baalajimaestro/ollama
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Merge https://github.com/ollama/ollama

Browse source
This commit is contained in:
baalajimaestro 2024-07-14 16:51:20 +05:30
commit 8c6402d194
Signed by: baalajimaestro
GPG key ID: F93C394FE9BBAFD5
109 changed files with 1730 additions and 1090 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
.github/workflows/release.yaml vendored
View file

@ -147,7 +147,7 @@ jobs:
run: |
$ErrorActionPreference = "Stop"
write-host "downloading AMD HIP Installer"
Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-23.Q4-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q3-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
write-host "Installing AMD HIP"
Start-Process "${env:RUNNER_TEMP}\rocm-install.exe" -ArgumentList '-install' -NoNewWindow -Wait
write-host "Completed AMD HIP"
@ -304,10 +304,6 @@ jobs:
write-host "Installing plugin"
& "${env:RUNNER_TEMP}\plugin\*\kmscng.msi" /quiet
write-host "plugin installed"
- name: remove unwanted mingw dll.a files
run: |
Remove-Item "C:\mingw64\x86_64-w64-mingw32\lib\libpthread.dll.a"
Remove-Item "C:\mingw64\x86_64-w64-mingw32\lib\libwinpthread.dll.a"
- uses: actions/setup-go@v5
with:
go-version-file: go.mod

2
.github/workflows/test.yaml vendored
View file

@ -169,7 +169,7 @@ jobs:
run: |
$ErrorActionPreference = "Stop"
write-host "downloading AMD HIP Installer"
Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-23.Q4-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q3-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
write-host "Installing AMD HIP"
Start-Process "${env:RUNNER_TEMP}\rocm-install.exe" -ArgumentList '-install' -NoNewWindow -Wait
write-host "Completed AMD HIP"

1
README.md
View file

@ -293,6 +293,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
- [OllamaSpring](https://github.com/CrazyNeil/OllamaSpring) (Ollama Client for macOS)
- [LLocal.in](https://github.com/kartikm7/llocal) (Easy to use Electron Desktop Client for Ollama)
- [Ollama with Google Mesop](https://github.com/rapidarchitect/ollama_mesop/) (Mesop Chat Client implementation with Ollama)
- [Kerlig AI](https://www.kerlig.com/) (AI writing assistant for macOS)
### Terminal

2
api/types.go
View file

@ -221,6 +221,8 @@ type DeleteRequest struct {
type ShowRequest struct {
Model string `json:"model"`
System string `json:"system"`
// Template is deprecated
Template string `json:"template"`
Verbose bool `json:"verbose"`

4
app/ollama.iss
View file

@ -127,6 +127,10 @@ Type: filesandordirs; Name: "{%USERPROFILE}\.ollama\models"
Type: filesandordirs; Name: "{%USERPROFILE}\.ollama\history"
; NOTE: if the user has a custom OLLAMA_MODELS it will be preserved
[InstallDelete]
Type: filesandordirs; Name: "{%TEMP}\ollama*"
Type: filesandordirs; Name: "{%LOCALAPPDATA}\Programs\Ollama"
[Messages]
WizardReady=Ollama Windows Preview
ReadyLabel1=%nLet's get you up and running with your own large language models.

2
cmd/cmd.go
View file

@ -843,7 +843,6 @@ type runOptions struct {
WordWrap bool
Format string
System string
Template string
Images []api.ImageData
Options map[string]interface{}
MultiModal bool
@ -1037,7 +1036,6 @@ func generate(cmd *cobra.Command, opts runOptions) error {
Images: opts.Images,
Format: opts.Format,
System: opts.System,
Template: opts.Template,
Options: opts.Options,
KeepAlive: opts.KeepAlive,
}

52
cmd/interactive.go
View file

@ -27,7 +27,6 @@ const (
MultilineNone MultilineState = iota
MultilinePrompt
MultilineSystem
MultilineTemplate
)
func loadModel(cmd *cobra.Command, opts *runOptions) error {
@ -94,7 +93,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
fmt.Fprintln(os.Stderr, "Available Commands:")
fmt.Fprintln(os.Stderr, " /set parameter ... Set a parameter")
fmt.Fprintln(os.Stderr, " /set system <string> Set system message")
fmt.Fprintln(os.Stderr, " /set template <string> Set prompt template")
fmt.Fprintln(os.Stderr, " /set history Enable history")
fmt.Fprintln(os.Stderr, " /set nohistory Disable history")
fmt.Fprintln(os.Stderr, " /set wordwrap Enable wordwrap")
@ -204,10 +202,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
opts.Messages = append(opts.Messages, api.Message{Role: "system", Content: opts.System})
fmt.Println("Set system message.")
sb.Reset()
case MultilineTemplate:
opts.Template = sb.String()
fmt.Println("Set prompt template.")
sb.Reset()
}
multiline = MultilineNone
@ -326,17 +320,13 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
}
fmt.Printf("Set parameter '%s' to '%s'\n", args[2], strings.Join(params, ", "))
opts.Options[args[2]] = fp[args[2]]
case "system", "template":
case "system":
if len(args) < 3 {
usageSet()
continue
}
if args[1] == "system" {
multiline = MultilineSystem
} else if args[1] == "template" {
multiline = MultilineTemplate
}
multiline = MultilineSystem
line := strings.Join(args[2:], " ")
line, ok := strings.CutPrefix(line, `"""`)
@ -356,23 +346,17 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
continue
}
if args[1] == "system" {
opts.System = sb.String() // for display in modelfile
newMessage := api.Message{Role: "system", Content: sb.String()}
// Check if the slice is not empty and the last message is from 'system'
if len(opts.Messages) > 0 && opts.Messages[len(opts.Messages)-1].Role == "system" {
// Replace the last message
opts.Messages[len(opts.Messages)-1] = newMessage
} else {
opts.Messages = append(opts.Messages, newMessage)
}
fmt.Println("Set system message.")
sb.Reset()
} else if args[1] == "template" {
opts.Template = sb.String()
fmt.Println("Set prompt template.")
sb.Reset()
opts.System = sb.String() // for display in modelfile
newMessage := api.Message{Role: "system", Content: sb.String()}
// Check if the slice is not empty and the last message is from 'system'
if len(opts.Messages) > 0 && opts.Messages[len(opts.Messages)-1].Role == "system" {
// Replace the last message
opts.Messages[len(opts.Messages)-1] = newMessage
} else {
opts.Messages = append(opts.Messages, newMessage)
}
fmt.Println("Set system message.")
sb.Reset()
sb.Reset()
continue
@ -393,7 +377,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
req := &api.ShowRequest{
Name: opts.Model,
System: opts.System,
Template: opts.Template,
Options: opts.Options,
}
resp, err := client.Show(cmd.Context(), req)
@ -437,12 +420,9 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
fmt.Println("No system message was specified for this model.")
}
case "template":
switch {
case opts.Template != "":
fmt.Println(opts.Template + "\n")
case resp.Template != "":
if resp.Template != "" {
fmt.Println(resp.Template)
default:
} else {
fmt.Println("No prompt template was specified for this model.")
}
default:
@ -536,10 +516,6 @@ func buildModelfile(opts runOptions) string {
fmt.Fprintf(&mf, "SYSTEM \"\"\"%s\"\"\"\n", opts.System)
}
if opts.Template != "" {
fmt.Fprintf(&mf, "TEMPLATE \"\"\"%s\"\"\"\n", opts.Template)
}
keys := make([]string, 0)
for k := range opts.Options {
keys = append(keys, k)

