Skip to content

Commit 1bc35c3

Browse files
graehlgraehl
committed
Merge branch 'master' into finelayer
2 parents 6b51522 + 053b153 commit 1bc35c3

File tree

134 files changed

+6577
-4909
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

134 files changed

+6577
-4909
lines changed

‎.editorconfig‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -49,6 +49,6 @@ charset = unset
4949
trim_trailing_whitespace = unset
5050
insert_final_newline = unset
5151

52-
[tools/mtmd/miniaudio.h]
52+
[vendor/miniaudio/miniaudio.h]
5353
trim_trailing_whitespace = unset
5454
insert_final_newline = unset

‎.github/workflows/build-linux-cross.yml‎

Lines changed: 15 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -26,12 +26,12 @@ jobs:
2626
sudo apt-get install -y --no-install-recommends \
2727
build-essential \
2828
gcc-14-riscv64-linux-gnu \
29-
g++-14-riscv64-linux-gnu \
30-
libcurl4-openssl-dev:riscv64
29+
g++-14-riscv64-linux-gnu
3130
3231
- name: Build
3332
run: |
34-
cmake -B build -DCMAKE_BUILD_TYPE=Release \
33+
cmake -B build -DLLAMA_CURL=OFF \
34+
-DCMAKE_BUILD_TYPE=Release \
3535
-DGGML_OPENMP=OFF \
3636
-DLLAMA_BUILD_EXAMPLES=ON \
3737
-DLLAMA_BUILD_TOOLS=ON \
@@ -72,12 +72,12 @@ jobs:
7272
glslc \
7373
gcc-14-riscv64-linux-gnu \
7474
g++-14-riscv64-linux-gnu \
75-
libvulkan-dev:riscv64 \
76-
libcurl4-openssl-dev:riscv64
75+
libvulkan-dev:riscv64
7776
7877
- name: Build
7978
run: |
80-
cmake -B build -DCMAKE_BUILD_TYPE=Release \
79+
cmake -B build -DLLAMA_CURL=OFF \
80+
-DCMAKE_BUILD_TYPE=Release \
8181
-DGGML_VULKAN=ON \
8282
-DGGML_OPENMP=OFF \
8383
-DLLAMA_BUILD_EXAMPLES=ON \
@@ -118,12 +118,12 @@ jobs:
118118
build-essential \
119119
glslc \
120120
crossbuild-essential-arm64 \
121-
libvulkan-dev:arm64 \
122-
libcurl4-openssl-dev:arm64
121+
libvulkan-dev:arm64
123122
124123
- name: Build
125124
run: |
126-
cmake -B build -DCMAKE_BUILD_TYPE=Release \
125+
cmake -B build -DLLAMA_CURL=OFF \
126+
-DCMAKE_BUILD_TYPE=Release \
127127
-DGGML_VULKAN=ON \
128128
-DGGML_OPENMP=OFF \
129129
-DLLAMA_BUILD_EXAMPLES=ON \
@@ -163,12 +163,12 @@ jobs:
163163
sudo apt-get install -y --no-install-recommends \
164164
build-essential \
165165
gcc-14-powerpc64le-linux-gnu \
166-
g++-14-powerpc64le-linux-gnu \
167-
libcurl4-openssl-dev:ppc64el
166+
g++-14-powerpc64le-linux-gnu
168167
169168
- name: Build
170169
run: |
171-
cmake -B build -DCMAKE_BUILD_TYPE=Release \
170+
cmake -B build -DLLAMA_CURL=OFF \
171+
-DCMAKE_BUILD_TYPE=Release \
172172
-DGGML_OPENMP=OFF \
173173
-DLLAMA_BUILD_EXAMPLES=ON \
174174
-DLLAMA_BUILD_TOOLS=ON \
@@ -209,12 +209,12 @@ jobs:
209209
glslc \
210210
gcc-14-powerpc64le-linux-gnu \
211211
g++-14-powerpc64le-linux-gnu \
212-
libvulkan-dev:ppc64el \
213-
libcurl4-openssl-dev:ppc64el
212+
libvulkan-dev:ppc64el
214213
215214
- name: Build
216215
run: |
217-
cmake -B build -DCMAKE_BUILD_TYPE=Release \
216+
cmake -B build -DLLAMA_CURL=OFF \
217+
-DCMAKE_BUILD_TYPE=Release \
218218
-DGGML_VULKAN=ON \
219219
-DGGML_OPENMP=OFF \
220220
-DLLAMA_BUILD_EXAMPLES=ON \

