bigdl-core-npu 2.6.0b20250206__cp310-cp310-win_amd64.whl → 2.6.0b20250208__cp310-cp310-win_amd64.whl

bigdl-core-npu/include/common.h

@@ -70,7 +70,7 @@ struct npu_model_params {
     std::string config;
     std::string low_bit;
     std::string lm_head_low_bit;
-    bool layernorm_const;
+    bool const_parameter;
     std::string model_type;
     bool transpose_value_cache;
     bool qkv_bias;

bigdl-core-npu/include/llamacpp/common.h

@@ -344,7 +344,7 @@ struct gpt_params {
     bool batched_bench_output_jsonl = false;
 
     // npu convert
-    std::string low_bit = "sym_int4";
+    std::string low_bit = "Q4_0";
     int32_t quantization_group_size = 0;
     int32_t max_context_len = 1024;
     int32_t max_prompt_len = 512;

bigdl-core-npu/include/llamacpp/ggml.h

@@ -2661,9 +2661,19 @@ extern "C" {
 
     GGML_API void ggml_dequantize_to_fp32(struct ggml_tensor * tensor, float *y, size_t n);
 
-    GGML_API void ggml_requantize_to_npu(struct ggml_tensor * tensor, float* y, const char* layer, const int weight_idx, enum gguf_npu_qtype npu_type, const char* dir_path);
+    GGML_API void ggml_requantize_to_npu(struct ggml_tensor * tensor, float* y, const char* layer, const int weight_idx, enum gguf_npu_qtype npu_type, const char* dir_path, int n_splits);
 
     GGML_API void ggml_convert_to_fp16(struct ggml_tensor * tensor, const char* layer, const int weight_idx, const char* dir_path);
+
+    GGML_API size_t ggml_quantize_chunk_with_splits(
+            enum ggml_type   type,
+               const float * src,
+                      void * dst,
+                   int64_t   start,
+                   int64_t   nrows,
+                   int64_t   n_per_row,
+               const float * imatrix,
+                       int   n_splits);
 #ifdef  __cplusplus
 }
 #endif

bigdl-core-npu/include/llamacpp/llama.h

@@ -8,6 +8,7 @@
 #include <stdint.h>
 #include <stdio.h>
 #include <stdbool.h>
+#include <string>
 
 #ifdef LLAMA_SHARED
 #    if defined(_WIN32) && !defined(__MINGW32__)
@@ -310,6 +311,8 @@ extern "C" {
         bool use_mmap;      // use mmap if possible
         bool use_mlock;     // force system to keep model in RAM
         bool check_tensors; // validate model tensor data
+        std::string npu_model;
+        std::string qtype;
     };
 
     // NOTE: changing the default values of parameters marked as [EXPERIMENTAL] may cause crashes or incorrect results in certain configurations
@@ -1217,6 +1220,8 @@ extern "C" {
 
     LLAMA_API void convert_gguf_to_npu_weight(llama_model* model, const char* weight_path, enum gguf_npu_qtype npu_type);
 
+    LLAMA_API std::string get_npu_model_dir(llama_model * model, std::string qtype);
+
 #ifdef __cplusplus
 }
 #endif

{bigdl_core_npu-2.6.0b20250206.dist-info → bigdl_core_npu-2.6.0b20250208.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: bigdl-core-npu
-Version: 2.6.0b20250206
+Version: 2.6.0b20250208
 Summary: Intel® NPU Acceleration Library
 Home-page: https://github.com/intel/intel-npu-acceleration-library
 Author: Alessandro Palla

{bigdl_core_npu-2.6.0b20250206.dist-info → bigdl_core_npu-2.6.0b20250208.dist-info}/RECORD

