Merge branch 'master' of https://github.com/VJHack/llama.cpp

Author: VJHack

.github/workflows/build.yml

@@ -23,6 +23,9 @@ env:
   BRANCH_NAME: ${{ github.head_ref || github.ref_name }}
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
   macOS-latest-cmake-arm64:

.github/workflows/server.yml

@@ -20,6 +20,12 @@ on:
     types: [opened, synchronize, reopened]
     paths: ['.github/workflows/server.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m', 'examples/server/**.*']
 
+env:
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_VERBOSITY: 10
+
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || github.run_id }}
   cancel-in-progress: true

CMakeLists.txt

@@ -82,11 +82,11 @@ set(GGML_FATAL_WARNINGS     ${LLAMA_FATAL_WARNINGS})
 
 # change the default for these ggml options
 if (NOT DEFINED GGML_LLAMAFILE)
-    set(GGML_LLAMAFILE ON)
+    set(GGML_LLAMAFILE_DEFAULT ON)
 endif()
 
-if (NOT DEFINED GGML_CUDA_USE_GRAPHS)
-    set(GGML_CUDA_USE_GRAPHS ON)
+if (NOT DEFINED GGML_CUDA_GRAPHS)
+    set(GGML_CUDA_GRAPHS_DEFAULT ON)
 endif()
 
 # transition helpers

Makefile

@@ -54,6 +54,7 @@ TEST_TARGETS = \
 	tests/test-grammar-parser \
 	tests/test-json-schema-to-grammar \
 	tests/test-llama-grammar \
+	tests/test-log \
 	tests/test-model-load-cancel \
 	tests/test-opt \
 	tests/test-quantize-fns \
@@ -148,6 +149,14 @@ GGML_NO_METAL := 1
 DEPRECATE_WARNING := 1
 endif
 
+ifdef LLAMA_DISABLE_LOGS
+REMOVE_WARNING := 1
+endif
+
+ifdef LLAMA_SERVER_VERBOSE
+REMOVE_WARNING := 1
+endif
+
 ifndef UNAME_S
 UNAME_S := $(shell uname -s)
 endif
@@ -351,19 +360,11 @@ ifdef LLAMA_SANITIZE_UNDEFINED
 	MK_LDFLAGS  += -fsanitize=undefined -g
 endif
 
-ifdef LLAMA_SERVER_VERBOSE
-	MK_CPPFLAGS += -DSERVER_VERBOSE=$(LLAMA_SERVER_VERBOSE)
-endif
-
 ifdef LLAMA_SERVER_SSL
 	MK_CPPFLAGS += -DCPPHTTPLIB_OPENSSL_SUPPORT
 	MK_LDFLAGS += -lssl -lcrypto
 endif
 
-ifdef LLAMA_DISABLE_LOGS
-	MK_CPPFLAGS += -DLOG_DISABLE_LOGS
-endif # LLAMA_DISABLE_LOGS
-
 # warnings
 WARN_FLAGS = \
 	-Wall \
@@ -618,7 +619,7 @@ ifdef GGML_CUDA
 			CUDA_PATH ?= /usr/local/cuda
 		endif
 
-		MK_CPPFLAGS  += -DGGML_USE_CUDA -I$(CUDA_PATH)/include -I$(CUDA_PATH)/targets/$(UNAME_M)-linux/include -DGGML_CUDA_USE_GRAPHS
+		MK_CPPFLAGS  += -DGGML_USE_CUDA -DGGML_CUDA_USE_GRAPHS -I$(CUDA_PATH)/include -I$(CUDA_PATH)/targets/$(UNAME_M)-linux/include
 		MK_LDFLAGS   += -lcuda -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L$(CUDA_PATH)/lib64 -L/usr/lib64 -L$(CUDA_PATH)/targets/$(UNAME_M)-linux/lib -L$(CUDA_PATH)/lib64/stubs -L/usr/lib/wsl/lib
 		MK_NVCCFLAGS += -use_fast_math
 	endif # GGML_MUSA
@@ -931,6 +932,7 @@ OBJ_LLAMA = \
 OBJ_COMMON = \
 	common/common.o \
 	common/arg.o \
+	common/log.o \
 	common/console.o \
 	common/ngram-cache.o \
 	common/sampling.o \
@@ -1027,6 +1029,14 @@ $(info   - LLAMA_NO_CCACHE)
 $(info )
 endif
 
