ipex-llm 2.2.0b20250207__py3-none-win_amd64.whl → 2.2.0b20250208__py3-none-win_amd64.whl

ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py

@@ -201,7 +201,7 @@ def convert_llm(model: torch.nn.Module,
                 keep_ir: bool=False,
                 compile_blob: bool=True):
     # whether to set layernorm weight as const
-    layernorm_const = os.environ.get("IPEX_LLM_NPU_LAYERNORM_CONST", "1") == "1"
+    const_parameter = os.environ.get("IPEX_LLM_NPU_CONST_PARAMETER", "1") == "1"
     if group_size == 0:
         n_splits_linear = 1
         if qtype in ["sym_int8_rtn", "asym_int4_rtn"]:
@@ -240,7 +240,7 @@ def convert_llm(model: torch.nn.Module,
             for layer_idx in range(0, layer_num):
                 param_list.append((model, layer_idx, n_splits_linear, n_splits_down_proj,
                                    temp_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                                   layernorm_const))
+                                   const_parameter))
             with Pool() as pool:
                 result = pool.starmap(convert_llama_layer, param_list)
 
@@ -267,7 +267,7 @@ def convert_llm(model: torch.nn.Module,
                 res = InitLLMPipeline(model_type, kv_len, model.num_head, model.head_dim, layer_num,
                                       model.vocab_size, weight_dir, "model",
                                       first_blob_path, last_blob_path,
-                                      os.path.join(temp_dir, "decoder_layer"), layernorm_const)
+                                      os.path.join(temp_dir, "decoder_layer"), const_parameter)
             except:
                 invalidInputError(False,
                                   "False to InitLLMPipeline.")
@@ -284,7 +284,7 @@ def convert_llm(model: torch.nn.Module,
             for layer_idx in range(0, layer_num):
                 param_list.append((model, layer_idx, n_splits_linear, n_splits_down_proj,
                                   temp_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                                  layernorm_const))
+                                  const_parameter))
             with Pool() as pool:
                 result = pool.starmap(convert_baichuan_layer, param_list)
 
@@ -308,7 +308,7 @@ def convert_llm(model: torch.nn.Module,
                 res = InitLLMPipeline("baichuan", kv_len, model.num_head, model.head_dim, layer_num,
                                       model.vocab_size, weight_dir, "model",
                                       first_blob_path, last_blob_path,
-                                      os.path.join(temp_dir, "decoder_layer"), layernorm_const)
+                                      os.path.join(temp_dir, "decoder_layer"), const_parameter)
             except:
                 invalidInputError(False,
                                   "False to InitLLMPipeline.")
@@ -325,7 +325,7 @@ def convert_llm(model: torch.nn.Module,
             for layer_idx in range(0, layer_num):
                 param_list.append((model, layer_idx, n_splits_linear, n_splits_down_proj,
                                    temp_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                                   layernorm_const))
+                                   const_parameter))
             with Pool() as pool:
                 result = pool.starmap(convert_minicpm_layer, param_list)
 
@@ -348,12 +348,12 @@ def convert_llm(model: torch.nn.Module,
                 res = InitLLMPipeline("minicpm", kv_len, model.num_head, model.head_dim, layer_num,
                                       model.vocab_size, weight_dir, "model",
                                       first_blob_path, last_blob_path,
-                                      os.path.join(temp_dir, "decoder_layer"), layernorm_const)
+                                      os.path.join(temp_dir, "decoder_layer"), const_parameter)
             except:
                 invalidInputError(False,
                                   "False to InitLLMPipeline.")
     elif model.config.model_type == "qwen2":
-        layernorm_const = os.environ.get("IPEX_LLM_NPU_LAYERNORM_CONST", "0") == "1"
+        const_parameter = os.environ.get("IPEX_LLM_NPU_CONST_PARAMETER", "0") == "1"
         with tempfile.TemporaryDirectory() as temp_dir:
             if save_directory is not None:
                 temp_dir = save_directory
@@ -371,7 +371,7 @@ def convert_llm(model: torch.nn.Module,
             for layer_idx in range(0, layer_num):
                 param_list.append((model, layer_idx, n_splits_linear, n_splits_down_proj,
                                   temp_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                                  layernorm_const))
+                                  const_parameter))
             with Pool() as pool:
                 result = pool.starmap(convert_qwen_layer, param_list)
 