3
cmd/interactive_test.go
View file

@ -59,7 +59,6 @@ func TestModelfileBuilder(t *testing.T) {
opts := runOptions{
Model: "hork",
System: "You are part horse and part shark, but all hork. Do horklike things",
Template: "This is a template.",
Messages: []api.Message{
{Role: "user", Content: "Hey there hork!"},
{Role: "assistant", Content: "Yes it is true, I am half horse, half shark."},
@ -75,7 +74,6 @@ func TestModelfileBuilder(t *testing.T) {
mf := buildModelfile(opts)
expectedModelfile := `FROM {{.Model}}
SYSTEM """{{.System}}"""
TEMPLATE """{{.Template}}"""
PARAMETER penalize_newline false
PARAMETER seed 42
PARAMETER stop [hi there]
@ -97,7 +95,6 @@ MESSAGE assistant """Yes it is true, I am half horse, half shark."""
mf = buildModelfile(opts)
expectedModelfile = `FROM {{.ParentModel}}
SYSTEM """{{.System}}"""
TEMPLATE """{{.Template}}"""
PARAMETER penalize_newline false
PARAMETER seed 42
PARAMETER stop [hi there]

2
docs/faq.md
View file

@ -272,4 +272,4 @@ The following server settings may be used to adjust how Ollama handles concurren
- `OLLAMA_NUM_PARALLEL` - The maximum number of parallel requests each model will process at the same time. The default will auto-select either 4 or 1 based on available memory.
- `OLLAMA_MAX_QUEUE` - The maximum number of requests Ollama will queue when busy before rejecting additional requests. The default is 512
Note: Windows with Radeon GPUs currently default to 1 model maximum due to limitations in ROCm v5.7 for available VRAM reporting. Once ROCm v6 is available, Windows Radeon will follow the defaults above. You may enable concurrent model loads on Radeon on Windows, but ensure you don't load more models than will fit into your GPUs VRAM.
Note: Windows with Radeon GPUs currently default to 1 model maximum due to limitations in ROCm v5.7 for available VRAM reporting. Once ROCm v6.2 is available, Windows Radeon will follow the defaults above. You may enable concurrent model loads on Radeon on Windows, but ensure you don't load more models than will fit into your GPUs VRAM.

3
go.mod
View file

@ -18,6 +18,7 @@ require (
require (
github.com/agnivade/levenshtein v1.1.1
github.com/d4l3k/go-bfloat16 v0.0.0-20211005043715-690c3bdd05f1
github.com/google/go-cmp v0.6.0
github.com/mattn/go-runewidth v0.0.14
github.com/nlpodyssey/gopickle v0.3.0
github.com/pdevine/tensor v0.0.0-20240510204454-f88f4562727c
@ -71,7 +72,7 @@ require (
golang.org/x/net v0.25.0 // indirect
golang.org/x/sys v0.20.0
golang.org/x/term v0.20.0
golang.org/x/text v0.15.0 // indirect
golang.org/x/text v0.15.0
google.golang.org/protobuf v1.34.1
gopkg.in/yaml.v3 v3.0.1 // indirect
)

23
gpu/amd_common.go
View file

@ -49,9 +49,17 @@ func rocmGetVisibleDevicesEnv(gpuInfo []GpuInfo) (string, string) {
}
func commonAMDValidateLibDir() (string, error) {
// We try to favor system paths first, so that we can wire up the subprocess to use
// the system version. Only use our bundled version if the system version doesn't work
// This gives users a more recovery options if versions have subtle problems at runtime
// Favor our bundled version
// Installer payload location if we're running the installed binary
exe, err := os.Executable()
if err == nil {
rocmTargetDir := filepath.Join(filepath.Dir(exe), "rocm")
if rocmLibUsable(rocmTargetDir) {
slog.Debug("detected ROCM next to ollama executable " + rocmTargetDir)
return rocmTargetDir, nil
}
}
// Prefer explicit HIP env var
hipPath := os.Getenv("HIP_PATH")
@ -87,14 +95,5 @@ func commonAMDValidateLibDir() (string, error) {
}
}
// Installer payload location if we're running the installed binary
exe, err := os.Executable()
if err == nil {
rocmTargetDir := filepath.Join(filepath.Dir(exe), "rocm")
if rocmLibUsable(rocmTargetDir) {
slog.Debug("detected ROCM next to ollama executable " + rocmTargetDir)
return rocmTargetDir, nil
}
}
return "", fmt.Errorf("no suitable rocm found, falling back to CPU")
}

