Merge 'origin/master' into hipblas

LostRuins · Apr 28, 2023 · 3b4a531 · 3b4a531
2 parents a1caa48 + 0b2da20
commit 3b4a531
Show file tree

Hide file tree

Showing 10 changed files with 862 additions and 56 deletions.
diff --git a/.devops/tools.sh b/.devops/tools.sh
@@ -23,7 +23,7 @@ elif [[ $arg1 == '--all-in-one' || $arg1 == '-a' ]]; then
             echo "Skip model quantization, it already exists: ${i/f16/q4_0}"
         else
             echo "Converting PTH to GGML: $i into ${i/f16/q4_0}..."
-            ./quantize "$i" "${i/f16/q4_0}" 2
+            ./quantize "$i" "${i/f16/q4_0}" q4_0
         fi
     done
 else

diff --git a/.gitignore b/.gitignore
@@ -15,6 +15,7 @@ build-em/
 build-debug/
 build-release/
 build-static/
+build-cublas/
 build-no-accel/
 build-sanitize-addr/
 build-sanitize-thread/

diff --git a/README.md b/README.md
@@ -7,31 +7,27 @@
 
 Inference of [LLaMA](https://arxiv.org/abs/2302.13971) model in pure C/C++
 
-**Warnings**
-
-- `Q4_2` and `Q4_3` are still in development. Do not expect any kind of backward compatibility until they are finalized
-
 **Hot topics:**
 
+- [New quantization methods](https://github.com/ggerganov/llama.cpp#quantization)
 - [Added LoRA support](https://github.com/ggerganov/llama.cpp/pull/820)
 - [Add GPU support to ggml](https://github.com/ggerganov/llama.cpp/discussions/915)
 - [Roadmap Apr 2023](https://github.com/ggerganov/llama.cpp/discussions/784)
 
 ## Description
 
-The main goal of llama.cpp is to run the llama model using 4-bit quantization on a MacBook.
+The main goal of `llama.cpp` is to run the LLaMA model using 4-bit integer quantization on a MacBook
 
 - Plain C/C++ implementation without dependencies
 - Apple silicon first-class citizen - optimized via ARM NEON and Accelerate framework
 - AVX2 support for x86 architectures
 - Mixed F16 / F32 precision
-- 4-bit quantization support
+- 4-bit integer quantization support
 - Runs on the CPU
 
-This was [hacked in an evening](https://github.com/ggerganov/llama.cpp/issues/33#issuecomment-1465108022) - I have no idea if it works correctly.
-Please do not make conclusions about the models based on the results from this implementation.
-For all I know, it can be completely wrong. This project is for educational purposes.
-New features will probably be added mostly through community contributions.
+The original implementation of `llama.cpp` was [hacked in an evening](https://github.com/ggerganov/llama.cpp/issues/33#issuecomment-1465108022).
+Since then, the project has improved significantly thanks to many contributions. This project is for educational purposes and serves
+as the main playground for developing new features for the [ggml](https://github.com/ggerganov/ggml) library.
 
 **Supported platforms:**
 
@@ -167,15 +163,27 @@ cd llama.cpp
 
 ### Build
 
-Note: For Windows, CMake or Zig can be used.
+In order to build llama.cpp you have three different options.
 
-1. Use `make`
+- Using `make`:
+  - On Linux or MacOS:
 
-    ```bash
-    make
-    ```
+      ```bash
+      make
+      ```
 
-1. Use CMake
+  - On Windows:
+
+    1. Download the latest fortran version of [w64devkit](https://github.com/seeto/w64devkit/releases).
+    2. Extract `w64devkit` on your pc.
+    3. Run `w64devkit.exe`.
+    4. Use the `cd` command to reach the `llama.cpp` folder.
+    5. From here you can run:
+        ```bash
+        make
+        ```
+
+- Using `CMake`:
 
     ```bash
     mkdir build
@@ -184,12 +192,71 @@ Note: For Windows, CMake or Zig can be used.
     cmake --build . --config Release
     ```
 
-1. Use Zig
+- Using `Zig`:
 