@@ -396,7 +396,7 @@ def convert_llm(model: torch.nn.Module,
                                "head_dim": model.head_dim,
                                "transpose_value_cache": transpose_value_cache,
                                "max_prompt_len": max_prompt_len,
-                               "layernorm_const": layernorm_const,
+                               "const_parameter": const_parameter,
                                "group_size":  group_size}
                 model.config.update(update_dict)
                 model.config.save_pretrained(save_directory)
@@ -405,7 +405,7 @@ def convert_llm(model: torch.nn.Module,
                 res = InitLLMPipeline("qwen", kv_len, model.num_head, model.head_dim, layer_num,
                                       model.vocab_size, weight_dir, "model",
                                       first_blob_path, last_blob_path,
-                                      os.path.join(temp_dir, "decoder_layer"), layernorm_const)
+                                      os.path.join(temp_dir, "decoder_layer"), const_parameter)
             except:
                 invalidInputError(False,
                                   "False to InitLLMPipeline.")
@@ -441,7 +441,9 @@ def convert_llm_for_deploy(model: torch.nn.Module,
     weight_dir = os.path.join(save_directory, "model_weights")
     if not os.path.exists(weight_dir):
         os.mkdir(weight_dir)
-    layernorm_const = os.environ.get("IPEX_LLM_NPU_LAYERNORM_CONST", "1") == "1"
+    const_parameter = os.environ.get("IPEX_LLM_NPU_CONST_PARAMETER", "1") == "1"
+    if keep_ir:
+        const_parameter = False
 
