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

Merge branch 'ggerganov:master' into master

Browse source
This commit is contained in:
R.Kaufmann 2023-03-27 21:11:08 +02:00 • committed by GitHub
commit 926e49e78e
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
10 changed files with 34 additions and 39 deletions
Download patch file Download diff file Expand all files Collapse all files

6
.github/workflows/build.yml vendored
View file

@ -72,6 +72,8 @@ jobs:
strategy:
matrix:
sanitizer: [ADDRESS, THREAD, UNDEFINED]
build_type: [Debug, Release]
accelerate: [ON, OFF]
steps:
- name: Clone
@ -89,8 +91,8 @@ jobs:
run: |
mkdir build
cd build
cmake .. -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON
cmake --build . --config Release
cmake .. -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DLLAMA_ACCELERATE=${{ matrix.accelerate }}
cmake --build . --config ${{ matrix.build_type }}
- name: Test
id: cmake_test

2
.github/workflows/docker.yml vendored
View file

@ -49,6 +49,7 @@ jobs:
with:
context: .
push: true
platforms: linux/amd64,linux/arm64
tags: "ghcr.io/ggerganov/llama.cpp:${{ matrix.config.tag }}-${{ env.COMMIT_SHA }}"
file: ${{ matrix.config.dockerfile }}
@ -57,5 +58,6 @@ jobs:
with:
context: .
push: ${{ github.event_name == 'push' }}
platforms: linux/amd64,linux/arm64
tags: "ghcr.io/ggerganov/llama.cpp:${{ matrix.config.tag }}"
file: ${{ matrix.config.dockerfile }}

6
README.md
View file

@ -248,7 +248,7 @@ cadaver, cauliflower, cabbage (vegetable), catalpa (tree) and Cailleach.
### Perplexity (Measuring model quality)
You can pass `--perplexity` as a command line option to measure perplexity over the given prompt. For more background,
You can use the `perplexity` example to measure perplexity over the given prompt. For more background,
see https://huggingface.co/docs/transformers/perplexity. However, in general, lower perplexity is better for LLMs.
#### Latest measurements
@ -271,10 +271,10 @@ Perplexity - model options
#### How to run
1. Download/extract: https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-2-raw-v1.zip?ref=salesforce-research
2. Run `./main --perplexity -m models/7B/ggml-model-q4_0.bin -f wiki.test.raw`
2. Run `./perplexity -m models/7B/ggml-model-q4_0.bin -f wiki.test.raw`
3. Output:
```
Calculating perplexity over 655 chunks
perplexity : calculating perplexity over 655 chunks
24.43 seconds per pass - ETA 4.45 hours
[1]4.5970,[2]5.1807,[3]6.0382,...
```

2
examples/embedding/CMakeLists.txt
View file

@ -1,4 +1,4 @@
set(TARGET embedding)
add_executable(${TARGET} embedding.cpp)
target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
target_link_libraries(${TARGET} PRIVATE common llama ggml ${CMAKE_THREAD_LIBS_INIT})
target_compile_features(${TARGET} PRIVATE cxx_std_11)

2
examples/main/CMakeLists.txt
View file

@ -1,4 +1,4 @@
set(TARGET main)
add_executable(${TARGET} main.cpp)
target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
target_link_libraries(${TARGET} PRIVATE common llama ggml ${CMAKE_THREAD_LIBS_INIT})
target_compile_features(${TARGET} PRIVATE cxx_std_11)

2
examples/main/main.cpp
View file

@ -493,7 +493,7 @@ int main(int argc, char ** argv) {
}
// In interactive mode, respect the maximum number of tokens and drop back to user input when reached.
if (params.interactive && n_remain <= 0) {
if (params.interactive && n_remain <= 0 && params.n_predict != -1) {
n_remain = params.n_predict;
is_interacting = true;
}

2
examples/perplexity/CMakeLists.txt
View file

@ -1,4 +1,4 @@
set(TARGET perplexity)
add_executable(${TARGET} perplexity.cpp)
target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
target_link_libraries(${TARGET} PRIVATE common llama ggml ${CMAKE_THREAD_LIBS_INIT})
target_compile_features(${TARGET} PRIVATE cxx_std_11)

2
examples/perplexity/perplexity.cpp
View file

@ -19,7 +19,7 @@ std::vector<double> softmax(const std::vector<float>& logits) {
void perplexity(llama_context * ctx, const gpt_params & params) {
// Download: https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-2-raw-v1.zip?ref=salesforce-research
// Run `./main --perplexity -m models/7B/ggml-model-q4_0.bin -f wiki.test.raw`
// Run `./perplexity -m models/7B/ggml-model-q4_0.bin -f wiki.test.raw`
// Output: `perplexity: 13.5106 [114/114]`
auto tokens = ::llama_tokenize(ctx, params.prompt, true);

2
examples/quantize/CMakeLists.txt
View file

@ -1,4 +1,4 @@
set(TARGET quantize)
add_executable(${TARGET} quantize.cpp)
target_link_libraries(${TARGET} PRIVATE llama ${CMAKE_THREAD_LIBS_INIT})
target_link_libraries(${TARGET} PRIVATE llama ggml ${CMAKE_THREAD_LIBS_INIT})
target_compile_features(${TARGET} PRIVATE cxx_std_11)