+ifdef REMOVE_WARNING
+$(info !!! REMOVAL WARNING !!!)
+$(info The following LLAMA_ options have been removed and are no longer supported)
+$(info   - LLAMA_DISABLE_LOGS   (https://github.com/ggerganov/llama.cpp/pull/9418))
+$(info   - LLAMA_SERVER_VERBOSE (https://github.com/ggerganov/llama.cpp/pull/9418))
+$(info )
+endif
+
 #
 # Build libraries
 #
@@ -1168,6 +1178,11 @@ common/arg.o: \
 	common/arg.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 
+common/log.o: \
+	common/log.cpp \
+	common/log.h
+	$(CXX) $(CXXFLAGS) -c $< -o $@
+
 common/sampling.o: \
 	common/sampling.cpp \
 	common/sampling.h \
@@ -1346,7 +1361,7 @@ llama-cvector-generator: examples/cvector-generator/cvector-generator.cpp \
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
 llama-convert-llama2c-to-ggml: examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp \
-	$(OBJ_GGML) $(OBJ_LLAMA)
+	$(OBJ_ALL)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
@@ -1528,6 +1543,11 @@ tests/test-llama-grammar: tests/test-llama-grammar.cpp \
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
+tests/test-log: tests/test-log.cpp \
+	$(OBJ_ALL)
+	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
+	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
+
 tests/test-grammar-parser: tests/test-grammar-parser.cpp \
 	$(OBJ_ALL)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)

README.md

@@ -77,6 +77,7 @@ Typically finetunes of the base models below are supported as well.
 - [x] [SEA-LION](https://huggingface.co/models?search=sea-lion)
 - [x] [GritLM-7B](https://huggingface.co/GritLM/GritLM-7B) + [GritLM-8x7B](https://huggingface.co/GritLM/GritLM-8x7B)
 - [x] [OLMo](https://allenai.org/olmo)
+- [x] [OLMoE](https://huggingface.co/allenai/OLMoE-1B-7B-0924)
 - [x] [Granite models](https://huggingface.co/collections/ibm-granite/granite-code-models-6624c5cec322e4c148c8b330)
 - [x] [GPT-NeoX](https://github.com/EleutherAI/gpt-neox) + [Pythia](https://github.com/EleutherAI/pythia)
 - [x] [Snowflake-Arctic MoE](https://huggingface.co/collections/Snowflake/arctic-66290090abe542894a5ac520)

ci/run.sh

@@ -737,6 +737,9 @@ function gg_sum_embd_bge_small {
 
 ## main
 
+export LLAMA_LOG_PREFIX=1
+export LLAMA_LOG_TIMESTAMPS=1
+
 if [ -z ${GG_BUILD_LOW_PERF} ]; then
     # Create symlink: ./llama.cpp/models-mnt -> $MNT/models/models-mnt
     rm -rf ${SRC}/models-mnt

common/CMakeLists.txt

@@ -51,21 +51,23 @@ endif()
 set(TARGET common)
 
 add_library(${TARGET} STATIC
+    arg.cpp
+    arg.h
     base64.hpp
-    common.h
     common.cpp
-    arg.h
-    arg.cpp
-    sampling.h
-    sampling.cpp
-    console.h
+    common.h
     console.cpp
-    json.hpp
+    console.h
     json-schema-to-grammar.cpp
-    train.h
-    train.cpp
-    ngram-cache.h
+    json.hpp
+    log.cpp
+    log.h
     ngram-cache.cpp
+    ngram-cache.h
+    sampling.cpp
+    sampling.h
+    train.cpp
+    train.h
     )
 
 if (BUILD_SHARED_LIBS)

common/arg.cpp

@@ -1,15 +1,17 @@
 #include "arg.h"
 
+#include "log.h"
 #include "sampling.h"
 
 #include <algorithm>
-#include <string>
-#include <vector>
-#include <set>
+#include <climits>
+#include <cstdarg>
 #include <fstream>
 #include <regex>
-#include <cstdarg>
-#include <climits>
+#include <set>
+#include <string>
+#include <thread>
+#include <vector>
 
 #include "json-schema-to-grammar.h"
 