5
gpu/amd_hip_windows.go
View file

@ -84,9 +84,8 @@ func (hl *HipLib) AMDDriverVersion() (driverMajor, driverMinor int, err error) {
}
slog.Debug("hipDriverGetVersion", "version", version)
// TODO - this isn't actually right, but the docs claim hipDriverGetVersion isn't accurate anyway...
driverMajor = version / 1000
driverMinor = (version - (driverMajor * 1000)) / 10
driverMajor = version / 10000000
driverMinor = (version - (driverMajor * 10000000)) / 100000
return driverMajor, driverMinor, nil
}

21
gpu/amd_windows.go
View file

@ -22,8 +22,8 @@ const (
var (
// Used to validate if the given ROCm lib is usable
ROCmLibGlobs = []string{"hipblas.dll", "rocblas"} // TODO - probably include more coverage of files here...
RocmStandardLocations = []string{"C:\\Program Files\\AMD\\ROCm\\5.7\\bin"} // TODO glob?
ROCmLibGlobs = []string{"hipblas.dll", "rocblas"} // This is not sufficient to discern v5 vs v6
RocmStandardLocations = []string{"C:\\Program Files\\AMD\\ROCm\\6.1\\bin"} // TODO glob?
)
func AMDGetGPUInfo() []RocmGPUInfo {
@ -35,12 +35,11 @@ func AMDGetGPUInfo() []RocmGPUInfo {
}
defer hl.Release()
// TODO - this reports incorrect version information, so omitting for now
// driverMajor, driverMinor, err := hl.AMDDriverVersion()
// if err != nil {
// // For now this is benign, but we may eventually need to fail compatibility checks
// slog.Debug("error looking up amd driver version", "error", err)
// }
driverMajor, driverMinor, err := hl.AMDDriverVersion()
if err != nil {
// For now this is benign, but we may eventually need to fail compatibility checks
slog.Debug("error looking up amd driver version", "error", err)
}
// Note: the HIP library automatically handles subsetting to any HIP_VISIBLE_DEVICES the user specified
count := hl.HipGetDeviceCount()
@ -132,10 +131,8 @@ func AMDGetGPUInfo() []RocmGPUInfo {
MinimumMemory: rocmMinimumMemory,
Name: name,
Compute: gfx,
// TODO - this information isn't accurate on windows, so don't report it until we find the right way to retrieve
// DriverMajor: driverMajor,
// DriverMinor: driverMinor,
DriverMajor: driverMajor,
DriverMinor: driverMinor,
},
index: i,
}

30
gpu/gpu.go
View file

@ -274,6 +274,28 @@ func GetGPUInfo() GpuInfoList {
gpuInfo.DriverMajor = driverMajor
gpuInfo.DriverMinor = driverMinor
// query the management library as well so we can record any skew between the two
// which represents overhead on the GPU we must set aside on subsequent updates
if cHandles.nvml != nil {
C.nvml_get_free(*cHandles.nvml, C.int(gpuInfo.index), &memInfo.free, &memInfo.total, &memInfo.used)
if memInfo.err != nil {
slog.Warn("error looking up nvidia GPU memory", "error", C.GoString(memInfo.err))
C.free(unsafe.Pointer(memInfo.err))
} else {
if memInfo.free != 0 && uint64(memInfo.free) > gpuInfo.FreeMemory {
gpuInfo.OSOverhead = uint64(memInfo.free) - gpuInfo.FreeMemory
slog.Info("detected OS VRAM overhead",
"id", gpuInfo.ID,
"library", gpuInfo.Library,
"compute", gpuInfo.Compute,
"driver", fmt.Sprintf("%d.%d", gpuInfo.DriverMajor, gpuInfo.DriverMinor),
"name", gpuInfo.Name,
"overhead", format.HumanBytes2(gpuInfo.OSOverhead),
)
}
}
}
// TODO potentially sort on our own algorithm instead of what the underlying GPU library does...
cudaGPUs = append(cudaGPUs, gpuInfo)
}
@ -338,14 +360,17 @@ func GetGPUInfo() GpuInfoList {
"before",
"total", format.HumanBytes2(cpus[0].TotalMemory),
"free", format.HumanBytes2(cpus[0].FreeMemory),
"free_swap", format.HumanBytes2(cpus[0].FreeSwap),
),
slog.Group(
"now",
"total", format.HumanBytes2(mem.TotalMemory),
"free", format.HumanBytes2(mem.FreeMemory),
"free_swap", format.HumanBytes2(mem.FreeSwap),
),
)
cpus[0].FreeMemory = mem.FreeMemory
cpus[0].FreeSwap = mem.FreeSwap
}
var memInfo C.mem_info_t
@ -374,9 +399,14 @@ func GetGPUInfo() GpuInfoList {
slog.Warn("error looking up nvidia GPU memory")
continue
}
if cHandles.nvml != nil && gpu.OSOverhead > 0 {
// When using the management library update based on recorded overhead
memInfo.free -= C.uint64_t(gpu.OSOverhead)
}
slog.Debug("updating cuda memory data",
"gpu", gpu.ID,
"name", gpu.Name,
"overhead", format.HumanBytes2(gpu.OSOverhead),
slog.Group(
"before",
"total", format.HumanBytes2(gpu.TotalMemory),

3
gpu/gpu_darwin.go
View file

@ -56,7 +56,8 @@ func GetCPUInfo() GpuInfoList {
func GetCPUMem() (memInfo, error) {
return memInfo{
TotalMemory: uint64(C.getPhysicalMemory()),
FreeMemory: 0,
FreeMemory: uint64(C.getFreeMemory()),
// FreeSwap omitted as Darwin uses dynamic paging
}, nil
}

1
gpu/gpu_info_darwin.h
View file

@ -2,3 +2,4 @@
#include <stdint.h>
uint64_t getRecommendedMaxVRAM();
uint64_t getPhysicalMemory();
uint64_t getFreeMemory();

