Merge branch 'master' of github.com:ggerganov/llama.cpp into grammar-…

…example

* 'master' of github.com:ggerganov/llama.cpp:
  convert : remove bug in convert.py permute function (ggerganov#3364)
  make-ggml.py : compatibility with more models and GGUF (ggerganov#3290)
  gguf : fix a few general keys (ggerganov#3341)
  metal : reusing llama.cpp logging (ggerganov#3152)
  build : add ACCELERATE_NEW_LAPACK to fix warning on macOS Sonoma (ggerganov#3342)
  readme : add some recent perplexity and bpw measurements to READMES, link for k-quants (ggerganov#3340)
  cmake : fix build-info.h on MSVC (ggerganov#3309)
  docs: Fix typo CLBlast_DIR var. (ggerganov#3330)
  nix : add cuda, use a symlinked toolkit for cmake (ggerganov#3202)

CMakeLists.txt

@@ -118,7 +118,7 @@ if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/.git")
     add_custom_command(
         OUTPUT "${CMAKE_CURRENT_SOURCE_DIR}/build-info.h"
         COMMENT "Generating build details from Git"
-        COMMAND ${CMAKE_COMMAND} -P "${CMAKE_CURRENT_SOURCE_DIR}/scripts/build-info.cmake"
+        COMMAND ${CMAKE_COMMAND} -DMSVC=${MSVC} -DCMAKE_C_COMPILER_VERSION=${CMAKE_C_COMPILER_VERSION} -DCMAKE_C_COMPILER_ID=${CMAKE_C_COMPILER_ID} -DCMAKE_VS_PLATFORM_NAME=${CMAKE_VS_PLATFORM_NAME} -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER} -P "${CMAKE_CURRENT_SOURCE_DIR}/scripts/build-info.cmake"
         WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
         DEPENDS "${GIT_DIR}/index"
         VERBATIM
@@ -162,6 +162,8 @@ if (APPLE AND LLAMA_ACCELERATE)
         message(STATUS "Accelerate framework found")
 
         add_compile_definitions(GGML_USE_ACCELERATE)
+        add_compile_definitions(ACCELERATE_NEW_LAPACK)
+        add_compile_definitions(ACCELERATE_LAPACK_ILP64)
         set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} ${ACCELERATE_FRAMEWORK})
     else()
         message(WARNING "Accelerate framework not found")

Makefile

@@ -305,6 +305,8 @@ ifndef LLAMA_NO_ACCELERATE
 	# `-framework Accelerate` works both with Apple Silicon and Mac Intel
 	ifeq ($(UNAME_S),Darwin)
 		MK_CPPFLAGS += -DGGML_USE_ACCELERATE
+		MK_CPPFLAGS += -DACCELERATE_NEW_LAPACK
+		MK_CPPFLAGS += -DACCELERATE_LAPACK_ILP64
 		MK_LDFLAGS  += -framework Accelerate
 	endif
 endif # LLAMA_NO_ACCELERATE

Package.swift

@@ -45,6 +45,8 @@ let package = Package(
                 .unsafeFlags(["-Wno-shorten-64-to-32"]),
                 .define("GGML_USE_K_QUANTS"),
                 .define("GGML_USE_ACCELERATE")
+                .define("ACCELERATE_NEW_LAPACK")
+                .define("ACCELERATE_LAPACK_ILP64")
             ] + additionalSettings,
             linkerSettings: [
                 .linkedFramework("Accelerate")

README.md

@@ -501,7 +501,7 @@ Building the program with BLAS support may lead to some performance improvements
     ```sh
     mkdir build
     cd build
-    cmake .. -DLLAMA_CLBLAST=ON -DCLBlast_dir=/some/path
+    cmake .. -DLLAMA_CLBLAST=ON -DCLBlast_DIR=/some/path
     cmake --build . --config Release
     ```
   - CMake (Windows):
@@ -597,6 +597,11 @@ Several quantization methods are supported. They differ in the resulting model d
 |   13B | ms/tok @ 8th |      - |     73 |     82 |     98 |    105 |    128 |
 |   13B | bits/weight  |   16.0 |    4.5 |    5.0 |    5.5 |    6.0 |    8.5 |
 
+- [k-quants](https://github.com/ggerganov/llama.cpp/pull/1684)
+- recent k-quants improvements
+  - [#2707](https://github.com/ggerganov/llama.cpp/pull/2707)
+  - [#2807](https://github.com/ggerganov/llama.cpp/pull/2807)
+
 ### Perplexity (measuring model quality)
 