47
ggml.c
View file

@ -1698,8 +1698,6 @@ inline static void ggml_vec_dot_q4_0(const int n, float * restrict s, const void
// Horizontal sum of all lanes of the accumulator
sumf = _mm512_reduce_add_ps( acc0 ) + _mm512_reduce_add_ps( acc1 );
#elif defined(__AVX2__)
const size_t countBlocks = nb;
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
@ -5806,23 +5804,28 @@ static void ggml_compute_forward_mul_mat_f32(
const int ne02 = src0->ne[2];
const int ne03 = src0->ne[3];
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
const int ne10 = src1->ne[0];
#endif
const int ne11 = src1->ne[1];
//const int ne12 = src1->ne[2];
//const int ne13 = src1->ne[3];
#ifndef NDEBUG
const int ne12 = src1->ne[2];
const int ne13 = src1->ne[3];
//const int ne0 = dst->ne[0];
//const int ne1 = dst->ne[1];
//const int ne2 = dst->ne[2];
//const int ne3 = dst->ne[3];
//const int ne = ne0*ne1*ne2*ne3;
const int ne0 = dst->ne[0];
const int ne1 = dst->ne[1];
const int ne2 = dst->ne[2];
const int ne3 = dst->ne[3];
//const int nb00 = src0->nb[0];
const int nb00 = src0->nb[0];
#endif
const int nb01 = src0->nb[1];
const int nb02 = src0->nb[2];
const int nb03 = src0->nb[3];
#ifndef NDEBUG
const int nb10 = src1->nb[0];
#endif
const int nb11 = src1->nb[1];
const int nb12 = src1->nb[2];
const int nb13 = src1->nb[3];
@ -5840,8 +5843,9 @@ static void ggml_compute_forward_mul_mat_f32(
assert(ne2 == ne12);
assert(ne3 == ne13);
// TODO: we don't support permuted src0
// we don't support permuted src0 or src1
assert(nb00 == sizeof(float));
assert(nb10 == sizeof(float));
// dst cannot be transposed or permuted
assert(nb0 == sizeof(float));
@ -5859,8 +5863,6 @@ static void ggml_compute_forward_mul_mat_f32(
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
if (ggml_compute_forward_mul_mat_use_blas(src0, src1, dst)) {
GGML_ASSERT(nb10 == sizeof(float));
if (params->ith != 0) {
return;
}
@ -5903,9 +5905,6 @@ static void ggml_compute_forward_mul_mat_f32(
return;
}
// TODO: do not support transposed src1
assert(nb10 == sizeof(float));
// parallelize by src0 rows using ggml_vec_dot_f32
// total rows in src0
@ -6169,7 +6168,6 @@ static void ggml_compute_forward_mul_mat_q4_0_f32(
const int ne1 = dst->ne[1];
const int ne2 = dst->ne[2];
const int ne3 = dst->ne[3];
//const int ne = ne0*ne1*ne2*ne3;
const int nb00 = src0->nb[0];
const int nb01 = src0->nb[1];
@ -6194,8 +6192,9 @@ static void ggml_compute_forward_mul_mat_q4_0_f32(
GGML_ASSERT(ne2 == ne12);
GGML_ASSERT(ne3 == ne13);
// TODO: we don't support permuted src0
// we don't support permuted src0 or src1
GGML_ASSERT(nb00 == (int) GGML_TYPE_SIZE[GGML_TYPE_Q4_0]);
GGML_ASSERT(nb10 == sizeof(float));
// dst cannot be transposed or permuted
GGML_ASSERT(nb0 == sizeof(float));
@ -6213,8 +6212,6 @@ static void ggml_compute_forward_mul_mat_q4_0_f32(
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
if (ggml_compute_forward_mul_mat_use_blas(src0, src1, dst)) {
GGML_ASSERT(nb10 == sizeof(float));
if (params->ith != 0) {
return;
}
@ -6278,8 +6275,6 @@ static void ggml_compute_forward_mul_mat_q4_0_f32(
return;
}
// TODO: do not support transposed src1
// parallelize by src0 rows using ggml_vec_dot_q4_0
// total rows in src0
@ -6354,7 +6349,6 @@ static void ggml_compute_forward_mul_mat_q4_1_f32(
const int ne1 = dst->ne[1];
const int ne2 = dst->ne[2];
const int ne3 = dst->ne[3];
//const int ne = ne0*ne1*ne2*ne3;
const int nb00 = src0->nb[0];
const int nb01 = src0->nb[1];
@ -6379,8 +6373,9 @@ static void ggml_compute_forward_mul_mat_q4_1_f32(
GGML_ASSERT(ne2 == ne12);
GGML_ASSERT(ne3 == ne13);
// TODO: we don't support permuted src0
// we don't support permuted src0 or src1
GGML_ASSERT(nb00 == (int) GGML_TYPE_SIZE[GGML_TYPE_Q4_1]);
GGML_ASSERT(nb10 == sizeof(float));
// dst cannot be transposed or permuted
GGML_ASSERT(nb0 == sizeof(float));
@ -6398,8 +6393,6 @@ static void ggml_compute_forward_mul_mat_q4_1_f32(
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
if (ggml_compute_forward_mul_mat_use_blas(src0, src1, dst)) {
GGML_ASSERT(nb10 == sizeof(float));
if (params->ith != 0) {
return;
}
@ -6466,8 +6459,6 @@ static void ggml_compute_forward_mul_mat_q4_1_f32(
return;
}
// TODO: do not support transposed src1
// parallelize by src0 rows using ggml_vec_dot_q4_1
// total rows in src0