‎common/CMakeLists.txt‎

Lines changed: 5 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -58,23 +58,20 @@ add_library(${TARGET} STATIC
5858
arg.cpp
5959
arg.h
6060
base64.hpp
61-
chat.cpp
62-
chat.h
6361
chat-parser.cpp
6462
chat-parser.h
63+
chat.cpp
64+
chat.h
6565
common.cpp
6666
common.h
6767
console.cpp
6868
console.h
69-
json-schema-to-grammar.cpp
70-
json.hpp
71-
json-partial.h
7269
json-partial.cpp
70+
json-partial.h
71+
json-schema-to-grammar.cpp
7372
llguidance.cpp
7473
log.cpp
7574
log.h
76-
minja/chat-template.hpp
77-
minja/minja.hpp
7875
ngram-cache.cpp
7976
ngram-cache.h
8077
regex-partial.cpp
@@ -147,7 +144,7 @@ if (LLAMA_LLGUIDANCE)
147144
set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} llguidance ${LLGUIDANCE_PLATFORM_LIBS})
148145
endif ()
149146

150-
target_include_directories(${TARGET} PUBLIC .)
147+
target_include_directories(${TARGET} PUBLIC . ../vendor)
151148
target_compile_features (${TARGET} PUBLIC cxx_std_17)
152149
target_link_libraries (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
153150

‎common/arg.cpp‎

Lines changed: 21 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -16,6 +16,9 @@
1616
# include <windows.h>
1717
#endif
1818

19+
#define JSON_ASSERT GGML_ASSERT
20+
#include <nlohmann/json.hpp>
21+
1922
#include <algorithm>
2023
#include <climits>
2124
#include <cstdarg>
@@ -36,8 +39,6 @@
3639
# include <future>
3740
#endif
3841

39-
#include "json-schema-to-grammar.h"
40-
4142
using json = nlohmann::ordered_json;
4243

4344
std::initializer_list<enum llama_example> mmproj_examples = {
@@ -1300,6 +1301,7 @@ common_params_context common_params_parser_init(common_params & params,
13001301
sampler_type_names.pop_back();
13011302

13021303
params.optimize = ggml_opt_get_default_optimizer_params(NULL);
1304+
params.optimize.lr = 1e-8;
13031305

13041306
/**
13051307
* filter options by example
@@ -1385,19 +1387,22 @@ common_params_context common_params_parser_init(common_params & params,
13851387
add_opt(common_arg(
13861388
{"--prio"}, "N",
13871389
string_format("set process/thread priority : low(-1), normal(0), medium(1), high(2), realtime(3) (default: %d)\n", params.cpuparams.priority),
1388-
[](common_params & params, int prio) {
1389-
if (prio < 0 || prio > 3) {
1390-
throw std::invalid_argument("invalid value");
1391-
}
1392-
params.cpuparams.priority = (enum ggml_sched_priority) prio;
1393-
}));
1394-
add_opt(common_arg({ "--poll" },
1395-
"<0...100>",
1396-
string_format(
1397-
"use polling level to wait for work (0 - no polling, default: %u)\n", (unsigned) params.cpuparams.poll),
1398-
[](common_params & params, const std::string & value) { params.cpuparams.poll = std::stoul(value); }));
1399-
add_opt(common_arg({ "-Cb", "--cpu-mask-batch" },
1400-
"M",
1390+
[](common_params & params, int prio) {
1391+
if (prio < GGML_SCHED_PRIO_LOW || prio > GGML_SCHED_PRIO_REALTIME) {
1392+
throw std::invalid_argument("invalid value");
1393+
}
1394+
params.cpuparams.priority = (enum ggml_sched_priority) prio;
1395+
}
1396+
));
1397+
add_opt(common_arg(
1398+
{"--poll"}, "<0...100>",
1399+
string_format("use polling level to wait for work (0 - no polling, default: %u)\n", (unsigned) params.cpuparams.poll),
1400+
[](common_params & params, const std::string & value) {
1401+
params.cpuparams.poll = std::stoul(value);
1402+
}
1403+
));
1404+
add_opt(common_arg(
1405+
{"-Cb", "--cpu-mask-batch"}, "M",
14011406
"CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch (default: same as --cpu-mask)",
14021407
[](common_params & params, const std::string & mask) {
14031408
params.cpuparams_batch.mask_valid = true;
@@ -2138,7 +2143,7 @@ common_params_context common_params_parser_init(common_params & params,
21382143
.set_examples({ LLAMA_EXAMPLE_FINETUNE }));
21392144
add_opt(common_arg({ "-opt", "--optimizer" },
21402145
"sgd|adamw",
2141-
"adamw or //TODO:sgd",
2146+
"adamw or sgd",
21422147
[](common_params & params, const std::string & name) {
21432148
params.optimize.optimizer = ggml_opt_get_optimizer(name.c_str());
21442149
if (params.optimize.optimizer == GGML_OPT_OPTIMIZER_COUNT) {