 You can use the `perplexity` example to measure perplexity over a given prompt (lower perplexity is better).

convert.py

@@ -439,7 +439,7 @@ def __repr__(self) -> str:
 def permute(weights: NDArray, n_head: int, n_head_kv: int) -> NDArray:
     #print( "permute debug " + str(weights.shape[0]) + " x " + str(weights.shape[1]) + " nhead " + str(n_head) + " nheadkv " + str(n_kv_head) )
     if n_head_kv is not None and n_head != n_head_kv:
-        n_head //= n_head_kv
+        n_head = n_head_kv
     return (weights.reshape(n_head, 2, weights.shape[0] // n_head // 2, *weights.shape[1:])
                 .swapaxes(1, 2)
                 .reshape(weights.shape))

examples/gptneox-wip/falcon-main.cpp

@@ -367,10 +367,10 @@ bool falcon_model_load(const std::string & fname, falcon_model & model, gpt2bpe_
         keyidx = gguf_find_key(ggufctx, "general.architecture");
         if (keyidx != -1) { printf("%s: model architecture   = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); }
         keyidx = gguf_find_key(ggufctx, "general.file_type");
-        if (keyidx != -1) { printf("%s: model file type      = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); }
+        if (keyidx != -1) { printf("%s: model file type      = %" PRIu32 "\n", __func__, gguf_get_val_u32(ggufctx, keyidx)); }
         keyidx = gguf_find_key(ggufctx, "gptneox.tensor_data_layout");
         if (keyidx != -1) { printf("%s: model data layout    = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); }
-        keyidx = gguf_find_key(ggufctx, "general.source.hugginface.repository");
+        keyidx = gguf_find_key(ggufctx, "general.source.huggingface.repository");
         if (keyidx != -1) { printf("%s: model source HF repo = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); }
     }
 

examples/gptneox-wip/gptneox-main.cpp

@@ -380,10 +380,10 @@ bool gpt_neox_model_load(const std::string & fname, gpt_neox_model & model, gpt2
         keyidx = gguf_find_key(ggufctx, "general.architecture");
         if (keyidx != -1) { printf("%s: model architecture   = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); }
         keyidx = gguf_find_key(ggufctx, "general.file_type");
-        if (keyidx != -1) { printf("%s: model file type      = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); }
+        if (keyidx != -1) { printf("%s: model file type      = %" PRIu32 "\n", __func__, gguf_get_val_u32(ggufctx, keyidx)); }
         keyidx = gguf_find_key(ggufctx, "gptneox.tensor_data_layout");
         if (keyidx != -1) { printf("%s: model data layout    = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); }
-        keyidx = gguf_find_key(ggufctx, "general.source.hugginface.repository");
+        keyidx = gguf_find_key(ggufctx, "general.source.huggingface.repository");
         if (keyidx != -1) { printf("%s: model source HF repo = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); }
     }
 

examples/llama-bench/llama-bench.cpp

@@ -903,7 +903,7 @@ static void test_gen(llama_context * ctx, int n_gen, int n_past, int n_threads)
     }
 }
 
-static void llama_null_log_callback(enum llama_log_level level, const char * text, void * user_data) {
+static void llama_null_log_callback(enum ggml_log_level level, const char * text, void * user_data) {
     (void) level;
     (void) text;
     (void) user_data;

examples/make-ggml.py

@@ -1,22 +1,25 @@
 #!/usr/bin/env python3
 """
-This script converts Hugging Face llama models to GGML and quantizes them.
+This script converts Hugging Face Llama, StarCoder, Falcon, Baichuan, and GPT-NeoX models to GGUF and quantizes them.
 
 Usage:
-python make-ggml.py --model {model_dir_or_hf_repo_name} [--outname {output_name} (Optional)] [--outdir {output_directory} (Optional)] [--quants {quant_types} (Optional)] [--keep_fp16 (Optional)]
+python make-ggml.py {model_dir_or_hf_repo_name} --model_type {model_type} [--outname {output_name} (Optional)] [--outdir {output_directory} (Optional)] [--quants {quant_types} (Optional)] [--keep_fp16 (Optional)]
 
 Arguments:
-- --model: (Required) The directory of the downloaded Hugging Face model or the name of the Hugging Face model repository. If the model directory does not exist, it will be downloaded from the Hugging Face model hub.
+- model: (Required) The directory of the downloaded Hugging Face model or the name of the Hugging Face model repository. If the model directory does not exist, it will be downloaded from the Hugging Face model hub.
+- --model_type: (Required) The type of the model to be converted. Choose from llama, starcoder, falcon, baichuan, or gptneox.
 - --outname: (Optional) The name of the output model. If not specified, the last part of the model directory path or the Hugging Face model repo name will be used.
 - --outdir: (Optional) The directory where the output model(s) will be stored. If not specified, '../models/{outname}' will be used.
 - --quants: (Optional) The types of quantization to apply. This should be a space-separated list. The default is 'Q4_K_M Q5_K_S'.
 - --keep_fp16: (Optional) If specified, the FP16 model will not be deleted after the quantized models are created.
 