     lm_head_low_bit = getattr(model.config, "bigdl_transformers_low_bit", "sym_int4_rtn")
     if hasattr(model, "lm_head") and not isinstance(model.lm_head, SlicedLMHead):
@@ -472,7 +474,7 @@ def convert_llm_for_deploy(model: torch.nn.Module,
                        "head_dim": model.model.layers[0].self_attn.head_dim,
                        "transpose_value_cache": transpose_value_cache,
                        "max_prompt_len": max_prompt_len,
-                       "layernorm_const": layernorm_const,
+                       "const_parameter": const_parameter,
                        "group_size":  group_size,
                        "fused_layers": fused_layers,
                        "qkv_bias": True,
@@ -490,12 +492,12 @@ def convert_llm_for_deploy(model: torch.nn.Module,
         # save fused_layers blobs of fused decoder layers
         convert_fused_qwen_layer(model, fused_layers, n_splits_linear, n_splits_down_proj,
                                  save_directory, weight_dir, transpose_value_cache, kv_len,
-                                 group_size, layernorm_const, "decode",
+                                 group_size, const_parameter, "decode",
                                  keep_ir=keep_ir, compile_blob=compile_blob)
         # save blob of single prefill layer
         convert_qwen_layer(model, 0, n_splits_linear, n_splits_down_proj,
                            save_directory, weight_dir, transpose_value_cache, max_prompt_len,
-                           group_size, layernorm_const, "prefill",
+                           group_size, const_parameter, "prefill",
                            keep_ir=keep_ir, compile_blob=compile_blob)
         # save blob of lmhead and bin of embedding
         convert_lm_head_and_embedding(model, save_directory, weight_dir, convert_model=True,
@@ -535,7 +537,7 @@ def convert_llm_for_deploy(model: torch.nn.Module,
                        "head_dim": model.model.layers[0].self_attn.head_dim,
                        "transpose_value_cache": transpose_value_cache,
                        "max_prompt_len": max_prompt_len,
-                       "layernorm_const": layernorm_const,
+                       "const_parameter": const_parameter,
                        "group_size":  group_size,
                        "fused_layers": fused_layers,
                        "qkv_bias": False,
@@ -559,12 +561,12 @@ def convert_llm_for_deploy(model: torch.nn.Module,
         # save fused_layers blobs of fused decoder layers
         convert_fused_llama_layer(model, fused_layers, n_splits_linear, n_splits_down_proj,
                                   save_directory, weight_dir, transpose_value_cache, kv_len,
-                                  group_size, layernorm_const, "decode",
+                                  group_size, const_parameter, "decode",
                                   keep_ir=keep_ir, compile_blob=compile_blob)
         # save blob of single prefill layer
         convert_llama_layer(model, 0, n_splits_linear, n_splits_down_proj,
                             save_directory, weight_dir, transpose_value_cache, max_prompt_len,
-                            group_size, layernorm_const, "prefill",
+                            group_size, const_parameter, "prefill",
                             keep_ir=keep_ir, compile_blob=compile_blob)
     elif model.config.model_type == "minicpm":
         if group_size == 0:
@@ -576,7 +578,7 @@ def convert_llm_for_deploy(model: torch.nn.Module,
                        "head_dim": model.model.layers[0].self_attn.head_dim,
                        "transpose_value_cache": transpose_value_cache,
                        "max_prompt_len": max_prompt_len,
-                       "layernorm_const": layernorm_const,
+                       "const_parameter": const_parameter,
                        "group_size":  group_size,
                        "fused_layers": fused_layers,
                        "qkv_bias": False,
@@ -594,12 +596,12 @@ def convert_llm_for_deploy(model: torch.nn.Module,
         # save fused_layers blobs of fused decoder layers
         convert_fused_minicpm_layer(model, fused_layers, n_splits_linear, n_splits_down_proj,
                                     save_directory, weight_dir, transpose_value_cache, kv_len,
-                                    group_size, layernorm_const, "decode",
+                                    group_size, const_parameter, "decode",
                                     keep_ir=keep_ir, compile_blob=compile_blob)
         # save blob of single prefill layer
         convert_minicpm_layer(model, 0, n_splits_linear, n_splits_down_proj,
                               save_directory, weight_dir, transpose_value_cache, max_prompt_len,
-                              group_size, layernorm_const, "prefill",
+                              group_size, const_parameter, "prefill",
                               keep_ir=keep_ir, compile_blob=compile_blob)
         # save blob of lmhead and bin of embedding and embedding_post
         convert_lm_head_and_embedding(model, n_splits_linear,

ipex_llm/transformers/npu_pipeline_model/llama.py

@@ -107,7 +107,7 @@ def convert_lm_head_and_embedding(model, n_splits_linear, temp_dir, weight_dir,
 
 def convert_llama_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
                         temp_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                        layernorm_const, mode="decode",
+                        const_parameter, mode="decode",
                         keep_ir=False, compile_blob=True):
     num_heads = model.model.layers[0].self_attn.num_heads
     num_key_value_heads = model.model.layers[0].self_attn.num_key_value_heads
@@ -145,14 +145,14 @@ def convert_llama_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
     else:
         input_len = kv_len
         decoder_name = "decoder_layer_prefill"
-        layernorm_const = False
+        const_parameter = False
         keep_position_ids = False
         npu_dpu_groups = 6
 
     single_decoder = LowBitLlamaMultiDecoderlayer(
         [1, input_len, num_heads * head_dim],
-        input_layernorm_weights=[layer_norm_0] if layernorm_const else None,
-        post_attn_layernorm_weights=[layer_norm_1] if layernorm_const else None,
+        input_layernorm_weights=[layer_norm_0] if const_parameter else None,
+        post_attn_layernorm_weights=[layer_norm_1] if const_parameter else None,
         cached_cos=cached_cos,
         cached_sin=cached_sin,
         num_heads=num_heads,
@@ -182,7 +182,7 @@ def convert_llama_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
     if mode == "decode":
         if hasattr(curr_layer.self_attn.rotary_emb, "cos_cached"):
             # llama-2-7B & llama-3-8B
-            if layernorm_const:
+            if const_parameter:
                 st_idx = 5
             else:
                 input_lm_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_3.bin")
@@ -192,7 +192,7 @@ def convert_llama_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
                 st_idx = 7
         else:
             # llama-3.2-3B & llama-3.2-1B
-            if layernorm_const:
+            if const_parameter:
                 st_idx = 6
             else:
                 input_lm_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_4.bin")
@@ -223,7 +223,7 @@ def convert_llama_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
 
