mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 22:33:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama: update vendored code to commit 40c6d79f (#7875)

Browse Source
This commit is contained in:
Jeffrey Morgan
2024-12-10 19:21:34 -08:00
committed by GitHub
parent a37f4a86a7
commit 527cc97899
289 changed files with 58552 additions and 41806 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
llama/llama.go
View File

@@ -3,12 +3,12 @@ package llama
//go:generate make -j 8
/*
#cgo CFLAGS: -O2 -std=c11 -DGGML_BUILD=1 -DNDEBUG -DLOG_DISABLE_LOGS -DGGML_USE_LLAMAFILE
#cgo CXXFLAGS: -O2 -std=c++11 -DGGML_BUILD=1 -DNDEBUG -DLOG_DISABLE_LOGS -DGGML_USE_LLAMAFILE
#cgo CFLAGS: -O3 -std=c17 -DGGML_BUILD=1 -DNDEBUG -DLOG_DISABLE_LOGS -DGGML_USE_LLAMAFILE -DGGML_USE_CPU -DGGML_USE_CPU_AARCH64
#cgo CXXFLAGS: -O3 -std=c++17 -DGGML_BUILD=1 -DNDEBUG -DLOG_DISABLE_LOGS -DGGML_USE_LLAMAFILE -DGGML_USE_CPU -DGGML_USE_CPU_AARCH64
#cgo amd64,avx CFLAGS: -mavx
#cgo amd64,avx CXXFLAGS: -mavx
#cgo amd64,avx2 CFLAGS: -mavx2 -mfma
#cgo amd64,avx2 CXXFLAGS: -mavx2 -mfma
#cgo amd64,avx2 CFLAGS: -mavx2 -mfma -mf16c
#cgo amd64,avx2 CXXFLAGS: -mavx2 -mfma -mf16c
#cgo amd64,avx512 CFLAGS: -mavx512f -mavx512dq -mavx512bw
#cgo amd64,avx512 CXXFLAGS: -mavx512f -mavx512dq -mavx512bw
#cgo amd64,avx512bf16 CFLAGS: -mavx512bf16 -D__AVX512BF16__
@@ -33,21 +33,22 @@ package llama
#cgo darwin,amd64,avx2 CFLAGS: -DGGML_USE_ACCELERATE -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64
#cgo darwin,amd64,avx2 CXXFLAGS: -DGGML_USE_ACCELERATE -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64
#cgo darwin,amd64,avx2 LDFLAGS: -framework Accelerate
#cgo darwin,arm64 CFLAGS: -DGGML_USE_METAL -DGGML_USE_ACCELERATE -DGGML_METAL_EMBED_LIBRARY -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64 -DGGML_USE_BLAS
#cgo darwin,arm64 CXXFLAGS: -DGGML_USE_METAL -DGGML_USE_ACCELERATE -DGGML_METAL_EMBED_LIBRARY -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64 -DGGML_USE_BLAS
#cgo darwin,arm64 CFLAGS: -DGGML_USE_METAL -DGGML_USE_ACCELERATE -DGGML_METAL_EMBED_LIBRARY -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64 -DGGML_USE_BLAS -DGGML_BLAS_USE_ACCELERATE
#cgo darwin,arm64 CXXFLAGS: -DGGML_USE_METAL -DGGML_USE_ACCELERATE -DGGML_METAL_EMBED_LIBRARY -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64 -DGGML_USE_BLAS -DGGML_BLAS_USE_ACCELERATE
#cgo darwin,arm64 LDFLAGS: -framework Foundation -framework Metal -framework MetalKit -framework Accelerate
#cgo linux CFLAGS: -D_GNU_SOURCE
#cgo linux CXXFLAGS: -D_GNU_SOURCE
#cgo linux LDFLAGS: -ldl
#cgo linux,amd64 LDFLAGS: -L${SRCDIR}/build/linux-amd64
#cgo linux,arm64 CFLAGS: -D__aarch64__ -D__ARM_NEON -D__ARM_FEATURE_FMA
#cgo linux,arm64 CXXFLAGS: -D__aarch64__ -D__ARM_NEON -D__ARM_FEATURE_FMA
#cgo linux,arm64 LDFLAGS: -L${SRCDIR}/build/linux-arm64
#cgo linux,arm64,sve CFLAGS: -march=armv8.6-a+sve
#cgo linux,arm64,sve CXXFLAGS: -march=armv8.6-a+sve
#cgo linux,cuda LDFLAGS: -lcuda -lcudart -lcublas -lcublasLt -lpthread -ldl -lrt -lresolv
#cgo linux,rocm LDFLAGS: -lpthread -ldl -lrt -lresolv
#cgo rocm CFLAGS: -DGGML_USE_CUDA -DGGML_USE_HIPBLAS -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1
#cgo rocm CXXFLAGS: -DGGML_USE_CUDA -DGGML_USE_HIPBLAS -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1
#cgo linux,cuda LDFLAGS: -lcuda -lcudart -lcublas -lcublasLt -lpthread -lrt -lresolv
#cgo linux,rocm LDFLAGS: -lpthread -lrt -lresolv