     ```bash
     zig build -Drelease-fast
     ```
 
+### BLAS Build
+
+Building the program with BLAS support may lead to some performance improvements in prompt processing using batch sizes higher than 32 (the default is 512). BLAS doesn't affect the normal generation performance. There are currently three different implementations of it:
+
+- Accelerate Framework:
+
+  This is only available on Mac PCs and it's enabled by default. You can just build using the normal instructions.
+
+- OpenBLAS:
+
+  This provides BLAS acceleration using only the CPU. Make sure to have OpenBLAS installed on your machine.
+
+  - Using `make`:
+    - On Linux:
+      ```bash
+      make LLAMA_OPENBLAS=1
+      ```
+      Note: In order to build on Arch Linux with OpenBLAS support enabled you must edit the Makefile adding at the end of the line 105: `-lcblas`
+
+    - On Windows:
+
+      1. Download the latest fortran version of [w64devkit](https://github.com/skeeto/w64devkit/releases).
+      2. Download the latest version of [OpenBLAS for Windows](https://github.com/xianyi/OpenBLAS/releases).
+      3. Extract `w64devkit` on your pc.
+      4. From the OpenBLAS zip that you just downloaded copy `libopenblas.a`, located inside the `lib` folder, inside `w64devkit\x86_64-w64-mingw32\lib`.
+      5. From the same OpenBLAS zip copy the content of the `include` folder inside `w64devkit\x86_64-w64-mingw32\include`.
+      6. Run `w64devkit.exe`.
+      7. Use the `cd` command to reach the `llama.cpp` folder.
+      8. From here you can run:
+
+          ```bash
+          make LLAMA_OPENBLAS=1
+          ```
+
+  - Using `CMake` on Linux:
+
+      ```bash
+      mkdir build
+      cd build
+      cmake .. -DLLAMA_OPENBLAS=ON
+      cmake --build . --config Release
+      ```
+
+- cuBLAS
+
+  This provides BLAS acceleration using the CUDA cores of your Nvidia GPU. Make sure to have the CUDA toolkit installed. You can download it from your Linux distro's package manager or from here: [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads).
+  - Using `make`:
+    ```bash
+    make LLAMA_CUBLAS=1
+    ```
+  - Using `CMake`:
+
+    ```bash
+    mkdir build
+    cd build
+    cmake .. -DLLAMA_CUBLAS=ON
+    cmake --build . --config Release
+    ```
+
 ### Prepare Data & Run
 
 ```bash
@@ -203,8 +270,8 @@ python3 -m pip install -r requirements.txt
 # convert the 7B model to ggml FP16 format
 python3 convert.py models/7B/
 
-# quantize the model to 4-bits (using method 2 = q4_0)
-./quantize ./models/7B/ggml-model-f16.bin ./models/7B/ggml-model-q4_0.bin 2
+# quantize the model to 4-bits (using q4_0 method)
+./quantize ./models/7B/ggml-model-f16.bin ./models/7B/ggml-model-q4_0.bin q4_0
 
 # run the inference
 ./main -m ./models/7B/ggml-model-q4_0.bin -n 128
@@ -223,6 +290,24 @@ As the models are currently fully loaded into memory, you will need adequate dis
 | 30B   | 60 GB         | 19.5 GB                |
 | 65B   | 120 GB        | 38.5 GB                |
 
+### Quantization
+
+Several quantization methods are supported. They differ in the resulting model disk size and inference speed.
+
+Model | F16 | Q4_0 | Q4_1 | Q4_2 | Q4_3 | Q5_0 | Q5_1 | Q8_0
+-- | -- | -- | -- | -- | -- | -- | -- | --
+7B (ppl) | 5.9565 | 6.2103 | 6.1286 | 6.1698 | 6.0617 | 6.0139 | 5.9934 | 5.9571
+7B (size) | 13.0G | 4.0G | 4.8G | 4.0G | 4.8G | 4.4G | 4.8G | 7.1G
+7B (ms/tok @ 4th) | 128 | 56 | 61 | 84 | 91 | 91 | 95 | 75