 def convert_fused_llama_layer(model, fused_layers, n_splits_linear, n_splits_down_proj,
                               save_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                              layernorm_const, mode="decode",
+                              const_parameter, mode="decode",
                               keep_ir=False, compile_blob=True):
     num_heads = model.model.layers[0].self_attn.num_heads
     num_key_value_heads = model.model.layers[0].self_attn.num_key_value_heads
@@ -294,6 +294,10 @@ def convert_fused_llama_layer(model, fused_layers, n_splits_linear, n_splits_dow
         else:  # FP16 Linear
             np_dtype = np.float16
 
+        if not const_parameter:
+            input_layer_norm_weights = None
+            post_attn_layernorm_weights = None
+
         fused_decoder = LowBitLlamaMultiDecoderlayer(
             [1, 1, num_heads * head_dim],
             input_layernorm_weights=input_layer_norm_weights,

ipex_llm/transformers/npu_pipeline_model/minicpm.py

@@ -301,7 +301,7 @@ def convert_lm_head_and_embedding(model, n_splits_linear, temp_dir, weight_dir,
 
 def convert_minicpm_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
                           temp_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                          layernorm_const, mode="decode",
+                          const_parameter, mode="decode",
                           keep_ir=False, compile_blob=True):
     num_heads = model.model.layers[0].self_attn.num_heads
     num_key_value_heads = model.model.layers[0].self_attn.num_key_value_heads
@@ -333,12 +333,12 @@ def convert_minicpm_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
     else:
         input_len = kv_len
         decoder_name = "decoder_layer_prefill"
-        layernorm_const = False
+        const_parameter = False
 
     single_decoder = LowBitMinicpmMultiDecoderlayer(
         [1, input_len, num_heads * head_dim],
-        input_layernorm_weights=[layer_norm_0] if layernorm_const else None,
-        post_attn_layernorm_weights=[layer_norm_1] if layernorm_const else None,
+        input_layernorm_weights=[layer_norm_0] if const_parameter else None,
+        post_attn_layernorm_weights=[layer_norm_1] if const_parameter else None,
         cached_cos=cached_cos,
         cached_sin=cached_sin,
         num_heads=num_heads,
@@ -364,7 +364,7 @@ def convert_minicpm_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
     os.remove(os.path.join(temp_dir, decoder_name + ".bin"))
 
     if mode == "decode":
-        if layernorm_const:
+        if const_parameter:
             st_idx = 5
         else:
             input_lm_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_3.bin")
@@ -394,7 +394,7 @@ def convert_minicpm_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
 
 def convert_fused_minicpm_layer(model, fused_layers, n_splits_linear, n_splits_down_proj,
                                 save_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                                layernorm_const, mode="decode",
+                                const_parameter, mode="decode",
                                 keep_ir=False, compile_blob=True):
     num_heads = model.model.layers[0].self_attn.num_heads
     num_key_value_heads = model.model.layers[0].self_attn.num_key_value_heads
@@ -461,6 +461,10 @@ def convert_fused_minicpm_layer(model, fused_layers, n_splits_linear, n_splits_d
         else:  # FP16 Linear
             np_dtype = np.float16
 