26
gpu/gpu_info_darwin.m
View file

@ -1,4 +1,5 @@
// go:build darwin
#import <Foundation/Foundation.h>
#import <mach/mach.h>
#include "gpu_info_darwin.h"
uint64_t getRecommendedMaxVRAM() {
@ -8,6 +9,27 @@ uint64_t getRecommendedMaxVRAM() {
return result;
}
// getPhysicalMemory returns the total physical memory in bytes
uint64_t getPhysicalMemory() {
return [[NSProcessInfo processInfo] physicalMemory];
return [NSProcessInfo processInfo].physicalMemory;
}
// getFreeMemory returns the total free memory in bytes, including inactive
// memory that can be reclaimed by the system.
uint64_t getFreeMemory() {
mach_port_t host_port = mach_host_self();
mach_msg_type_number_t host_size = sizeof(vm_statistics64_data_t) / sizeof(integer_t);
vm_size_t pagesize;
vm_statistics64_data_t vm_stat;
host_page_size(host_port, &pagesize);
if (host_statistics64(host_port, HOST_VM_INFO64, (host_info64_t)&vm_stat, &host_size) != KERN_SUCCESS) {
return 0;
}
uint64_t free_memory = (uint64_t)vm_stat.free_count * pagesize;
free_memory += (uint64_t)vm_stat.speculative_count * pagesize;
free_memory += (uint64_t)vm_stat.inactive_count * pagesize;
return free_memory;
}

17
gpu/gpu_linux.go
View file

@ -50,7 +50,7 @@ var OneapiMgmtName = "libze_intel_gpu.so"
func GetCPUMem() (memInfo, error) {
var mem memInfo
var total, available, free, buffers, cached uint64
var total, available, free, buffers, cached, freeSwap uint64
f, err := os.Open("/proc/meminfo")
if err != nil {
return mem, err
@ -70,20 +70,21 @@ func GetCPUMem() (memInfo, error) {
_, err = fmt.Sscanf(line, "Buffers:%d", &buffers)
case strings.HasPrefix(line, "Cached:"):
_, err = fmt.Sscanf(line, "Cached:%d", &cached)
case strings.HasPrefix(line, "SwapFree:"):
_, err = fmt.Sscanf(line, "SwapFree:%d", &freeSwap)
default:
continue
}
if err != nil {
return mem, err
}
if total > 0 && available > 0 {
mem.TotalMemory = total * format.KibiByte
mem.FreeMemory = available * format.KibiByte
return mem, nil
}
}
mem.TotalMemory = total * format.KibiByte
mem.FreeMemory = (free + buffers + cached) * format.KibiByte
mem.FreeSwap = freeSwap * format.KibiByte
if available > 0 {
mem.FreeMemory = available * format.KibiByte
} else {
mem.FreeMemory = (free + buffers + cached) * format.KibiByte
}
return mem, nil
}

2
gpu/gpu_windows.go
View file

@ -51,5 +51,5 @@ func GetCPUMem() (memInfo, error) {
if r1 == 0 {
return memInfo{}, fmt.Errorf("GlobalMemoryStatusEx failed: %w", err)
}
return memInfo{TotalMemory: memStatus.TotalPhys, FreeMemory: memStatus.AvailPhys}, nil
return memInfo{TotalMemory: memStatus.TotalPhys, FreeMemory: memStatus.AvailPhys, FreeSwap: memStatus.AvailPageFile}, nil
}

4
gpu/types.go
View file

@ -10,6 +10,7 @@ import (
type memInfo struct {
TotalMemory uint64 `json:"total_memory,omitempty"`
FreeMemory uint64 `json:"free_memory,omitempty"`
FreeSwap uint64 `json:"free_swap,omitempty"`
}
// Beginning of an `ollama info` command
@ -52,7 +53,8 @@ type CPUInfo struct {
type CudaGPUInfo struct {
GpuInfo
index int //nolint:unused,nolintlint
OSOverhead uint64 // Memory overhead between the driver library and management library
index int //nolint:unused,nolintlint
}
type CudaGPUInfoList []CudaGPUInfo

31
llm/ext_server/server.cpp vendored
View file

@ -1413,7 +1413,7 @@ struct llama_server_context
return get_slot(-1);
}
LOG_INFO("slot with common prefix found", {{
LOG_DEBUG("slot with common prefix found", {{
"slot_id", slot->id,
"characters", longest
}});
@ -1688,22 +1688,8 @@ struct llama_server_context
}
slot.params.n_keep = std::min(slot.n_ctx - 4, slot.params.n_keep);
char buf[256];
llama_model_meta_val_str(model, "general.architecture", buf, 256);
bool gemma2 = strcmp(buf, "gemma2") == 0;
int32_t truncate_at = slot.n_ctx;
// truncate at 2/3 of the context length for gemma2 models
// as they do not support context shifts (from the sliding window implementation).
// this way, prompts that almost fit the context length can still generate a full
// response without a sudden stop from hitting the context limit
if (gemma2) {
truncate_at = 2 * slot.n_ctx / 3;
}
// if input prompt is too big, truncate it, if group attention self-extend is disabled
if (slot.ga_n == 1 && slot.n_prompt_tokens >= truncate_at)
if (slot.ga_n == 1 && slot.n_prompt_tokens >= slot.n_ctx)
{
const int n_left = slot.n_ctx - slot.params.n_keep;
const int n_shift = n_left / 2;
@ -1731,19 +1717,6 @@ struct llama_server_context
GGML_ASSERT(slot.n_prompt_tokens < slot.n_ctx);
}
// Models with sliding window attention do not work with context shifts, so
// limit their prediction to the context length
if (gemma2) {
int32_t limit = slot.n_ctx - slot.n_prompt_tokens;
slot.n_predict = limit;
slot.params.n_predict = limit;
LOG_INFO("model does not support sliding window, limiting generation", {
{"n_ctx", slot.n_ctx},
{"n_prompt_tokens", slot.n_prompt_tokens},
{"n_predict", slot.n_predict}
});
}
if (!slot.params.cache_prompt)
{
llama_sampling_reset(slot.ctx_sampling);