#cgo rocm CFLAGS: -DGGML_USE_CUDA -DGGML_USE_HIP -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1
#cgo rocm CXXFLAGS: -DGGML_USE_CUDA -DGGML_USE_HIP -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1
#cgo rocm LDFLAGS: -L${SRCDIR} -lggml_rocm -lhipblas -lamdhip64 -lrocblas
#cgo windows CFLAGS: -Wno-discarded-qualifiers -D_WIN32_WINNT=0x602
#cgo windows CXXFLAGS: -D_WIN32_WINNT=0x602
@@ -67,7 +68,8 @@ package llama
#include "mllama.h"
#include "sampling_ext.h"
bool llamaProgressCallback(float progress, void *user_data);
extern bool llamaProgressCallback(float progress, void *user_data);
extern void llamaLog(int level, char* text, void* user_data);
typedef enum {COMP_UNKNOWN,COMP_GCC,COMP_CLANG} COMPILER;
COMPILER inline get_compiler() {
@@ -79,6 +81,7 @@ COMPILER inline get_compiler() {
return UNKNOWN_COMPILER;
#endif
}
*/
import "C"
@@ -93,6 +96,7 @@ import (
"runtime/cgo"
"slices"
"strings"
"sync/atomic"
"unsafe"
)
@@ -113,6 +117,26 @@ func PrintSystemInfo() string {
return C.GoString(C.llama_print_system_info()) + compiler
}
var logLevel atomic.Int32
func init() {
logLevel.Store(int32(C.GGML_LOG_LEVEL_INFO))
C.llama_log_set((C.ggml_log_callback)(C.llamaLog), nil)
}
func EnableDebug() {
logLevel.Store(int32(C.GGML_LOG_LEVEL_DEBUG))
}
//export llamaLog
func llamaLog(level int32, text *C.char, _ unsafe.Pointer) {
if level < logLevel.Load() {
return
}
fmt.Print(C.GoString(text))
}
func GetModelArch(modelPath string) (string, error) {
mp := C.CString(modelPath)
defer C.free(unsafe.Pointer(mp))
@@ -633,14 +657,13 @@ func (c *Context) Synchronize() {
// sampling
// TODO: this is a temporary wrapper to allow calling C++ code from CGo
type SamplingContext struct {
c *C.struct_gpt_sampler
c *C.struct_common_sampler
}
type SamplingParams struct {
TopK int
TopP float32
MinP float32
TfsZ float32
TypicalP float32
Temp float32
RepeatLastN int
@@ -656,11 +679,10 @@ type SamplingParams struct {
}
func NewSamplingContext(model *Model, params SamplingParams) (*SamplingContext, error) {
var cparams C.struct_gpt_sampler_cparams
var cparams C.struct_common_sampler_cparams
cparams.top_k = C.int32_t(params.TopK)
cparams.top_p = C.float(params.TopP)
cparams.min_p = C.float(params.MinP)
cparams.tfs_z = C.float(params.TfsZ)
cparams.typical_p = C.float(params.TypicalP)
cparams.temp = C.float(params.Temp)
cparams.penalty_last_n = C.int32_t(params.RepeatLastN)
@@ -677,26 +699,26 @@ func NewSamplingContext(model *Model, params SamplingParams) (*SamplingContext,
defer C.free(unsafe.Pointer(grammar))
cparams.grammar = grammar
context := &SamplingContext{c: C.gpt_sampler_cinit(model.c, &cparams)}
context := &SamplingContext{c: C.common_sampler_cinit(model.c, &cparams)}
if context.c == nil {
return nil, errors.New("unable to create sampling context")
}
runtime.SetFinalizer(context, func(s *SamplingContext) { C.gpt_sampler_cfree(s.c) })
runtime.SetFinalizer(context, func(s *SamplingContext) { C.common_sampler_cfree(s.c) })
return context, nil
}
func (s *SamplingContext) Reset() {
C.gpt_sampler_creset(s.c)
C.common_sampler_creset(s.c)
}
func (s *SamplingContext) Sample(llamaContext *Context, idx int) int {
return int(C.gpt_sampler_csample(s.c, llamaContext.c, C.int(idx)))
return int(C.common_sampler_csample(s.c, llamaContext.c, C.int(idx)))
}
func (s *SamplingContext) Accept(id int, applyGrammar bool) {
C.gpt_sampler_caccept(s.c, C.llama_token(id), C.bool(applyGrammar))
C.common_sampler_caccept(s.c, C.llama_token(id), C.bool(applyGrammar))
}
type JsonSchema struct {