+        if not const_parameter:
+            input_layer_norm_weights = None
+            post_attn_layernorm_weights = None
+
         fused_decoder = LowBitMinicpmMultiDecoderlayer(
             [1, 1, num_heads * head_dim],
             input_layernorm_weights=input_layer_norm_weights,

ipex_llm/transformers/npu_pipeline_model/qwen.py

@@ -117,7 +117,7 @@ def convert_lm_head_and_embedding(model, temp_dir, weight_dir,
 
 def convert_qwen_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
                        temp_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                       layernorm_const, mode="decode",
+                       const_parameter, mode="decode",
                        keep_ir=False, compile_blob=True):
     num_heads = model.model.layers[0].self_attn.num_heads
     num_key_value_heads = model.model.layers[0].self_attn.num_key_value_heads
@@ -193,7 +193,7 @@ def convert_qwen_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
     # 0, 1, 2 are input_embed/attention_mask/position_id
     if mode == "decode":
         if hasattr(curr_layer.self_attn.rotary_emb, "cos_cached"):
-            if layernorm_const:
+            if const_parameter:
                 st_idx = 3
             else:
                 input_lm_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_3.bin")
@@ -203,7 +203,7 @@ def convert_qwen_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
                 st_idx = 5
         else:
             # transformers >= 4.45.0
-            if layernorm_const:
+            if const_parameter:
                 st_idx = 4
             else:
                 input_lm_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_4.bin")
@@ -241,7 +241,7 @@ def convert_qwen_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
 
 def convert_fused_qwen_layer(model, fused_layers, n_splits_linear, n_splits_down_proj,
                              save_dir, weight_dir, transpose_value_cache, kv_len, group_size,
-                             layernorm_const, mode="decode",
+                             const_parameter, mode="decode",
                              keep_ir=False, compile_blob=True):
     num_heads = model.model.layers[0].self_attn.num_heads
     num_key_value_heads = model.model.layers[0].self_attn.num_key_value_heads
@@ -325,6 +325,13 @@ def convert_fused_qwen_layer(model, fused_layers, n_splits_linear, n_splits_down
         else:  # FP16 Linear
             np_dtype = np.float16
 
+        if not const_parameter:
+            input_layer_norm_weights = None
+            post_attn_layernorm_weights = None
+            q_biases = None
+            k_biases = None
+            v_biases = None
+
         fused_decoder = LowBitQwenMultiDecoderlayer(
             [1, 1, num_heads * head_dim],
             input_layernorm_weights=input_layer_norm_weights,