8
llm/generate/gen_linux.sh
View file

@ -77,7 +77,7 @@ if [ -z "${OLLAMA_SKIP_CPU_GENERATE}" ]; then
if [ -n "${OLLAMA_CUSTOM_CPU_DEFS}" ]; then
init_vars
echo "OLLAMA_CUSTOM_CPU_DEFS=\"${OLLAMA_CUSTOM_CPU_DEFS}\""
CMAKE_DEFS="${OLLAMA_CUSTOM_CPU_DEFS} -DCMAKE_POSITION_INDEPENDENT_CODE=on ${CMAKE_DEFS}"
CMAKE_DEFS="${OLLAMA_CUSTOM_CPU_DEFS} -DBUILD_SHARED_LIBS=off -DCMAKE_POSITION_INDEPENDENT_CODE=on ${CMAKE_DEFS}"
BUILD_DIR="../build/linux/${ARCH}/cpu"
echo "Building custom CPU"
build
@ -93,7 +93,7 @@ if [ -z "${OLLAMA_SKIP_CPU_GENERATE}" ]; then
# -DGGML_AVX512_VBMI -- 2018 Intel Cannon Lake
# -DGGML_AVX512_VNNI -- 2021 Intel Alder Lake
COMMON_CPU_DEFS="-DCMAKE_POSITION_INDEPENDENT_CODE=on -DGGML_NATIVE=off -DGGML_OPENMP=off"
COMMON_CPU_DEFS="-DBUILD_SHARED_LIBS=off -DCMAKE_POSITION_INDEPENDENT_CODE=on -DGGML_NATIVE=off -DGGML_OPENMP=off"
if [ -z "${OLLAMA_CPU_TARGET}" -o "${OLLAMA_CPU_TARGET}" = "cpu" ]; then
#
# CPU first for the default library, set up as lowest common denominator for maximum compatibility (including Rosetta)
@ -178,7 +178,7 @@ if [ -z "${OLLAMA_SKIP_CUDA_GENERATE}" -a -d "${CUDA_LIB_DIR}" ]; then
CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} ${OLLAMA_CUSTOM_CUDA_DEFS}"
echo "Building custom CUDA GPU"
else
CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_FLAGS=-t8 -DGGML_CUDA_FORCE_MMQ=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} -DCMAKE_LIBRARY_PATH=/usr/local/cuda/compat"
CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_FLAGS=-t8 -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES}"
fi
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} ${ARM64_DEFS} ${CMAKE_CUDA_DEFS}"
BUILD_DIR="../build/linux/${ARCH}/cuda${CUDA_VARIANT}"
@ -254,7 +254,7 @@ if [ -z "${OLLAMA_SKIP_ROCM_GENERATE}" -a -d "${ROCM_PATH}" ]; then
ROCM_VARIANT=_v$(ls ${ROCM_PATH}/lib/librocblas.so.*.*.????? | cut -f5 -d. || true)
fi
init_vars
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DGGML_HIPBLAS=on -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ -DAMDGPU_TARGETS=$(amdGPUs) -DGPU_TARGETS=$(amdGPUs)"
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DGGML_HIPBLAS=on -DLLAMA_CUDA_NO_PEER_COPY=on -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ -DAMDGPU_TARGETS=$(amdGPUs) -DGPU_TARGETS=$(amdGPUs)"
# Users building from source can tune the exact flags we pass to cmake for configuring llama.cpp
if [ -n "${OLLAMA_CUSTOM_ROCM_DEFS}" ]; then
echo "OLLAMA_CUSTOM_ROCM_DEFS=\"${OLLAMA_CUSTOM_ROCM_DEFS}\""

13
llm/generate/gen_windows.ps1
View file

@ -6,18 +6,9 @@ function amdGPUs {
if ($env:AMDGPU_TARGETS) {
return $env:AMDGPU_TARGETS
}
# TODO - load from some common data file for linux + windows build consistency
# Current supported rocblas list from ROCm v6.1.2 on windows
$GPU_LIST = @(
"gfx900"
"gfx906:xnack-"
"gfx908:xnack-"
"gfx90a:xnack+"
"gfx90a:xnack-"
"gfx940"
"gfx941"
"gfx942"
"gfx1010"
"gfx1012"
"gfx1030"
"gfx1100"
"gfx1101"
@ -366,6 +357,7 @@ function build_rocm() {
"-DCMAKE_C_COMPILER=clang.exe",
"-DCMAKE_CXX_COMPILER=clang++.exe",
"-DGGML_HIPBLAS=on",
"-DLLAMA_CUDA_NO_PEER_COPY=on",
"-DHIP_PLATFORM=amd",
"-DGGML_AVX=on",
"-DGGML_AVX2=off",
@ -394,7 +386,6 @@ function build_rocm() {
sign
install
# Assumes v5.7, may need adjustments for v6
rm -ea 0 -recurse -force -path "${script:SRC_DIR}\dist\windows-${script:ARCH}\rocm\"
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\rocm\rocblas\library\" -ea 0 > $null
cp "${env:HIP_PATH}\bin\hipblas.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\rocm\"

26
llm/ggml.go
View file

@ -424,6 +424,32 @@ func (llm GGML) GraphSize(context, batch uint64) (partialOffload, fullOffload ui
4*batch*(3*embedding+vocab)+embedding*vocab*105/128,
4*batch*(2*embedding+1+2*embeddingHeadsK*headsKV+context+context*headsKV)+4*embeddingHeadsK*context*headsKV+embedding*embeddingHeadsK*headsKV*9/16,
)
case "chatglm":
fullOffload = 4 * batch * (embedding + vocab)
partialOffload = 4*batch*(embedding+vocab) + embedding*vocab*105/128
if qkvBias, ok := layers["blk.0"]["attn_qkv.bias"]; ok {
fullOffload = max(
fullOffload,
4*batch*(2+
2*embedding+
context+
context*heads+
embeddingHeadsK*heads+
qkvBias.Shape[0]),
)
partialOffload = max(
partialOffload,
4*batch*(1+
2*embedding+
embeddingHeadsK*heads+
context+
context*heads)+
4*embeddingHeadsK*context+
4*context*embeddingHeadsK+
4*qkvBias.Shape[0],
)
}
}
return