@@ -383,20 +385,6 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             exit(0);
         }
     ));
-    add_opt(llama_arg(
-        {"-v", "--verbose"},
-        "print verbose information",
-        [](gpt_params & params) {
-            params.verbosity = 1;
-        }
-    ));
-    add_opt(llama_arg(
-        {"--verbosity"}, "N",
-        format("set specific verbosity level (default: %d)", params.verbosity),
-        [](gpt_params & params, int value) {
-            params.verbosity = value;
-        }
-    ));
     add_opt(llama_arg(
         {"--verbose-prompt"},
         format("print a verbose prompt before generation (default: %s)", params.verbose_prompt ? "true" : "false"),
@@ -417,7 +405,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
         [](gpt_params & params) {
             params.use_color = true;
         }
-    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_INFILL}));
+    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_INFILL, LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP}));
     add_opt(llama_arg(
         {"-t", "--threads"}, "N",
         format("number of threads to use during generation (default: %d)", params.cpuparams.n_threads),
@@ -697,6 +685,13 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             params.n_keep = value;
         }
     ));
+    add_opt(llama_arg(
+        {"--no-context-shift"},
+        format("disables context shift on inifinite text generation (default: %s)", params.ctx_shift ? "disabled" : "enabled"),
+        [](gpt_params & params) {
+            params.ctx_shift = false;
+        }
+    ).set_examples({LLAMA_EXAMPLE_MAIN}));
     add_opt(llama_arg(
         {"--chunks"}, "N",
         format("max number of chunks to process (default: %d, -1 = all)", params.n_chunks),
@@ -876,15 +871,15 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             params.input_prefix = value;
             params.enable_chat_template = false;
         }
-    ).set_examples({LLAMA_EXAMPLE_MAIN}));
+    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_INFILL}));
     add_opt(llama_arg(
         {"--in-suffix"}, "STRING",
         "string to suffix after user inputs with (default: empty)",
         [](gpt_params & params, const std::string & value) {
             params.input_suffix = value;
             params.enable_chat_template = false;
         }
-    ).set_examples({LLAMA_EXAMPLE_MAIN}));
+    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_INFILL}));
     add_opt(llama_arg(
         {"--no-warmup"},
         "skip warming up the model with an empty run",
@@ -1824,19 +1819,6 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             params.system_prompt = system_prompt;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}));
-    add_opt(llama_arg(
-        {"--log-format"}, "{text, json}",
-        "log output format: json or text (default: json)",
-        [](gpt_params & params, const std::string & value) {
-            if (value == "json") {
-                params.log_json = true;
-            } else if (value == "text") {
-                params.log_json = false;
-            } else {
-                throw std::invalid_argument("invalid value");
-            }
-        }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}));
     add_opt(llama_arg(
         {"--metrics"},
         format("enable prometheus compatible metrics endpoint (default: %s)", params.endpoint_metrics ? "enabled" : "disabled"),
@@ -1956,40 +1938,57 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             else { std::invalid_argument("invalid value"); }
         }
     ).set_examples({LLAMA_EXAMPLE_BENCH}));
-#ifndef LOG_DISABLE_LOGS
-    // TODO: make this looks less weird
-    add_opt(llama_arg(
-        {"--log-test"},
-        "Log test",
-        [](gpt_params &) { log_param_single_parse("--log-test"); }
-    ));
     add_opt(llama_arg(
         {"--log-disable"},
         "Log disable",
-        [](gpt_params &) { log_param_single_parse("--log-disable"); }
+        [](gpt_params &) {
+            gpt_log_pause(gpt_log_main());
+        }
     ));
     add_opt(llama_arg(
-        {"--log-enable"},
-        "Log enable",
-        [](gpt_params &) { log_param_single_parse("--log-enable"); }
+        {"--log-file"}, "FNAME",
+        "Log to file",
+        [](gpt_params &, const std::string & value) {
+            gpt_log_set_file(gpt_log_main(), value.c_str());
+        }
     ));
     add_opt(llama_arg(
-        {"--log-new"},
-        "Log new",
-        [](gpt_params &) { log_param_single_parse("--log-new"); }
-    ));
+        {"--log-colors"},
+        "Enable colored logging",
+        [](gpt_params &) {
+            gpt_log_set_colors(gpt_log_main(), true);
+        }
+    ).set_env("LLAMA_LOG_COLORS"));
     add_opt(llama_arg(
-        {"--log-append"},
-        "Log append",
-        [](gpt_params &) { log_param_single_parse("--log-append"); }
+        {"-v", "--verbose", "--log-verbose"},
+        "Set verbosity level to infinity (i.e. log all messages, useful for debugging)",
+        [](gpt_params & params) {
+            params.verbosity = INT_MAX;
+            gpt_log_set_verbosity_thold(INT_MAX);
+        }
     ));
     add_opt(llama_arg(
-        {"--log-file"}, "FNAME",
-        "Log file",
-        [](gpt_params &, const std::string & value) { log_param_pair_parse(false, "--log-file", value); }
-    ));
-#endif // LOG_DISABLE_LOGS
+        {"-lv", "--verbosity", "--log-verbosity"}, "N",
+        "Set the verbosity threshold. Messages with a higher verbosity will be ignored.",
+        [](gpt_params & params, int value) {
+            params.verbosity = value;
+            gpt_log_set_verbosity_thold(value);
+        }
+    ).set_env("LLAMA_LOG_VERBOSITY"));
+    add_opt(llama_arg(
+        {"--log-prefix"},
+        "Enable prefx in log messages",
+        [](gpt_params &) {
+            gpt_log_set_prefix(gpt_log_main(), true);
+        }
+    ).set_env("LLAMA_LOG_PREFIX"));
+    add_opt(llama_arg(
+        {"--log-timestamps"},
+        "Enable timestamps in log messages",
+        [](gpt_params &) {
+            gpt_log_set_timestamps(gpt_log_main(), true);
+        }
+    ).set_env("LLAMA_LOG_TIMESTAMPS"));
 
     return ctx_arg;
 }
-