"); for (auto id : inp) { - LOG("%s", llama_token_to_piece(ctx_tgt, id).c_str()); + LOG("%s", common_token_to_piece(ctx_tgt, id).c_str()); } const int n_input = inp.size(); @@ -277,13 +277,13 @@ int main(int argc, char ** argv) { s_keep = s; accept = true; token_id = drafts[s].tokens[i_dft]; - token_str = llama_token_to_piece(ctx_tgt, token_id); + token_str = common_token_to_piece(ctx_tgt, token_id); gpt_sampler_accept(smpl, token_id, true); LOG_DBG("draft token %d of sequence %d (%d, '%s') accepted\n", i_dft, s, token_id, token_str.c_str()); break; } else { - LOG_DBG("draft token %d of sequence %d (%d, '%s') rejected\n", i_dft, s, drafts[s].tokens[i_dft], llama_token_to_piece(ctx_tgt, drafts[s].tokens[i_dft]).c_str()); + LOG_DBG("draft token %d of sequence %d (%d, '%s') rejected\n", i_dft, s, drafts[s].tokens[i_dft], common_token_to_piece(ctx_tgt, drafts[s].tokens[i_dft]).c_str()); drafts[s].active = false; // calculate residual probability @@ -350,7 +350,7 @@ int main(int argc, char ** argv) { token_id = dist_tgt.data[idx].id; gpt_sampler_accept(smpl, token_id, true); - token_str = llama_token_to_piece(ctx_tgt, token_id); + token_str = common_token_to_piece(ctx_tgt, token_id); } } else { // greedy verification @@ -361,7 +361,7 @@ int main(int argc, char ** argv) { gpt_sampler_accept(smpl, token_id, true); - token_str = llama_token_to_piece(ctx_tgt, token_id); + token_str = common_token_to_piece(ctx_tgt, token_id); for (int s = 0; s < n_seq_dft; ++s) { if (!drafts[s].active) { @@ -431,8 +431,8 @@ int main(int argc, char ** argv) { drafts[0].dists.push_back(std::vector()); drafts[0].i_batch_tgt.push_back(0); - llama_batch_clear(batch_dft); - llama_batch_add (batch_dft, token_id, n_past_dft, { 0 }, true); + common_batch_clear(batch_dft); + common_batch_add (batch_dft, token_id, n_past_dft, { 0 }, true); llama_kv_cache_seq_rm(ctx_dft, 0, n_past_dft, -1); // LOG_DBG("dft batch: %s\n", LOG_BATCH_TOSTR_PRETTY(ctx_dft, batch_dft).c_str()); @@ -461,8 +461,8 @@ int main(int argc, char ** argv) { drafts[0].drafting = true; drafts[0].i_batch_dft = 0; - llama_batch_clear(batch_tgt); - llama_batch_add (batch_tgt, drafts[0].tokens[0], n_past_tgt, { 0 }, true); + common_batch_clear(batch_tgt); + common_batch_add (batch_tgt, drafts[0].tokens[0], n_past_tgt, { 0 }, true); // sample n_draft tokens from the draft model using tree-based sampling for (int i = 0; i < n_draft; ++i) { @@ -483,7 +483,7 @@ int main(int argc, char ** argv) { for (int k = 0; k < std::min(n_seq_dft + 3, (int) cur_p->size); ++k) { LOG_DBG(" - draft candidate %3d for seq %3d, pos %3d: %6d (%8.3f) '%s'\n", - k, s, i, cur_p->data[k].id, cur_p->data[k].p, llama_token_to_piece(ctx_dft, cur_p->data[k].id).c_str()); + k, s, i, cur_p->data[k].id, cur_p->data[k].p, common_token_to_piece(ctx_dft, cur_p->data[k].id).c_str()); } std::vector sa(1, s); @@ -545,12 +545,12 @@ int main(int argc, char ** argv) { // add unique drafted tokens to the target batch drafts[s].i_batch_tgt.push_back(batch_tgt.n_tokens); - llama_batch_add(batch_tgt, id, n_past_tgt + i + 1, { s }, true); + common_batch_add(batch_tgt, id, n_past_tgt + i + 1, { s }, true); // add the token to the batch for batched decoding with the draft model drafts[s].i_batch_dft = batch_dft.n_tokens; - llama_batch_add(batch_dft, id, n_past_cur, { s }, true); + common_batch_add(batch_dft, id, n_past_cur, { s }, true); if (batch_tgt.n_tokens > n_draft) { drafts[s].drafting = false; diff --git a/examples/tokenize/tokenize.cpp b/examples/tokenize/tokenize.cpp index a9af6471f..85cf44952 100644 --- a/examples/tokenize/tokenize.cpp +++ b/examples/tokenize/tokenize.cpp @@ -365,7 +365,7 @@ int main(int raw_argc, char ** raw_argv) { const bool parse_special = !no_parse_special; std::vector tokens; - tokens = ::llama_tokenize(model, prompt, add_bos, parse_special); + tokens = ::common_tokenize(model, prompt, add_bos, parse_special); if (printing_ids) { printf("["); @@ -380,7 +380,7 @@ int main(int raw_argc, char ** raw_argv) { } else { bool invalid_utf8 = false; printf("%6d -> '", tokens[i]); - write_utf8_cstr_to_stdout(llama_token_to_piece(ctx, tokens[i]).c_str(), invalid_utf8); + write_utf8_cstr_to_stdout(common_token_to_piece(ctx, tokens[i]).c_str(), invalid_utf8); if (invalid_utf8) { printf("' (utf-8 decode failure)\n"); } else { diff --git a/tests/test-chat-template.cpp b/tests/test-chat-template.cpp index a8222caee..6f046249f 100644 --- a/tests/test-chat-template.cpp +++ b/tests/test-chat-template.cpp @@ -140,11 +140,11 @@ int main(void) { // test llama_chat_format_single for system message printf("\n\n=== llama_chat_format_single (system message) ===\n\n"); - std::vector chat2; - llama_chat_msg sys_msg{"system", "You are a helpful assistant"}; + std::vector chat2; + common_chat_msg sys_msg{"system", "You are a helpful assistant"}; auto fmt_sys = [&](std::string tmpl) { - auto output = llama_chat_format_single(nullptr, tmpl, chat2, sys_msg, false); + auto output = common_chat_format_single(nullptr, tmpl, chat2, sys_msg, false); printf("fmt_sys(%s) : %s\n", tmpl.c_str(), output.c_str()); printf("-------------------------\n"); return output; @@ -160,10 +160,10 @@ int main(void) { chat2.push_back({"system", "You are a helpful assistant"}); chat2.push_back({"user", "Hello"}); chat2.push_back({"assistant", "I am assistant"}); - llama_chat_msg new_msg{"user", "How are you"}; + common_chat_msg new_msg{"user", "How are you"}; auto fmt_single = [&](std::string tmpl) { - auto output = llama_chat_format_single(nullptr, tmpl, chat2, new_msg, true); + auto output = common_chat_format_single(nullptr, tmpl, chat2, new_msg, true); printf("fmt_single(%s) : %s\n", tmpl.c_str(), output.c_str()); printf("-------------------------\n"); return output; diff --git a/tests/test-tokenizer-0.cpp b/tests/test-tokenizer-0.cpp index 4d49850c9..0af85f002 100644 --- a/tests/test-tokenizer-0.cpp +++ b/tests/test-tokenizer-0.cpp @@ -202,7 +202,7 @@ int main(int argc, char **argv) { for (int i = 0; i < nthread; i++) { threads[i] = std::thread([&, i]() { for (const auto & test_kv : k_tests) { - const std::vector res = llama_tokenize(ctx, test_kv.first, add_special, false); + const std::vector res = common_tokenize(ctx, test_kv.first, add_special, false); // here only print the result of the first thread // because the other threads are running the same tests @@ -212,7 +212,7 @@ int main(int argc, char **argv) { printf("\n"); printf("src: '%s'\n", test_kv.first.c_str()); - printf("res: '%s'\n", llama_detokenize(ctx, res).c_str()); + printf("res: '%s'\n", common_detokenize(ctx, res).c_str()); printf("tok: "); for (const auto & tok : res) { printf("%d ", tok); @@ -229,16 +229,16 @@ int main(int argc, char **argv) { 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(ctx, res).c_str(), - llama_detokenize(ctx, test_kv.second).c_str()); + common_detokenize(ctx, res).c_str(), + common_detokenize(ctx, test_kv.second).c_str()); fprintf(stderr, "%s : expected tokens: ", __func__); for (const auto & t : test_kv.second) { - fprintf(stderr, "%6d '%s', ", t, llama_token_to_piece(ctx, t).c_str()); + fprintf(stderr, "%6d '%s', ", t, common_token_to_piece(ctx, t).c_str()); } fprintf(stderr, "\n"); fprintf(stderr, "%s : got tokens: ", __func__); for (const auto & t : res) { - fprintf(stderr, "%6d '%s', ", t, llama_token_to_piece(ctx, t).c_str()); + fprintf(stderr, "%6d '%s', ", t, common_token_to_piece(ctx, t).c_str()); } fprintf(stderr, "\n"); @@ -273,7 +273,7 @@ int main(int argc, char **argv) { { const auto t_start = ggml_time_us(); - res = llama_tokenize(ctx, text, add_special, false); + res = common_tokenize(ctx, text, add_special, false); const auto t_end = ggml_time_us(); diff --git a/tests/test-tokenizer-1-bpe.cpp b/tests/test-tokenizer-1-bpe.cpp index 9498387e0..0ff7fc833 100644 --- a/tests/test-tokenizer-1-bpe.cpp +++ b/tests/test-tokenizer-1-bpe.cpp @@ -78,10 +78,10 @@ int main(int argc, char **argv) { const int n_vocab = llama_n_vocab(model); for (int i = 0; i < n_vocab; ++i) { - std::string str = llama_detokenize(ctx, std::vector(1, i)); + std::string str = common_detokenize(ctx, std::vector(1, i)); try { auto cps = unicode_cpts_from_utf8(str); - std::vector tokens = llama_tokenize(ctx, str, false, true); + std::vector tokens = common_tokenize(ctx, str, false, true); if (ignore_merges && tokens.size() > 1) { fprintf(stderr, "%s : error: token %d detokenizes to '%s'(%zu) but " @@ -94,7 +94,7 @@ int main(int argc, char **argv) { fprintf(stderr, "]\n"); return 2; } - std::string check = llama_detokenize(ctx, tokens); + std::string check = common_detokenize(ctx, tokens); if (check != str) { fprintf(stderr, "%s : error: token %d detokenizes to '%s'(%zu) but tokenization of this detokenizes to '%s'(%zu)\n", __func__, i, str.c_str(), str.length(), check.c_str(), check.length()); @@ -123,8 +123,8 @@ int main(int argc, char **argv) { } std::string str = unicode_cpt_to_utf8(cp); - std::vector tokens = llama_tokenize(ctx, str, false); - std::string check = llama_detokenize(ctx, tokens); + std::vector tokens = common_tokenize(ctx, str, false); + std::string check = common_detokenize(ctx, tokens); if (cp != 9601 && str != check) { fprintf(stderr, "error: codepoint 0x%x detokenizes to '%s'(%zu) instead of '%s'(%zu)\n", cp, check.c_str(), check.length(), str.c_str(), str.length()); diff --git a/tests/test-tokenizer-1-spm.cpp b/tests/test-tokenizer-1-spm.cpp index 7ca9e2ca6..9b0716a43 100644 --- a/tests/test-tokenizer-1-spm.cpp +++ b/tests/test-tokenizer-1-spm.cpp @@ -66,9 +66,9 @@ int main(int argc, char ** argv) { const int n_vocab = llama_n_vocab(model); for (int i = 0; i < n_vocab; ++i) { - std::string str = llama_detokenize(ctx, std::vector(1, i), true); - std::vector tokens = llama_tokenize(ctx, str, false, true); - std::string check = llama_detokenize(ctx, tokens); + std::string str = common_detokenize(ctx, std::vector(1, i), true); + std::vector tokens = common_tokenize(ctx, str, false, true); + std::string check = common_detokenize(ctx, tokens); if (check != str) { fprintf(stderr, "%s : error: token %d detokenizes to '%s'(%zu) but tokenization of this detokenizes to '%s'(%zu)\n", __func__, i, str.c_str(), str.length(), check.c_str(), check.length()); @@ -93,8 +93,8 @@ int main(int argc, char ** argv) { } std::string str = unicode_cpt_to_utf8(cp); - std::vector tokens = llama_tokenize(ctx, str, false, true); - std::string check = llama_detokenize(ctx, tokens); + std::vector tokens = common_tokenize(ctx, str, false, true); + std::string check = common_detokenize(ctx, tokens); if (cp != 9601 && str != check) { fprintf(stderr, "error: codepoint 0x%x detokenizes to '%s'(%zu) instead of '%s'(%zu)\n", cp, check.c_str(), check.length(), str.c_str(), str.length());