2
llm/llama.cpp

@ -1 +1 @@
Subproject commit d7fd29fff16456ce9c3a23fd2d09a66256b05aff
Subproject commit a8db2a9ce64cd4417f6a312ab61858f17f0f8584

7
llm/llm.go
View file

@ -1,12 +1,11 @@
package llm
// #cgo CFLAGS: -Illama.cpp -Illama.cpp/include -Illama.cpp/ggml/include
// #cgo windows LDFLAGS: -static-libstdc++
// #cgo LDFLAGS: -lllama -lggml -lstdc++ -lpthread
// #cgo darwin,arm64 LDFLAGS: -L${SRCDIR}/build/darwin/arm64_static -L${SRCDIR}/build/darwin/arm64_static/src -L${SRCDIR}/build/darwin/arm64_static/ggml/src -framework Accelerate -framework Metal
// #cgo darwin,amd64 LDFLAGS: -L${SRCDIR}/build/darwin/x86_64_static -L${SRCDIR}/build/darwin/x86_64_static/src -L${SRCDIR}/build/darwin/x86_64_static/ggml/src
// #cgo windows,amd64 LDFLAGS: -L${SRCDIR}/build/windows/amd64_static -L${SRCDIR}/build/windows/amd64_static/src -L${SRCDIR}/build/windows/amd64_static/ggml/src
// #cgo windows,arm64 LDFLAGS: -L${SRCDIR}/build/windows/arm64_static -L${SRCDIR}/build/windows/arm64_static/src -L${SRCDIR}/build/windows/arm64_static/ggml/src
// #cgo windows,amd64 LDFLAGS: -static-libstdc++ -static-libgcc -static -L${SRCDIR}/build/windows/amd64_static -L${SRCDIR}/build/windows/amd64_static/src -L${SRCDIR}/build/windows/amd64_static/ggml/src
// #cgo windows,arm64 LDFLAGS: -static-libstdc++ -static-libgcc -static -L${SRCDIR}/build/windows/arm64_static -L${SRCDIR}/build/windows/arm64_static/src -L${SRCDIR}/build/windows/arm64_static/ggml/src
// #cgo linux,amd64 LDFLAGS: -L${SRCDIR}/build/linux/x86_64_static -L${SRCDIR}/build/linux/x86_64_static/src -L${SRCDIR}/build/linux/x86_64_static/ggml/src
// #cgo linux,arm64 LDFLAGS: -L${SRCDIR}/build/linux/arm64_static -L${SRCDIR}/build/linux/arm64_static/src -L${SRCDIR}/build/linux/arm64_static/ggml/src
// #include <stdlib.h>
@ -34,7 +33,7 @@ func Quantize(infile, outfile string, ftype fileType) error {
params.ftype = ftype.Value()
if rc := C.llama_model_quantize(cinfile, coutfile, &params); rc != 0 {
return fmt.Errorf("llama_model_quantize: %d", rc)
return fmt.Errorf("failed to quantize model. This model architecture may not be supported, or you may need to upgrade Ollama to the latest version")
}
return nil

10
llm/patches/05-default-pretokenizer.diff
View file

@ -1,11 +1,11 @@
diff --git a/src/llama.cpp b/src/llama.cpp
index 73f52435..2b81b4bd 100644
index 2b9ace28..172640e2 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -5092,16 +5092,7 @@ static void llm_load_vocab(
// for now, only BPE models have pre-tokenizers
@@ -5357,16 +5357,7 @@ static void llm_load_vocab(
if (vocab.type == LLAMA_VOCAB_TYPE_BPE) {
vocab.tokenizer_add_space_prefix = false;
vocab.tokenizer_clean_spaces = true;
- if (tokenizer_pre.empty()) {
- LLAMA_LOG_WARN("%s: missing pre-tokenizer type, using: 'default'\n", __func__);
- LLAMA_LOG_WARN("%s: \n", __func__);
@ -20,7 +20,7 @@ index 73f52435..2b81b4bd 100644
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
} else if (
tokenizer_pre == "llama3" ||
@@ -5164,7 +5155,8 @@ static void llm_load_vocab(
@@ -5439,7 +5430,8 @@ static void llm_load_vocab(
tokenizer_pre == "jais") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_JAIS;
} else {