-Quant types:
+Old quant types (some base model types require these):
 - Q4_0: small, very high quality loss - legacy, prefer using Q3_K_M
 - Q4_1: small, substantial quality loss - legacy, prefer using Q3_K_L
 - Q5_0: medium, balanced quality - legacy, prefer using Q4_K_M
 - Q5_1: medium, low quality loss - legacy, prefer using Q5_K_M
+
+New quant types (recommended):
 - Q2_K: smallest, extreme quality loss - not recommended
 - Q3_K: alias for Q3_K_M
 - Q3_K_S: very small, very high quality loss
@@ -40,9 +43,7 @@
 import os
 from huggingface_hub import snapshot_download
 
-def main(model, outname, outdir, quants, keep_fp16):
-    ggml_version = "v3"
-
+def main(model, model_type, outname, outdir, quants, keep_fp16):
     if not os.path.isdir(model):
         print(f"Model not found at {model}. Downloading...")
         try:
@@ -63,31 +64,35 @@ def main(model, outname, outdir, quants, keep_fp16):
     print("Building llama.cpp")
     subprocess.run(f"cd .. && make quantize", shell=True, check=True)
 
-    fp16 = f"{outdir}/{outname}.ggml{ggml_version}.fp16.bin"
+    fp16 = f"{outdir}/{outname}.gguf.fp16.bin"
 
-    print(f"Making unquantised GGML at {fp16}")
+    print(f"Making unquantised GGUF at {fp16}")
     if not os.path.isfile(fp16):
-        subprocess.run(f"python3 ../convert.py {model} --outtype f16 --outfile {fp16}", shell=True, check=True)
+        if model_type != "llama":
+            subprocess.run(f"python3 ../convert-{model_type}-hf-to-gguf.py {model} 1 --outfile {fp16}", shell=True, check=True)
+        else:
+            subprocess.run(f"python3 ../convert.py {model} --outtype f16 --outfile {fp16}", shell=True, check=True)
     else:
         print(f"Unquantised GGML already exists at: {fp16}")
 
     print("Making quants")
     for type in quants:
-        outfile = f"{outdir}/{outname}.ggml{ggml_version}.{type}.bin"
+        outfile = f"{outdir}/{outname}.gguf.{type}.bin"
         print(f"Making {type} : {outfile}")
         subprocess.run(f"../quantize {fp16} {outfile} {type}", shell=True, check=True)
 
     if not keep_fp16:
         os.remove(fp16)
 
 if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description='Convert/Quantize HF to GGML. If you have the HF model downloaded already, pass the path to the model dir. Otherwise, pass the Hugging Face model repo name. You need to be in the /examples folder for it to work.')
-    parser.add_argument('--model', required=True, help='Downloaded model dir or Hugging Face model repo name')
+    parser = argparse.ArgumentParser(description='Convert/Quantize HF models to GGUF. If you have the HF model downloaded already, pass the path to the model dir. Otherwise, pass the Hugging Face model repo name. You need to be in the /examples folder for it to work.')
+    parser.add_argument('model', help='Downloaded model dir or Hugging Face model repo name')
+    parser.add_argument('--model_type', required=True, choices=['llama', 'starcoder', 'falcon', 'baichuan', 'gptneox'], help='Type of the model to be converted. Choose from llama, starcoder, falcon, baichuan, or gptneox.')
     parser.add_argument('--outname', default=None, help='Output model(s) name')
     parser.add_argument('--outdir', default=None, help='Output directory')
     parser.add_argument('--quants', nargs='*', default=["Q4_K_M", "Q5_K_S"], help='Quant types')
     parser.add_argument('--keep_fp16', action='store_true', help='Keep fp16 model', default=False)
 
     args = parser.parse_args()
 
