Skip to content

Commit fe6b72b

Browse files
gabe-l-hartgabe-l-hart
committed
Merge remote-tracking branch 'origin/master' into gabe-l-hart/thinking-model-disabled-agent-prefill
* origin/master: (84 commits) CUDA: fastdiv, launch bounds for mmvq + q8_1 quant (ggml-org#15802) tests : add --list-ops and --show-coverage options (ggml-org#15745) gguf: gguf_writer refactor (ggml-org#15691) kv-cache : fix SWA checks + disable cacheless iSWA (ggml-org#15811) model-conversion : add --embeddings flag to modelcard.template [no ci] (ggml-org#15801) chat : fixed crash when Hermes 2 <tool_call> had a newline before it (ggml-org#15639) chat : nemotron thinking & toolcalling support (ggml-org#15676) scripts : add Jinja tester PySide6 simple app (ggml-org#15756) llama : add support for EmbeddingGemma 300m (ggml-org#15798) metal : Add template specialization for mul_mm_id w/ ne20 == 10 (ggml-org#15799) llama : set n_outputs to 1 to avoid 0 outputs mean-pooling (ggml-org#15791) CANN: Refactor ND to NZ workspace to be per-device (ggml-org#15763) server: add exceed_context_size_error type (ggml-org#15780) Document the new max GPU layers default in help (ggml-org#15771) ggml: add ops for WAN video model (cuda && cpu) (ggml-org#15669) CANN: Fix precision issue on 310I DUO multi-devices (ggml-org#15784) opencl: add hs=40 to FA (ggml-org#15758) CANN: fix acl_rstd allocation size in ggml_cann_rms_norm (ggml-org#15760) vulkan: fix mmv subgroup16 selection (ggml-org#15775) vulkan: don't use std::string in load_shaders, to improve compile time (ggml-org#15724) ...
2 parents e2a19d2 + 5143fa8 commit fe6b72b

File tree

142 files changed

+7178
-1767
lines changed

Some content is hidden

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

142 files changed

+7178
-1767
lines changed

‎.clang-format‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -22,7 +22,7 @@ AllowShortIfStatementsOnASingleLine: Never
2222
AllowShortLambdasOnASingleLine: Inline
2323
AllowShortLoopsOnASingleLine: false
2424
AlwaysBreakBeforeMultilineStrings: true
25-
BinPackArguments: false
25+
BinPackArguments: true
2626
BinPackParameters: false # OnePerLine
2727
BitFieldColonSpacing: Both
2828
BreakBeforeBraces: Custom # Attach

‎README.md‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -137,6 +137,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
137137
- [X] [Trillion-7B-preview](https://huggingface.co/trillionlabs/Trillion-7B-preview)
138138
- [x] [Ling models](https://huggingface.co/collections/inclusionAI/ling-67c51c85b34a7ea0aba94c32)
139139
- [x] [LFM2 models](https://huggingface.co/collections/LiquidAI/lfm2-686d721927015b2ad73eaa38)
140+
- [x] [Hunyuan models](https://huggingface.co/collections/tencent/hunyuan-dense-model-6890632cda26b19119c9c5e7)
140141

141142
#### Multimodal
142143

‎ci/run.sh‎

Lines changed: 10 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -386,10 +386,10 @@ function gg_run_open_llama_7b_v2 {
386386

387387
(time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
388388

389-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
390-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
391-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
392-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
389+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 -fa off ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
390+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 -fa on ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
391+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa off ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
392+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa on ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
393393

394394
function check_ppl {
395395
qnt="$1"
@@ -520,8 +520,8 @@ function gg_run_pythia_1_4b {
520520

521521
(time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
522522

523-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
524-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
523+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa off ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
524+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa on ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
525525

526526
function check_ppl {
527527
qnt="$1"
@@ -651,10 +651,10 @@ function gg_run_pythia_2_8b {
651651

652652
(time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
653653

654-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
655-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
656-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
657-
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
654+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 -fa off ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
655+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 -fa on ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
656+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa off ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
657+
(time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa on ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
658658

659659
function check_ppl {
660660
qnt="$1"