+7B (ms/tok @ 8th) | 128 | 47 | 55 | 48 | 53 | 53 | 59 | 75
+7B (bpw) | 16.0 | 5.0 | 6.0 | 5.0 | 6.0 | 5.5 | 6.0 | 9.0
+-- | -- | -- | -- | -- | -- | -- | -- | --
+13B (ppl) | 5.2455 | 5.3748 | 5.3471 | 5.3433 | 5.3234 | 5.2768 | 5.2582 | 5.2458
+13B (size) | 25.0G | 7.6G | 9.1G | 7.6G | 9.1G | 8.4G | 9.1G | 14G
+13B (ms/tok @ 4th) | 239 | 104 | 113 | 160 | 175 | 176 | 185 | 141
+13B (ms/tok @ 8th) | 240 | 85 | 99 | 97 | 114 | 108 | 117 | 147
+13B (bpw) | 16.0 | 5.0 | 6.0 | 5.0 | 6.0 | 5.5 | 6.0 | 9.0
+
 ### Interactive mode
 
 If you want a more ChatGPT-like experience, you can run in interactive mode by passing `-i` as a parameter.

diff --git a/examples/quantize/quantize.cpp b/examples/quantize/quantize.cpp
@@ -2,8 +2,19 @@
 #include "llama.h"
 
 #include <cstdio>
+#include <map>
 #include <string>
 
+static const std::map<std::string, enum llama_ftype> LLAMA_FTYPE_MAP = {
+  {"q4_0", LLAMA_FTYPE_MOSTLY_Q4_0},
+  {"q4_1", LLAMA_FTYPE_MOSTLY_Q4_1},
+  {"q4_2", LLAMA_FTYPE_MOSTLY_Q4_2},
+  {"q4_3", LLAMA_FTYPE_MOSTLY_Q4_3},
+  {"q5_0", LLAMA_FTYPE_MOSTLY_Q5_0},
+  {"q5_1", LLAMA_FTYPE_MOSTLY_Q5_1},
+  {"q8_0", LLAMA_FTYPE_MOSTLY_Q8_0},
+};
+
 // usage:
 //  ./quantize models/llama/ggml-model.bin models/llama/ggml-model-quant.bin type
 //
@@ -12,11 +23,9 @@ int main(int argc, char ** argv) {
 
     if (argc < 4) {
         fprintf(stderr, "usage: %s model-f32.bin model-quant.bin type [nthread]\n", argv[0]);
-        fprintf(stderr, "  type = %d - q4_0\n", LLAMA_FTYPE_MOSTLY_Q4_0);
-        fprintf(stderr, "  type = %d - q4_1\n", LLAMA_FTYPE_MOSTLY_Q4_1);
-        fprintf(stderr, "  type = %d - q4_2\n", LLAMA_FTYPE_MOSTLY_Q4_2);
-        fprintf(stderr, "  type = %d - q4_3\n", LLAMA_FTYPE_MOSTLY_Q4_3);
-        fprintf(stderr, "  type = %d - q8_0\n", LLAMA_FTYPE_MOSTLY_Q8_0);
+        for (auto it = LLAMA_FTYPE_MAP.begin(); it != LLAMA_FTYPE_MAP.end(); it++) {
+            fprintf(stderr, "  type = \"%s\" or %d\n", it->first.c_str(), it->second);
+        }
         return 1;
     }
 
@@ -30,7 +39,18 @@ int main(int argc, char ** argv) {
     const std::string fname_inp = argv[1];
     const std::string fname_out = argv[2];
 
-    const enum llama_ftype ftype = (enum llama_ftype)atoi(argv[3]);
+    enum llama_ftype ftype;
+    if (argv[3][0] == 'q') {
+        auto it = LLAMA_FTYPE_MAP.find(argv[3]);
+        if (it == LLAMA_FTYPE_MAP.end()) {
+            fprintf(stderr, "%s: unknown ftype '%s'\n", __func__, argv[3]);
+            return 1;
+        }
+        ftype = it->second;
+    } else {
+        ftype = (enum llama_ftype)atoi(argv[3]);
+    }
+
     int nthread = argc > 4 ? atoi(argv[4]) : 0;
 
     const int64_t t_main_start_us = ggml_time_us();

diff --git a/ggml-cuda.cu b/ggml-cuda.cu
@@ -41,6 +41,23 @@ typedef struct {
 } block_q4_3;
 static_assert(sizeof(block_q4_3) == 2 * sizeof(ggml_fp16_t) + QK4_3 / 2, "wrong q4_3 block size/padding");
 
+#define QK5_0 32
+typedef struct {
+    __half d;               // delta
+    uint8_t qh[4];          // 5-th bit of quants
+    uint8_t qs[QK5_0 / 2];  // nibbles / quants
+} block_q5_0;
+static_assert(sizeof(block_q5_0) == sizeof(ggml_fp16_t) + sizeof(uint32_t) + QK5_0 / 2, "wrong q5_0 block size/padding");
+
+#define QK5_1 32
+typedef struct {
+    __half d;               // delta
+    __half m;               // min
+    uint32_t qh;            // 5-th bit of quants