-    main(args.model, args.outname, args.outdir, args.quants, args.keep_fp16)
+    main(args.model, args.model_type, args.outname, args.outdir, args.quants, args.keep_fp16)

examples/perplexity/README.md

@@ -1,3 +1,21 @@
 # perplexity
 
 TODO
+
+## Llama 2 70B Scorechart
+Quantization | Model size (GiB) | Perplexity | Delta to fp16
+-- | -- | -- | --
+Q4_0 | 36.20 | 3.5550 | 3.61%
+Q4_1 | 40.20 | 3.5125 | 2.37%
+Q5_0 | 44.20 | 3.4744 | 1.26%
+Q2_K | 27.27 | 3.7339 | 8.82%
+Q3_K_S | 27.86 | 3.7019 | 7.89%
+Q3_K_M | 30.83 | 3.5932 | 4.72%
+Q3_K_L | 33.67 | 3.5617 | 3.80%
+Q4_K_S | 36.39 | 3.4852 | 1.57%
+Q4_K_M | 38.54 | 3.4725 | 1.20%
+Q5_K_S | 44.20 | 3.4483 | 0.50%
+Q5_K_M | 45.41 | 3.4451 | 0.40%
+Q6_K | 52.70 | 3.4367 | 0.16%
+fp16 | 128.5 | 3.4313 | -
+

examples/quantize/README.md

@@ -1,3 +1,44 @@
 # quantize
 
 TODO
+
+## Llama 2 7B
+
+Quantization | Bits per Weight (BPW)
+-- | --
+Q2_K | 3.35
+Q3_K_S | 3.50
+Q3_K_M | 3.91
+Q3_K_L | 4.27
+Q4_K_S | 4.58
+Q4_K_M | 4.84
+Q5_K_S | 5.52
+Q5_K_M | 5.68
+Q6_K | 6.56
+
+## Llama 2 13B
+Quantization | Bits per Weight (BPW)
+-- | --
+Q2_K | 3.34
+Q3_K_S | 3.48
+Q3_K_M | 3.89
+Q3_K_L | 4.26
+Q4_K_S | 4.56
+Q4_K_M | 4.83
+Q5_K_S | 5.51
+Q5_K_M | 5.67
+Q6_K | 6.56
+
+# Llama 2 70B
+
+Quantization | Bits per Weight (BPW)
+-- | --
+Q2_K | 3.40
+Q3_K_S | 3.47
+Q3_K_M | 3.85
+Q3_K_L | 4.19
+Q4_K_S | 4.53
+Q4_K_M | 4.80
+Q5_K_S | 5.50
+Q5_K_M | 5.65
+Q6_K | 6.56

flake.nix

@@ -35,6 +35,20 @@
         );
         pkgs = import nixpkgs { inherit system; };
         nativeBuildInputs = with pkgs; [ cmake ninja pkg-config ];
+        cudatoolkit_joined = with pkgs; symlinkJoin {
+          # HACK(Green-Sky): nix currently has issues with cmake findcudatoolkit
+          # see https://github.com/NixOS/nixpkgs/issues/224291
+          # copied from jaxlib
+          name = "${cudaPackages.cudatoolkit.name}-merged";
+          paths = [
+            cudaPackages.cudatoolkit.lib
+            cudaPackages.cudatoolkit.out
+          ] ++ lib.optionals (lib.versionOlder cudaPackages.cudatoolkit.version "11") [
+            # for some reason some of the required libs are in the targets/x86_64-linux
+            # directory; not sure why but this works around it
+            "${cudaPackages.cudatoolkit}/targets/${system}"
+          ];
+        };
         llama-python =
           pkgs.python3.withPackages (ps: with ps; [ numpy sentencepiece ]);
         postPatch = ''
@@ -70,6 +84,13 @@
             "-DLLAMA_CLBLAST=ON"
           ];
         };
+        packages.cuda = pkgs.stdenv.mkDerivation {
+          inherit name src meta postPatch nativeBuildInputs postInstall;
+          buildInputs = with pkgs; buildInputs ++ [ cudatoolkit_joined ];
+          cmakeFlags = cmakeFlags ++ [
+            "-DLLAMA_CUBLAS=ON"
+          ];
+        };
         packages.rocm = pkgs.stdenv.mkDerivation {
           inherit name src meta postPatch nativeBuildInputs postInstall;
           buildInputs = with pkgs; buildInputs ++ [ hip hipblas rocblas ];

ggml-metal.h

@@ -19,6 +19,8 @@
 
 #pragma once
 
+#include "ggml.h"
+
 #include <stddef.h>
 #include <stdbool.h>
 
@@ -33,6 +35,8 @@ struct ggml_cgraph;
 extern "C" {
 #endif
 
+void ggml_metal_log_set_callback(ggml_log_callback log_callback, void * user_data);
+
 struct ggml_metal_context;
 
 // number of command buffers to use