@@ -1,23 +1,24 @@
 bigdl-core-npu/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-bigdl-core-npu/common.lib,sha256=6gG0mv354_BjuENosPgN94WvkDtEuR-8ZZbxZg6aMcs,6647656
-bigdl-core-npu/ggml.dll,sha256=Mf_ehF2wyf41WratvU7OgL-9KaJjgWjsbJTiwysFa7Q,704000
-bigdl-core-npu/ggml.lib,sha256=U29OuxT-I6Ul4MQA7IaQqq7FN6d8C45Ee8SjjzjEt10,139602
-bigdl-core-npu/llama-cli-npu.exe,sha256=6dTgOLpASlJKJ3gTeLQkqwsPIe-CMnST__ijlg8IzLc,43520
-bigdl-core-npu/llama.dll,sha256=w98eAXeZ26y9vObJaup_IcutigrWREpW2o1xbE9kyxI,1574400
-bigdl-core-npu/llama.lib,sha256=Z2dQy5y6QJ1P2DubIjfo4cB-T6a2ZfAMEC7MDShrX8Q,1494020
-bigdl-core-npu/npu_llm.dll,sha256=9KbQuuK1bIIa6IuEE-pXlDSEVLRHUWKsioi_D0ZTl8c,3956736
+bigdl-core-npu/common.lib,sha256=MeeiDWPtA8cw-rdDsxPDdEqOV5ZdYNwNcP6jzS8Yc5M,6645900
+bigdl-core-npu/ggml.dll,sha256=qoyuaP5i1Gp0yyivaOf5DrdheOa4Lc86VvY3imSzVho,705024
+bigdl-core-npu/ggml.lib,sha256=YiA6rQgJZwvwdM0bXqMHNgAigoEPozcLs7jP3dYz1nk,139880
+bigdl-core-npu/llama-cli-npu.exe,sha256=8oPaOUWSlZzh54bttKjA5qRR9RivC8w5-gwRg-FjyBY,556032
+bigdl-core-npu/llama.dll,sha256=ERfdyRLtgCp7QWl1mcZw2rd08NS8hL71XuWK3Syc5qI,1591808
+bigdl-core-npu/llama.lib,sha256=G-H-iFwvni5O4_Lwjx2GteAE7B9kTC9pMVbM23dndC0,1515984
+bigdl-core-npu/llm-cli.exe,sha256=3lNbGbrjnAAobmq8YtZj_zp4-IwF7LJMyII-Y5J_iRw,44544
+bigdl-core-npu/npu_llm.dll,sha256=JihsqX2Y8QjcAyn77K_oHj2jlrligxveFddX5KO9nek,3962368
 bigdl-core-npu/npu_llm.lib,sha256=nNYF-btjrT9Pzcd31kQV5BsLBSNdA_DSeyK5RnmOc9s,44524
-bigdl-core-npu/include/common.h,sha256=xsTOeUtJPvwuv5F8Wn9Bk67GnvIk3RlMI3DN_W5UBEY,3206
+bigdl-core-npu/include/common.h,sha256=p5PTHTHiJlOA9lXXWH12eHf5JTvGi-mmf9GgFY7kv7E,3206
 bigdl-core-npu/include/npu_llm.h,sha256=siEFqkKnZlzR9cvq7qXiAy5Z65-X14QyWEw__wsLLpA,2605
 bigdl-core-npu/include/llamacpp/arg.h,sha256=XOLwSDqJsCJJgGCkOGSWcA9yGZ0eiiY7GffHdAxFYYY,3110
-bigdl-core-npu/include/llamacpp/common.h,sha256=SaGJ1cAmJWd5D-T_03udb6ADTcJ8VX-bMy0GYGDMpI4,25841
+bigdl-core-npu/include/llamacpp/common.h,sha256=pZ1DuXPcaIFui7juV0kB4-c5CW2I9BRI3Tlz8OT_bsk,25837
 bigdl-core-npu/include/llamacpp/ggml-alloc.h,sha256=kiWITcUF9Q7kvGSBeCfouV59YyGDNy56VE_0fXXnWHE,3088
 bigdl-core-npu/include/llamacpp/ggml-backend.h,sha256=cUJR-AC7hUbIxukKixR9vUSV38gmp-9eoViaBG8lyio,14347
-bigdl-core-npu/include/llamacpp/ggml.h,sha256=_K-zhVjT6uHa5lXMN3k3AwSrwb9W2q1A7v5HiGQc8iQ,105629
-bigdl-core-npu/include/llamacpp/llama.h,sha256=gMTNuRrEY7z0nkCewKLMZtf5urcSzkQm_HxEs3DJXHI,60967
+bigdl-core-npu/include/llamacpp/ggml.h,sha256=liFnnzpemuVWZ3lg7oTCoRIPCo2taBdA8cwMY6OV9FM,106000
+bigdl-core-npu/include/llamacpp/llama.h,sha256=OCMREKob7-NDw3ia9INyJ13bVKUT00gSp9JIayY5FnM,61134
 bigdl-core-npu/include/llamacpp/log.h,sha256=D6UiundA26yZOh3ci828u5U4ZaBxWb9CwSNu_3dt6CA,4295
 intel_npu_acceleration_library/__init__.py,sha256=ZKTIhGMDjF7P6pF-yX8KWcSXbeHWRk24AO_orsa18f8,536