23
llm/server.go
View file

@ -88,6 +88,7 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
var estimate MemoryEstimate
var systemTotalMemory uint64
var systemFreeMemory uint64
var systemSwapFreeMemory uint64
systemMemInfo, err := gpu.GetCPUMem()
if err != nil {
@ -95,7 +96,8 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
} else {
systemTotalMemory = systemMemInfo.TotalMemory
systemFreeMemory = systemMemInfo.FreeMemory
slog.Debug("system memory", "total", format.HumanBytes2(systemTotalMemory), "free", systemFreeMemory)
systemSwapFreeMemory = systemMemInfo.FreeSwap
slog.Debug("system memory", "total", format.HumanBytes2(systemTotalMemory), "free", format.HumanBytes2(systemFreeMemory), "free_swap", format.HumanBytes2(systemSwapFreeMemory))
}
// If the user wants zero GPU layers, reset the gpu list to be CPU/system ram info
@ -122,6 +124,16 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
}
}
// On linux, over-allocating CPU memory will almost always result in an error
if runtime.GOOS == "linux" {
systemMemoryRequired := estimate.TotalSize - estimate.VRAMSize
available := systemFreeMemory + systemSwapFreeMemory
if systemMemoryRequired > available {
slog.Warn("model request too large for system", "requested", format.HumanBytes2(systemMemoryRequired), "available", available, "total", format.HumanBytes2(systemTotalMemory), "free", format.HumanBytes2(systemFreeMemory), "swap", format.HumanBytes2(systemSwapFreeMemory))
return nil, fmt.Errorf("model requires more system memory (%s) than is available (%s)", format.HumanBytes2(systemMemoryRequired), format.HumanBytes2(available))
}
}
estimate.log()
// Loop through potential servers
@ -254,10 +266,6 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
params = append(params, "--tensor-split", estimate.TensorSplit)
}
if estimate.TensorSplit != "" {
params = append(params, "--tensor-split", estimate.TensorSplit)
}
for i := range len(servers) {
dir := availableServers[servers[i]]
if dir == "" {
@ -679,7 +687,7 @@ type CompletionRequest struct {
Prompt string
Format string
Images []ImageData
Options api.Options
Options *api.Options
}
type CompletionResponse struct {
@ -699,10 +707,9 @@ func (s *llmServer) Completion(ctx context.Context, req CompletionRequest, fn fu
}
defer s.sem.Release(1)
// only allow maximum 10 "context shifts" to avoid infinite generation
// put an upper limit on num_predict to avoid the model running on forever
if req.Options.NumPredict < 0 || req.Options.NumPredict > 10*s.options.NumCtx {
req.Options.NumPredict = 10 * s.options.NumCtx
slog.Debug("setting token limit to 10x num_ctx", "num_ctx", s.options.NumCtx, "num_predict", req.Options.NumPredict)
}
request := map[string]any{

90
openai/openai.go
View file

@ -3,11 +3,13 @@ package openai
import (
"bytes"
"encoding/base64"
"encoding/json"
"fmt"
"io"
"math/rand"
"net/http"
"strings"
"time"
"github.com/gin-gonic/gin"
@ -28,7 +30,7 @@ type ErrorResponse struct {
type Message struct {
Role string `json:"role"`
Content string `json:"content"`
Content any `json:"content"`
}
type Choice struct {
@ -269,10 +271,66 @@ func toModel(r api.ShowResponse, m string) Model {
}
}
func fromChatRequest(r ChatCompletionRequest) api.ChatRequest {
func fromChatRequest(r ChatCompletionRequest) (*api.ChatRequest, error) {
var messages []api.Message
for _, msg := range r.Messages {
messages = append(messages, api.Message{Role: msg.Role, Content: msg.Content})
switch content := msg.Content.(type) {
case string:
messages = append(messages, api.Message{Role: msg.Role, Content: content})
case []any:
message := api.Message{Role: msg.Role}
for _, c := range content {
data, ok := c.(map[string]any)
if !ok {
return nil, fmt.Errorf("invalid message format")
}
switch data["type"] {
case "text":
text, ok := data["text"].(string)
if !ok {
return nil, fmt.Errorf("invalid message format")
}
message.Content = text
case "image_url":
var url string
if urlMap, ok := data["image_url"].(map[string]any); ok {
if url, ok = urlMap["url"].(string); !ok {
return nil, fmt.Errorf("invalid message format")
}
} else {
if url, ok = data["image_url"].(string); !ok {
return nil, fmt.Errorf("invalid message format")
}
}
types := []string{"jpeg", "jpg", "png"}
valid := false
for _, t := range types {
prefix := "data:image/" + t + ";base64,"
if strings.HasPrefix(url, prefix) {
url = strings.TrimPrefix(url, prefix)
valid = true
break
}
}
if !valid {
return nil, fmt.Errorf("invalid image input")
}
img, err := base64.StdEncoding.DecodeString(url)
if err != nil {
return nil, fmt.Errorf("invalid message format")
}
message.Images = append(message.Images, img)
default:
return nil, fmt.Errorf("invalid message format")
}
}
messages = append(messages, message)
default:
return nil, fmt.Errorf("invalid message content type: %T", content)
}
}
options := make(map[string]interface{})
@ -323,13 +381,13 @@ func fromChatRequest(r ChatCompletionRequest) api.ChatRequest {
format = "json"
}
return api.ChatRequest{
return &api.ChatRequest{
Model: r.Model,
Messages: messages,
Format: format,
Options: options,
Stream: &r.Stream,
}
}, nil
}
func fromCompleteRequest(r CompletionRequest) (api.GenerateRequest, error) {
@ -338,12 +396,16 @@ func fromCompleteRequest(r CompletionRequest) (api.GenerateRequest, error) {
switch stop := r.Stop.(type) {
case string:
options["stop"] = []string{stop}
case []string:
options["stop"] = stop
default:
if r.Stop != nil {
return api.GenerateRequest{}, fmt.Errorf("invalid type for 'stop' field: %T", r.Stop)
case []any:
var stops []string
for _, s := range stop {
if str, ok := s.(string); ok {
stops = append(stops, str)
} else {
return api.GenerateRequest{}, fmt.Errorf("invalid type for 'stop' field: %T", s)
}
}
options["stop"] = stops
}
if r.MaxTokens != nil {
@ -652,7 +714,13 @@ func ChatMiddleware() gin.HandlerFunc {
}
var b bytes.Buffer
if err := json.NewEncoder(&b).Encode(fromChatRequest(req)); err != nil {
chatReq, err := fromChatRequest(req)
if err != nil {
c.AbortWithStatusJSON(http.StatusBadRequest, NewError(http.StatusBadRequest, err.Error()))
}
if err := json.NewEncoder(&b).Encode(chatReq); err != nil {
c.AbortWithStatusJSON(http.StatusInternalServerError, NewError(http.StatusInternalServerError, err.Error()))
return
}