‎common/chat-parser.cpp‎

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -154,9 +154,10 @@ bool common_chat_msg_parser::try_parse_reasoning(const std::string & start_think
154154
if (!rest.empty()) {
155155
handle_reasoning(rest, /* closed */ !is_partial());
156156
}
157-
if (!syntax_.thinking_forced_open) {
158-
throw common_chat_msg_partial_exception(end_think);
159-
}
157+
// Allow unclosed thinking tags, for now (https://github.com/ggml-org/llama.cpp/issues/13812, https://github.com/ggml-org/llama.cpp/issues/13877)
158+
// if (!syntax_.thinking_forced_open) {
159+
// throw common_chat_msg_partial_exception(end_think);
160+
// }
160161
return true;
161162
}
162163
}

‎common/chat-parser.h‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,9 +2,10 @@
22

33
#include "chat.h"
44
#include "json-partial.h"
5-
#include "json.hpp"
65
#include "regex-partial.h"
76

7+
#include <nlohmann/json.hpp>
8+
89
#include <optional>
910
#include <string>
1011
#include <vector>

‎common/chat.cpp‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,13 +1,14 @@
11
#include "chat.h"
22
#include "chat-parser.h"
33
#include "common.h"
4+
#include "json-partial.h"
45
#include "json-schema-to-grammar.h"
56
#include "log.h"
6-
#include "json-partial.h"
7-
#include "minja/chat-template.hpp"
8-
#include "minja/minja.hpp"
97
#include "regex-partial.h"
108

9+
#include <minja/chat-template.hpp>
10+
#include <minja/minja.hpp>
11+
1112
#include <cstdio>
1213
#include <exception>
1314
#include <iostream>
@@ -16,7 +17,6 @@
1617
#include <string>
1718
#include <vector>
1819

19-
2020
static std::string format_time(const std::chrono::system_clock::time_point & now, const std::string & format) {
2121
auto time = std::chrono::system_clock::to_time_t(now);
2222
auto local_time = *std::localtime(&time);

‎common/common.cpp‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -203,6 +203,7 @@ bool set_process_priority(enum ggml_sched_priority prio) {
203203

204204
DWORD p = NORMAL_PRIORITY_CLASS;
205205
switch (prio) {
206+
case GGML_SCHED_PRIO_LOW: p = BELOW_NORMAL_PRIORITY_CLASS; break;
206207
case GGML_SCHED_PRIO_NORMAL: p = NORMAL_PRIORITY_CLASS; break;
207208
case GGML_SCHED_PRIO_MEDIUM: p = ABOVE_NORMAL_PRIORITY_CLASS; break;
208209
case GGML_SCHED_PRIO_HIGH: p = HIGH_PRIORITY_CLASS; break;
@@ -228,6 +229,7 @@ bool set_process_priority(enum ggml_sched_priority prio) {
228229

229230
int p = 0;
230231
switch (prio) {
232+
case GGML_SCHED_PRIO_LOW: p = 5; break;
231233
case GGML_SCHED_PRIO_NORMAL: p = 0; break;
232234
case GGML_SCHED_PRIO_MEDIUM: p = -5; break;
233235
case GGML_SCHED_PRIO_HIGH: p = -10; break;

‎common/json-partial.cpp‎

Lines changed: 5 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,10 @@
1-
#include <json-partial.h>
2-
#include "ggml.h"
1+
#include "json-partial.h"
2+
33
#include "log.h"
4-
#include <string>
54

6-
#include <json.hpp>
5+
#include <nlohmann/json.hpp>
6+
7+
#include <string>
78

89
using json = nlohmann::ordered_json;
910

‎common/json-partial.h‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,6 @@
11
#pragma once
2-
#include <json.hpp>
2+
3+
#include <nlohmann/json.hpp>
34

45
// Healing marker (empty if the JSON was fully parsed / wasn't healed).
56
struct common_healing_marker {

0 commit comments

Comments
 (0)