-intel_npu_acceleration_library/_version.py,sha256=l2BonFE20zLZI17uMf5zSIVamDTrunPYZXc0hR7UQns,112
+intel_npu_acceleration_library/_version.py,sha256=4pFkd6f20XRFeIEE7uRcKsrNSMU_fJedu5xEWYA3v6k,112
 intel_npu_acceleration_library/compiler.py,sha256=3IdgqjamSC8MLexDBJypIeZRiWIcTFnvQSU1LPXUr7Y,6225
 intel_npu_acceleration_library/device.py,sha256=9bn8eVXJa5cXIqgfLsQAdkMVtVUQABb8z0-mQik5jRg,7424
 intel_npu_acceleration_library/dtypes.py,sha256=gdd06Wsc9zIZFHlauUEx4xcK9WGTn1Mu6GkuYDJeA-E,4683
@@ -234,7 +235,7 @@ intel_npu_acceleration_library/external/openvino/torch/__init__.py,sha256=RXLzsf
 intel_npu_acceleration_library/functional/__init__.py,sha256=WWKwKOh6Sgovv7mKctA872TbLP98Pg5m5-MREvUmlAA,204
 intel_npu_acceleration_library/functional/scaled_dot_product_attention.py,sha256=yGUcg4tDQOLuUnP1g74cl-ec8TRr2SuAMcNLlN6qLvE,1620
 intel_npu_acceleration_library/lib/Release/cache.json,sha256=CyrSqZUWo0Ec4_7ydOiuKIC0Gm8AybrGdozUqUuHxBw,8840377
-intel_npu_acceleration_library/lib/Release/intel_npu_acceleration_library.dll,sha256=q7r4RIG5hC2w6zuhqlHLgXMbevIU7dZhFxXxesy-UhQ,341504
+intel_npu_acceleration_library/lib/Release/intel_npu_acceleration_library.dll,sha256=eXhnETU-SufRnGBOfeMnYX4tEYSMiNA7uvXDGczHeZc,341504
 intel_npu_acceleration_library/lib/Release/openvino.dll,sha256=m7M119p3JBq2YYJJ2zzCaBDz6XivKK3nNykb8L1cvDU,13244768
 intel_npu_acceleration_library/lib/Release/openvino_auto_batch_plugin.dll,sha256=2v_I9P3Qo0St1bQZMEZscnFOUVvgZQQ0HvQlG3HtTd0,203104
 intel_npu_acceleration_library/lib/Release/openvino_auto_plugin.dll,sha256=e3Aj9CDRHN30dBEdPSk7OCWe52tWfhI4xeXgyFjuDHg,475488
@@ -264,7 +265,7 @@ intel_npu_acceleration_library/nn/functional.py,sha256=UfAKBc0u6RtyaMo14ldH2GpEn
 intel_npu_acceleration_library/nn/linear.py,sha256=Q06SoGQeLaI86nA_ky2GnFC6H2Fw1zyMDILKnpYC2eo,5739
 intel_npu_acceleration_library/nn/llm.py,sha256=P6dz36Yf6BHtzWcftaghC6QaMI_WeRfQwrCbO7fD6hk,15002
 intel_npu_acceleration_library/nn/module.py,sha256=EYxoTq6I_YgBDgTF76GPDxHrT8SupOTDGMzQaomBeq8,12667
-bigdl_core_npu-2.6.0b20250206.dist-info/METADATA,sha256=AvzIyOhFVctFq9nq_McYX_Fuib3-aqtLQeQcE8xmxZY,1762
-bigdl_core_npu-2.6.0b20250206.dist-info/WHEEL,sha256=rzGfZgUcGeKSgIHGYMuqg4xE4VPHxnaldXH6BG0zjVk,101
-bigdl_core_npu-2.6.0b20250206.dist-info/top_level.txt,sha256=iMQZlTsFPJjlD-Y0MqZEP_9ifI0LlbNCJIOTaMoGMjk,46
-bigdl_core_npu-2.6.0b20250206.dist-info/RECORD,,
+bigdl_core_npu-2.6.0b20250208.dist-info/METADATA,sha256=0eKBbcpupCbACu_wuUmjO8kgF0Uj0-C39YnxzYKopAI,1762
+bigdl_core_npu-2.6.0b20250208.dist-info/WHEEL,sha256=rzGfZgUcGeKSgIHGYMuqg4xE4VPHxnaldXH6BG0zjVk,101
+bigdl_core_npu-2.6.0b20250208.dist-info/top_level.txt,sha256=iMQZlTsFPJjlD-Y0MqZEP_9ifI0LlbNCJIOTaMoGMjk,46
+bigdl_core_npu-2.6.0b20250208.dist-info/RECORD,,

intel_npu_acceleration_library/_version.py

@@ -3,4 +3,4 @@
 # SPDX-License-Identifier: Apache 2.0
 #
 
-__version__ = "2.6.0b20250206"
+__version__ = "2.6.0b20250208"