{ipex_llm-2.2.0b20250207.dist-info → ipex_llm-2.2.0b20250208.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250207
+Version: 2.2.0b20250208
 Summary: Large Language Model Develop Toolkit
 Home-page: https://github.com/intel-analytics/ipex-llm
 Author: BigDL Authors
@@ -27,7 +27,7 @@ Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine
 Requires-Dist: torch ==2.1.2+cpu ; (platform_system == "Linux") and extra == 'all'
 Requires-Dist: torch ==2.1.2 ; (platform_system == "Windows") and extra == 'all'
 Provides-Extra: cpp
-Requires-Dist: bigdl-core-cpp ==2.6.0b20250207 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250208 ; extra == 'cpp'
 Requires-Dist: setuptools ; extra == 'cpp'
 Requires-Dist: onednn-devel ==2025.0.1 ; (platform_system == "Windows") and extra == 'cpp'
 Requires-Dist: onednn ==2025.0.1 ; (platform_system == "Windows") and extra == 'cpp'
@@ -60,7 +60,7 @@ Requires-Dist: transformers ==4.40.0 ; extra == 'npu'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'npu'
 Requires-Dist: torch ==2.1.2+cpu ; (platform_system == "Linux") and extra == 'npu'
 Requires-Dist: torch ==2.1.2 ; (platform_system == "Windows") and extra == 'npu'
-Requires-Dist: bigdl-core-npu ==2.6.0b20250207 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.6.0b20250208 ; (platform_system == "Windows") and extra == 'npu'
 Provides-Extra: serving
 Requires-Dist: py-cpuinfo ; extra == 'serving'
 Requires-Dist: fschat[model_worker,webui] ==0.2.36 ; extra == 'serving'
@@ -80,9 +80,9 @@ Requires-Dist: setuptools <70.0.0 ; extra == 'xpu'
 Requires-Dist: torch ==2.1.0a0 ; extra == 'xpu'
 Requires-Dist: torchvision ==0.16.0a0 ; extra == 'xpu'
 Requires-Dist: intel-extension-for-pytorch ==2.1.10+xpu ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250207 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250207 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250207 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250208 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250208 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250208 ; extra == 'xpu'
 Provides-Extra: xpu-2-1
 Requires-Dist: py-cpuinfo ; extra == 'xpu-2-1'
 Requires-Dist: protobuf ; extra == 'xpu-2-1'
@@ -97,9 +97,9 @@ Requires-Dist: setuptools <70.0.0 ; extra == 'xpu-2-1'
 Requires-Dist: torch ==2.1.0a0 ; extra == 'xpu-2-1'
 Requires-Dist: torchvision ==0.16.0a0 ; extra == 'xpu-2-1'
 Requires-Dist: intel-extension-for-pytorch ==2.1.10+xpu ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250207 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250207 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250207 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250208 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250208 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250208 ; extra == 'xpu-2-1'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'xpu-2-1'
 Requires-Dist: dpcpp-cpp-rt ==2024.0.2 ; (platform_system == "Windows") and extra == 'xpu-2-1'
 Requires-Dist: mkl-dpcpp ==2024.0.0 ; (platform_system == "Windows") and extra == 'xpu-2-1'
@@ -117,7 +117,7 @@ Requires-Dist: setuptools ; extra == 'xpu-2-6'
 Requires-Dist: torch ==2.6.0+xpu ; extra == 'xpu-2-6'
 Requires-Dist: torchvision ==0.21.0+xpu ; extra == 'xpu-2-6'
 Requires-Dist: torchaudio ==2.6.0+xpu ; extra == 'xpu-2-6'
-Requires-Dist: bigdl-core-xe-all ==2.6.0b20250207 ; extra == 'xpu-2-6'
+Requires-Dist: bigdl-core-xe-all ==2.6.0b20250208 ; extra == 'xpu-2-6'
 Requires-Dist: onednn-devel ==2025.0.1 ; extra == 'xpu-2-6'
 Requires-Dist: onednn ==2025.0.1 ; extra == 'xpu-2-6'
 Requires-Dist: dpcpp-cpp-rt ==2025.0.2 ; extra == 'xpu-2-6'
@@ -133,9 +133,9 @@ Requires-Dist: tokenizers ==0.15.2 ; extra == 'xpu-arc'
 Requires-Dist: accelerate ==0.23.0 ; extra == 'xpu-arc'
 Requires-Dist: tabulate ; extra == 'xpu-arc'
 Requires-Dist: setuptools ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250207 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250207 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250207 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250208 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250208 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250208 ; extra == 'xpu-arc'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'xpu-arc'
 Requires-Dist: torch ==2.3.1+cxx11.abi ; (platform_system == "Linux") and extra == 'xpu-arc'
 Requires-Dist: torchvision ==0.18.1+cxx11.abi ; (platform_system == "Linux") and extra == 'xpu-arc'
@@ -156,9 +156,9 @@ Requires-Dist: tokenizers ==0.15.2 ; extra == 'xpu-arl'
 Requires-Dist: accelerate ==0.23.0 ; extra == 'xpu-arl'
 Requires-Dist: tabulate ; extra == 'xpu-arl'
 Requires-Dist: setuptools ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250207 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250207 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250207 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250208 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250208 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250208 ; extra == 'xpu-arl'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'xpu-arl'
 Requires-Dist: torch ==2.3.1+cxx11.abi ; (platform_system == "Linux") and extra == 'xpu-arl'
 Requires-Dist: torchvision ==0.18.1+cxx11.abi ; (platform_system == "Linux") and extra == 'xpu-arl'
@@ -179,9 +179,9 @@ Requires-Dist: tokenizers ==0.15.2 ; extra == 'xpu-lnl'
 Requires-Dist: accelerate ==0.23.0 ; extra == 'xpu-lnl'
 Requires-Dist: tabulate ; extra == 'xpu-lnl'
 Requires-Dist: setuptools ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250207 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250207 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250207 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250208 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250208 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250208 ; extra == 'xpu-lnl'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'xpu-lnl'
 Requires-Dist: torch ==2.3.1+cxx11.abi ; (platform_system == "Linux") and extra == 'xpu-lnl'
 Requires-Dist: torchvision ==0.18.1+cxx11.abi ; (platform_system == "Linux") and extra == 'xpu-lnl'

