@fugood/llama.node 1.3.6 → 1.3.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/package.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "@fugood/llama.node",
3
3
  "access": "public",
4
- "version": "1.3.6",
4
+ "version": "1.3.7",
5
5
  "description": "An another Node binding of llama.cpp",
6
6
  "main": "lib/index.js",
7
7
  "scripts": {
@@ -72,20 +72,20 @@
72
72
  "CMakeLists.txt"
73
73
  ],
74
74
  "optionalDependencies": {
75
- "@fugood/node-llama-linux-x64": "1.3.6",
76
- "@fugood/node-llama-linux-x64-vulkan": "1.3.6",
77
- "@fugood/node-llama-linux-x64-cuda": "1.3.6",
78
- "@fugood/node-llama-linux-arm64-snapdragon": "1.3.6",
79
- "@fugood/node-llama-linux-arm64": "1.3.6",
80
- "@fugood/node-llama-linux-arm64-vulkan": "1.3.6",
81
- "@fugood/node-llama-linux-arm64-cuda": "1.3.6",
82
- "@fugood/node-llama-win32-x64": "1.3.6",
83
- "@fugood/node-llama-win32-x64-vulkan": "1.3.6",
84
- "@fugood/node-llama-win32-x64-cuda": "1.3.6",
85
- "@fugood/node-llama-win32-arm64": "1.3.6",
86
- "@fugood/node-llama-win32-arm64-vulkan": "1.3.6",
87
- "@fugood/node-llama-darwin-x64": "1.3.6",
88
- "@fugood/node-llama-darwin-arm64": "1.3.6"
75
+ "@fugood/node-llama-linux-x64": "1.3.7",
76
+ "@fugood/node-llama-linux-x64-vulkan": "1.3.7",
77
+ "@fugood/node-llama-linux-x64-cuda": "1.3.7",
78
+ "@fugood/node-llama-linux-arm64-snapdragon": "1.3.7",
79
+ "@fugood/node-llama-linux-arm64": "1.3.7",
80
+ "@fugood/node-llama-linux-arm64-vulkan": "1.3.7",
81
+ "@fugood/node-llama-linux-arm64-cuda": "1.3.7",
82
+ "@fugood/node-llama-win32-x64": "1.3.7",
83
+ "@fugood/node-llama-win32-x64-vulkan": "1.3.7",
84
+ "@fugood/node-llama-win32-x64-cuda": "1.3.7",
85
+ "@fugood/node-llama-win32-arm64": "1.3.7",
86
+ "@fugood/node-llama-win32-arm64-vulkan": "1.3.7",
87
+ "@fugood/node-llama-darwin-x64": "1.3.7",
88
+ "@fugood/node-llama-darwin-arm64": "1.3.7"
89
89
  },
90
90
  "devDependencies": {
91
91
  "@babel/preset-env": "^7.24.4",
@@ -122,18 +122,10 @@ index d0cab0bcb..48d532838 100644
122
122
  check_cxx_compiler_flag(-mfp16-format=ieee GGML_COMPILER_SUPPORTS_FP16_FORMAT_I3E)
123
123
  if (NOT "${GGML_COMPILER_SUPPORTS_FP16_FORMAT_I3E}" STREQUAL "")
124
124
  diff --git a/src/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp b/src/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp
125
- index cabd301ad..31eec134c 100644
125
+ index 72a82a891..7869ad323 100644
126
126
  --- a/src/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp
127
127
  +++ b/src/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp
128
- @@ -9,6 +9,7 @@
129
- #include <chrono>
130
- #include <mutex>
131
- #include <string>
132
- +#include <stdexcept>
133
-
134
- #ifdef _WIN32
135
- # include <sal.h>
136
- @@ -3682,6 +3683,8 @@ ggml_hexagon_registry::ggml_hexagon_registry(ggml_backend_reg_t reg) {
128
+ @@ -3417,6 +3417,8 @@ ggml_hexagon_registry::ggml_hexagon_registry(ggml_backend_reg_t reg) {
137
129
  } catch (std::exception const &exc) {
138
130
  GGML_LOG_ERROR("ggml-hex: failed to create device/session %zu\n", i);
139
131
  devices[i].context = nullptr;
@@ -1593,7 +1593,8 @@ void llama_model::load_hparams(llama_model_loader & ml) {
1593
1593
  } break;
1594
1594
  case LLM_ARCH_DEEPSEEK2:
1595
1595
  {
1596
- bool is_lite = (hparams.n_layer == 27);
1596
+ // lite variants include DeepSeek-V2-Lite, GigaChat3-10B-A1.8B
1597
+ bool is_lite = (hparams.n_layer == 27 || hparams.n_layer == 26);
1597
1598
  ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
1598
1599
  ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT, hparams.n_layer_dense_lead);
1599
1600
  if (!is_lite) {
@@ -4581,7 +4582,8 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
4581
4582
  } break;
4582
4583
  case LLM_ARCH_DEEPSEEK2:
4583
4584
  {
4584
- const bool is_lite = (hparams.n_layer == 27);
4585
+ // lite variants include DeepSeek-V2-Lite, GigaChat3-10B-A1.8B
4586
+ const bool is_lite = (hparams.n_layer == 27 || hparams.n_layer == 26);
4585
4587
 
4586
4588
  const bool is_mla = (hparams.n_embd_head_k_mla != 0 && hparams.n_embd_head_v_mla != 0);
4587
4589
 
@@ -4,7 +4,8 @@
4
4
 
5
5
  llm_build_deepseek2::llm_build_deepseek2(const llama_model & model, const llm_graph_params & params) :
6
6
  llm_graph_context(params) {
7
- bool is_lite = (hparams.n_layer == 27);
7
+ // lite variants include DeepSeek-V2-Lite, GigaChat3-10B-A1.8B
8
+ bool is_lite = (hparams.n_layer == 27 || hparams.n_layer == 26);
8
9
 
9
10
  const bool is_mla = (hparams.n_embd_head_k_mla != 0 && hparams.n_embd_head_v_mla != 0);
10
11