‎common/arg.cpp‎

Lines changed: 22 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -1545,10 +1545,18 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
15451545
}
15461546
).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_PERPLEXITY, LLAMA_EXAMPLE_RETRIEVAL}));
15471547
add_opt(common_arg(
1548-
{"-fa", "--flash-attn"},
1549-
string_format("enable Flash Attention (default: %s)", params.flash_attn ? "enabled" : "disabled"),
1550-
[](common_params & params) {
1551-
params.flash_attn = true;
1548+
{"-fa", "--flash-attn"}, "FA",
1549+
string_format("set Flash Attention use ('on', 'off', or 'auto', default: '%s')", llama_flash_attn_type_name(params.flash_attn_type)),
1550+
[](common_params & params, const std::string & value) {
1551+
if (value == "on" || value == "enabled" || value == "1") {
1552+
params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_ENABLED;
1553+
} else if (value == "off" || value == "disabled" || value == "0") {
1554+
params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_DISABLED;
1555+
} else if (value == "auto" || value == "-1") {
1556+
params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_AUTO;
1557+
} else {
1558+
throw std::runtime_error(string_format("error: unkown value for --flash-attn: '%s'\n", value.c_str()));
1559+
}
15521560
}
15531561
).set_env("LLAMA_ARG_FLASH_ATTN"));
15541562
add_opt(common_arg(
@@ -2458,7 +2466,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
24582466
).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_N_CPU_MOE_DRAFT"));
24592467
add_opt(common_arg(
24602468
{"-ngl", "--gpu-layers", "--n-gpu-layers"}, "N",
2461-
"number of layers to store in VRAM",
2469+
string_format("max. number of layers to store in VRAM (default: %d)", params.n_gpu_layers),
24622470
[](common_params & params, int value) {
24632471
params.n_gpu_layers = value;
24642472
if (!llama_supports_gpu_offload()) {
@@ -2555,15 +2563,15 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
25552563
{"--lora"}, "FNAME",
25562564
"path to LoRA adapter (can be repeated to use multiple adapters)",
25572565
[](common_params & params, const std::string & value) {
2558-
params.lora_adapters.push_back({ std::string(value), 1.0, nullptr });
2566+
params.lora_adapters.push_back({ std::string(value), 1.0, "", "", nullptr });
25592567
}
25602568
// we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
25612569
).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));
25622570
add_opt(common_arg(
25632571
{"--lora-scaled"}, "FNAME", "SCALE",
25642572
"path to LoRA adapter with user defined scaling (can be repeated to use multiple adapters)",
25652573
[](common_params & params, const std::string & fname, const std::string & scale) {
2566-
params.lora_adapters.push_back({ fname, std::stof(scale), nullptr });
2574+
params.lora_adapters.push_back({ fname, std::stof(scale), "", "", nullptr });
25672575
}
25682576
// we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
25692577
).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));
@@ -2954,20 +2962,20 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
29542962
params.endpoint_metrics = true;
29552963
}
29562964
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_METRICS"));
2957-
add_opt(common_arg(
2958-
{"--slots"},
2959-
string_format("enable slots monitoring endpoint (default: %s)", params.endpoint_slots ? "enabled" : "disabled"),
2960-
[](common_params & params) {
2961-
params.endpoint_slots = true;
2962-
}
2963-
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_SLOTS"));
29642965
add_opt(common_arg(
29652966
{"--props"},
29662967
string_format("enable changing global properties via POST /props (default: %s)", params.endpoint_props ? "enabled" : "disabled"),
29672968
[](common_params & params) {
29682969
params.endpoint_props = true;
29692970
}
29702971
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_PROPS"));
2972+
add_opt(common_arg(
2973+
{"--slots"},
2974+
string_format("enable slots monitoring endpoint (default: %s)", params.endpoint_slots ? "enabled" : "disabled"),
2975+
[](common_params & params) {
2976+
params.endpoint_slots = true;
2977+
}
2978+
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_SLOTS"));
29712979
add_opt(common_arg(
29722980
{"--no-slots"},
29732981
"disables slots monitoring endpoint",
@@ -3459,8 +3467,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
34593467
params.model.hf_repo = "ggml-org/Qwen2.5-Coder-1.5B-Q8_0-GGUF";
34603468
params.model.hf_file = "qwen2.5-coder-1.5b-q8_0.gguf";
34613469
params.port = 8012;
3462-
params.n_gpu_layers = 99;
3463-
params.flash_attn = true;
34643470
params.n_ubatch = 1024;
34653471
params.n_batch = 1024;
34663472
params.n_ctx = 0;
@@ -3475,8 +3481,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
34753481
params.model.hf_repo = "ggml-org/Qwen2.5-Coder-3B-Q8_0-GGUF";
34763482
params.model.hf_file = "qwen2.5-coder-3b-q8_0.gguf";
34773483
params.port = 8012;
3478-
params.n_gpu_layers = 99;
3479-
params.flash_attn = true;
34803484
params.n_ubatch = 1024;
34813485
params.n_batch = 1024;
34823486
params.n_ctx = 0;
@@ -3491,8 +3495,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
34913495
params.model.hf_repo = "ggml-org/Qwen2.5-Coder-7B-Q8_0-GGUF";
34923496
params.model.hf_file = "qwen2.5-coder-7b-q8_0.gguf";
34933497
params.port = 8012;
3494-
params.n_gpu_layers = 99;
3495-
params.flash_attn = true;
34963498
params.n_ubatch = 1024;
34973499
params.n_batch = 1024;
34983500
params.n_ctx = 0;
@@ -3508,10 +3510,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
35083510
params.model.hf_file = "qwen2.5-coder-7b-q8_0.gguf";
35093511
params.speculative.model.hf_repo = "ggml-org/Qwen2.5-Coder-0.5B-Q8_0-GGUF";
35103512
params.speculative.model.hf_file = "qwen2.5-coder-0.5b-q8_0.gguf";
3511-
params.speculative.n_gpu_layers = 99;
35123513
params.port = 8012;
3513-
params.n_gpu_layers = 99;
3514-
params.flash_attn = true;
35153514
params.n_ubatch = 1024;
35163515
params.n_batch = 1024;
35173516
params.n_ctx = 0;
@@ -3527,10 +3526,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
35273526
params.model.hf_file = "qwen2.5-coder-14b-q8_0.gguf";
35283527
params.speculative.model.hf_repo = "ggml-org/Qwen2.5-Coder-0.5B-Q8_0-GGUF";
35293528
params.speculative.model.hf_file = "qwen2.5-coder-0.5b-q8_0.gguf";
3530-
params.speculative.n_gpu_layers = 99;
35313529
params.port = 8012;
3532-
params.n_gpu_layers = 99;
3533-
params.flash_attn = true;
35343530
params.n_ubatch = 1024;
35353531
params.n_batch = 1024;
35363532
params.n_ctx = 0;
@@ -3545,8 +3541,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
35453541
params.model.hf_repo = "ggml-org/Qwen3-Coder-30B-A3B-Instruct-Q8_0-GGUF";
35463542
params.model.hf_file = "qwen3-coder-30b-a3b-instruct-q8_0.gguf";
35473543
params.port = 8012;
3548-
params.n_gpu_layers = 99;
3549-
params.flash_attn = true;
35503544
params.n_ubatch = 1024;
35513545
params.n_batch = 1024;
35523546
params.n_ctx = 0;

0 commit comments

Comments
 (0)