{ipex_llm-2.2.0b20250207.dist-info → ipex_llm-2.2.0b20250208.dist-info}/RECORD

@@ -41,35 +41,35 @@ ipex_llm/langchain/llms/transformerspipelinellm.py,sha256=vm522YPPwWxxAPVvQBtxRf
 ipex_llm/langchain/vllm/__init__.py,sha256=T-EbRT6GJ_8RCu-iLmSzcftOimXSPQf2d5X72AUAy2Y,874
 ipex_llm/langchain/vllm/vllm.py,sha256=6dxc-ZISZQrJilEa_HA827l75Dv9rcHpY_G6FdJ8BVs,7793
 ipex_llm/libs/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ipex_llm/libs/bloom-api.dll,sha256=R0zcv1M0D8y8inrrCUO2xCSTRb0IChVyLa6YQo9zne8,36352
-ipex_llm/libs/bloom.dll,sha256=eBzUhLMeOAb9InMPp9_KC5VhJC9F-YKNlJn6HyfOAb0,507904
-ipex_llm/libs/gptneox-api.dll,sha256=9_mq8IntnMiU7-_kDxiLojnEc1nu3rrxZZAIes7Nd4k,24576
-ipex_llm/libs/gptneox.dll,sha256=kR3dyhN7tNUxVIWoqudW57V0MIGqr-Mxkmw7kwR8VWs,568320
-ipex_llm/libs/libbloom_avx.dll,sha256=0iRHd_QIzEG_NI0RkFKmCX_HG-3E21t33sxrmbCpQwo,536576
-ipex_llm/libs/libbloom_vnni.dll,sha256=dL1TzKzoki8KDsCmka6QfzBH24T06WokxT3F4M5a3lk,508416
-ipex_llm/libs/libgptneox_avx.dll,sha256=SPi9xXxB5jLp63CfgVhmMA-rCoyCCji2nuWz-rv5y3E,596992
-ipex_llm/libs/libgptneox_vnni.dll,sha256=NV3xykgHJGxNTDWAA_yhwlBG_dbHPX0__5s9uHCPmfc,568832
-ipex_llm/libs/libllama_avx.dll,sha256=EbZ-lpHHtM-zS9aiuDU8cBVueVAtRi3UqerARH41qC8,591360
-ipex_llm/libs/libllama_vnni.dll,sha256=67XqNSyXI1nuaA1-xcSOhYIHZaH7aZBvwMetGpTriIk,563200
-ipex_llm/libs/libstarcoder_avx.dll,sha256=kAqXHfoZfmyqIbNbGpzQjXNCMz9pkG5KVRECzEDEwhM,627712
-ipex_llm/libs/libstarcoder_vnni.dll,sha256=c02B9jpBvST282jRXJtkRwJKkZnzhkz5MLdFfjH9T8I,599552
-ipex_llm/libs/llama-api.dll,sha256=SA2frHXocsnAN9z3LZfWT_FjY1waSMS26bHM6ot_07c,25600
-ipex_llm/libs/llama.dll,sha256=Ls7CKimo2SNy-uJt6lLz16yz1O9E358dRgP8E0svF98,562688
-ipex_llm/libs/main-bloom.exe,sha256=-HCik31DRGrozp_Uy420O1l-Sk_7e9V1bjg4XaLPFvA,103424
-ipex_llm/libs/main-gptneox.exe,sha256=pqxQCGKBrsoDtvuKhCwk6uOAGt4GGvzoAdQbHB9qrFI,98816
-ipex_llm/libs/main-llama.exe,sha256=sPKj3WRmI97jyNhO4A5Lz4eF-tsZZojv6z2VaNzAKAU,99840
-ipex_llm/libs/main-starcoder.exe,sha256=7vyW8v2qO1J_fkRq4uzk44UsV4AhDGmcWHUwMiez8WY,157696
-ipex_llm/libs/pipeline.dll,sha256=vHFtLO6vUZQVwtzXICv1Q5Ork32Dw5Ipqa8pbr6TtmM,72704