332
openai/openai_test.go
View file

@ -2,8 +2,8 @@ package openai
import (
"bytes"
"encoding/base64"
"encoding/json"
"fmt"
"io"
"net/http"
"net/http/httptest"
@ -16,7 +16,181 @@ import (
"github.com/stretchr/testify/assert"
)
func TestMiddleware(t *testing.T) {
const prefix = `data:image/jpeg;base64,`
const image = `iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+A8AAQUBAScY42YAAAAASUVORK5CYII=`
const imageURL = prefix + image
func TestMiddlewareRequests(t *testing.T) {
type testCase struct {
Name string
Method string
Path string
Handler func() gin.HandlerFunc
Setup func(t *testing.T, req *http.Request)
Expected func(t *testing.T, req *http.Request)
}
var capturedRequest *http.Request
captureRequestMiddleware := func() gin.HandlerFunc {
return func(c *gin.Context) {
bodyBytes, _ := io.ReadAll(c.Request.Body)
c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
capturedRequest = c.Request
c.Next()
}
}
testCases := []testCase{
{
Name: "chat handler",
Method: http.MethodPost,
Path: "/api/chat",
Handler: ChatMiddleware,
Setup: func(t *testing.T, req *http.Request) {
body := ChatCompletionRequest{
Model: "test-model",
Messages: []Message{{Role: "user", Content: "Hello"}},
}
bodyBytes, _ := json.Marshal(body)
req.Body = io.NopCloser(bytes.NewReader(bodyBytes))
req.Header.Set("Content-Type", "application/json")
},
Expected: func(t *testing.T, req *http.Request) {
var chatReq api.ChatRequest
if err := json.NewDecoder(req.Body).Decode(&chatReq); err != nil {
t.Fatal(err)
}
if chatReq.Messages[0].Role != "user" {
t.Fatalf("expected 'user', got %s", chatReq.Messages[0].Role)
}
if chatReq.Messages[0].Content != "Hello" {
t.Fatalf("expected 'Hello', got %s", chatReq.Messages[0].Content)
}
},
},
{
Name: "completions handler",
Method: http.MethodPost,
Path: "/api/generate",
Handler: CompletionsMiddleware,
Setup: func(t *testing.T, req *http.Request) {
temp := float32(0.8)
body := CompletionRequest{
Model: "test-model",
Prompt: "Hello",
Temperature: &temp,
Stop: []string{"\n", "stop"},
}
bodyBytes, _ := json.Marshal(body)
req.Body = io.NopCloser(bytes.NewReader(bodyBytes))
req.Header.Set("Content-Type", "application/json")
},
Expected: func(t *testing.T, req *http.Request) {
var genReq api.GenerateRequest
if err := json.NewDecoder(req.Body).Decode(&genReq); err != nil {
t.Fatal(err)
}
if genReq.Prompt != "Hello" {
t.Fatalf("expected 'Hello', got %s", genReq.Prompt)
}
if genReq.Options["temperature"] != 1.6 {
t.Fatalf("expected 1.6, got %f", genReq.Options["temperature"])
}
stopTokens, ok := genReq.Options["stop"].([]any)
if !ok {
t.Fatalf("expected stop tokens to be a list")
}
if stopTokens[0] != "\n" || stopTokens[1] != "stop" {
t.Fatalf("expected ['\\n', 'stop'], got %v", stopTokens)
}
},
},
{
Name: "chat handler with image content",
Method: http.MethodPost,
Path: "/api/chat",
Handler: ChatMiddleware,
Setup: func(t *testing.T, req *http.Request) {
body := ChatCompletionRequest{
Model: "test-model",
Messages: []Message{
{
Role: "user", Content: []map[string]any{
{"type": "text", "text": "Hello"},
{"type": "image_url", "image_url": map[string]string{"url": imageURL}},
},
},
},
}
bodyBytes, _ := json.Marshal(body)
req.Body = io.NopCloser(bytes.NewReader(bodyBytes))
req.Header.Set("Content-Type", "application/json")
},
Expected: func(t *testing.T, req *http.Request) {
var chatReq api.ChatRequest
if err := json.NewDecoder(req.Body).Decode(&chatReq); err != nil {
t.Fatal(err)
}
if chatReq.Messages[0].Role != "user" {
t.Fatalf("expected 'user', got %s", chatReq.Messages[0].Role)
}
if chatReq.Messages[0].Content != "Hello" {
t.Fatalf("expected 'Hello', got %s", chatReq.Messages[0].Content)
}
img, _ := base64.StdEncoding.DecodeString(imageURL[len(prefix):])
if !bytes.Equal(chatReq.Messages[0].Images[0], img) {
t.Fatalf("expected image encoding, got %s", chatReq.Messages[0].Images[0])
}
},
},
}
gin.SetMode(gin.TestMode)
router := gin.New()
endpoint := func(c *gin.Context) {
c.Status(http.StatusOK)
}
for _, tc := range testCases {
t.Run(tc.Name, func(t *testing.T) {
router = gin.New()
router.Use(captureRequestMiddleware())
router.Use(tc.Handler())
router.Handle(tc.Method, tc.Path, endpoint)
req, _ := http.NewRequest(tc.Method, tc.Path, nil)
if tc.Setup != nil {
tc.Setup(t, req)
}
resp := httptest.NewRecorder()
router.ServeHTTP(resp, req)
tc.Expected(t, capturedRequest)
})
}
}
func TestMiddlewareResponses(t *testing.T) {
type testCase struct {
Name string
Method string
@ -30,159 +204,7 @@ func TestMiddleware(t *testing.T) {
testCases := []testCase{
{
Name: "chat handler",
Method: http.MethodPost,
Path: "/api/chat",
TestPath: "/api/chat",
Handler: ChatMiddleware,
Endpoint: func(c *gin.Context) {
var chatReq api.ChatRequest
if err := c.ShouldBindJSON(&chatReq); err != nil {
c.JSON(http.StatusBadRequest, gin.H{"error": "invalid request"})
return
}
userMessage := chatReq.Messages[0].Content
var assistantMessage string
switch userMessage {
case "Hello":
assistantMessage = "Hello!"
default:
assistantMessage = "I'm not sure how to respond to that."
}
c.JSON(http.StatusOK, api.ChatResponse{
Message: api.Message{
Role: "assistant",