+    uint8_t qs[QK5_1 / 2];  // nibbles / quants
+} block_q5_1;
+static_assert(sizeof(block_q5_1) == 2 * sizeof(ggml_fp16_t) + sizeof(uint32_t) + QK5_1 / 2, "wrong q5_1 block size/padding");
+
 #define QK8_0 32
 typedef struct {
     float   d;              // delta
@@ -142,6 +159,64 @@ static __global__ void dequantize_block_q4_3(const void * vx, float * y) {
     }
 }
 
+static __global__ void dequantize_block_q5_0(const void * vx, float * y) {
+    const block_q5_0 * x = (const block_q5_0 *) vx;
+
+    const int i = blockIdx.x;
+
+    const float d = x[i].d;
+
+    const uint8_t * pp = x[i].qs;
+
+    uint32_t qh;
+    memcpy(&qh, x[i].qh, sizeof(qh));
+
+    for (int l = 0; l < QK5_0; l += 2) {
+        const uint8_t vi = pp[l/2];
+
+        const int8_t vh0 = ((qh & (1 << (l + 0))) >> (l + 0)) << 4;
+        const int8_t vh1 = ((qh & (1 << (l + 1))) >> (l + 1)) << 4;
+
+        const int8_t vi0 = ((vi & 0xf) | vh0);
+        const int8_t vi1 = ((vi >>  4) | vh1);
+
+        const float v0 = (vi0 - 16)*d;
+        const float v1 = (vi1 - 16)*d;
+
+        y[i*QK5_0 + l + 0] = v0;
+        y[i*QK5_0 + l + 1] = v1;
+    }
+}
+
+static __global__ void dequantize_block_q5_1(const void * vx, float * y) {
+    const block_q5_1 * x = (const block_q5_1 *) vx;
+
+    const int i = blockIdx.x;
+
+    const float d = x[i].d;
+    const float m = x[i].m;
+
+    const uint8_t * pp = x[i].qs;
+
+    const uint32_t qh = x[i].qh;
+
+    for (int l = 0; l < QK5_1; l += 2) {
+        const uint8_t vi = pp[l/2];
+
+        const int8_t vh0 = ((qh & (1 << (l + 0))) >> (l + 0)) << 4;
+        const int8_t vh1 = ((qh & (1 << (l + 1))) >> (l + 1)) << 4;
+
+        const int8_t vi0 = (vi & 0xf) | vh0;
+        const int8_t vi1 = (vi >>  4) | vh1;
+
+        const float v0 = vi0*d + m;
+        const float v1 = vi1*d + m;
+
+        y[i*QK5_1 + l + 0] = v0;
+        y[i*QK5_1 + l + 1] = v1;
+    }
+}
+
 static __global__ void dequantize_block_q8_0(const void * vx, float * y) {
     const block_q8_0 * x = (const block_q8_0 *) vx;
 
@@ -178,6 +253,16 @@ void dequantize_row_q4_3_cuda(const void * vx, float * y, int k, cudaStream_t st
     dequantize_block_q4_3<<<nb, 1, 0, stream>>>(vx, y);
 }
 
+void dequantize_row_q5_0_cuda(const void * vx, float * y, int k, cudaStream_t stream) {
+    const int nb = k / QK5_0;
+    dequantize_block_q5_0<<<nb, 1, 0, stream>>>(vx, y);
+}
+
+void dequantize_row_q5_1_cuda(const void * vx, float * y, int k, cudaStream_t stream) {
+    const int nb = k / QK5_1;
+    dequantize_block_q5_1<<<nb, 1, 0, stream>>>(vx, y);
+}
+
 void dequantize_row_q8_0_cuda(const void * vx, float * y, int k, cudaStream_t stream) {
     const int nb = k / QK8_0;
     dequantize_block_q8_0<<<nb, 1, 0, stream>>>(vx, y);

diff --git a/ggml-cuda.h b/ggml-cuda.h
@@ -77,6 +77,8 @@ void dequantize_row_q4_0_cuda(const void * vx, float * y, int k, cudaStream_t st
 void dequantize_row_q4_1_cuda(const void * vx, float * y, int k, cudaStream_t stream);
 void dequantize_row_q4_2_cuda(const void * vx, float * y, int k, cudaStream_t stream);
 void dequantize_row_q4_3_cuda(const void * vx, float * y, int k, cudaStream_t stream);
+void dequantize_row_q5_0_cuda(const void * vx, float * y, int k, cudaStream_t stream);
+void dequantize_row_q5_1_cuda(const void * vx, float * y, int k, cudaStream_t stream);
 void dequantize_row_q8_0_cuda(const void * vx, float * y, int k, cudaStream_t stream);
 
 #ifdef  __cplusplus