-ipex_llm/libs/quantize-bloom.exe,sha256=8rUxXU7Z4AZ7mFHI3sGpwGG18_DkapunwTzzUTjCCbo,126464
-ipex_llm/libs/quantize-bloom_vnni.exe,sha256=gA9kKUkmFOIzT_CmFFvG-fG6d6bZuEWSTeyPvhCsDLs,128000
-ipex_llm/libs/quantize-gptneox.exe,sha256=YsrviyLjQU9uxD1p6TfdBAPXG72-QzZFGpt7lDmK_gM,104448
-ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=mYmUHza3rZztjTogXv9FxuIM20z0gHfyjbF6b6ADEK0,104960
-ipex_llm/libs/quantize-llama.exe,sha256=h-7nbo0uIswViTdxf_vHmE3sZdnQ79dDMUHzqjtyMKs,110080
-ipex_llm/libs/quantize-llama_vnni.exe,sha256=OEPzGySIaa-O9IhPY-u2slHnhMDzp6mL8e_Qr2WUgKc,110592
-ipex_llm/libs/quantize-starcoder.exe,sha256=4U-jT0MC4Iz4kP_6WpKkMOSk_hTlqAwgSVlGLGa-imA,127488
-ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=Xc4jW9KH_RNSfJIYJinDRIx-BbWmqxx4h-kc9jowZpk,128512
-ipex_llm/libs/starcoder-api.dll,sha256=2lF73SE1AyICwtpQSZUfkiAbE1WJQ5gEbikL1Lsvzhg,21504
-ipex_llm/libs/starcoder.dll,sha256=NBh51OQS90ppaqMAJAFCa6HptcUnnPx7tUL1J95QwMk,599040
+ipex_llm/libs/bloom-api.dll,sha256=6JzFpTxSCIDgyHljiTDIwiyj8tNjyIimr00FUg18eTw,36352
+ipex_llm/libs/bloom.dll,sha256=2W5NmA0RyQMElLCyunvANpTUBTnzahAtlFwf1IQYb_k,507904
+ipex_llm/libs/gptneox-api.dll,sha256=jlJcF2GR9ZHenAOcbcKirLRejWsdiQDhUpWbl5ulVz8,24576
+ipex_llm/libs/gptneox.dll,sha256=8f1rs4aF7OGVtIjZZv-v1bSamG6C_b68G_O2sd_dzAU,568320
+ipex_llm/libs/libbloom_avx.dll,sha256=RJ-jYk9S6Jd34ktr1B3FbEL3MuPyngfekS2HLHc3nHE,536576
+ipex_llm/libs/libbloom_vnni.dll,sha256=LCvnLsZ2uisySM4nL8cDDggRC_6Bw3OFkP-2vikhTVQ,508416
+ipex_llm/libs/libgptneox_avx.dll,sha256=aDdomyYNf9unY6S9ix6CuX2ICY6uphptRm294nRm9AU,596992
+ipex_llm/libs/libgptneox_vnni.dll,sha256=GMlR-de0HcAdw8YJzcXX1JndXgv3PmZ4nipoqyeh7ls,568832
+ipex_llm/libs/libllama_avx.dll,sha256=Y7AJTriTvqWJH8hIA6gwYsL2u_uGA35NmRPoSQSkn78,591360
+ipex_llm/libs/libllama_vnni.dll,sha256=XGd5lij9PZu63CBadK3R6mhkfqpi21wRrU9eGFpReaM,563200
+ipex_llm/libs/libstarcoder_avx.dll,sha256=ZzZFh1IW_QlHrNcE79vSsqPcA9YZSDn1LCEpCwtv0oo,627712
+ipex_llm/libs/libstarcoder_vnni.dll,sha256=dF3mLHFXKAe8TiTb2HqIbjvkQoCVK2BpRwbohFSDZ_Y,599552
+ipex_llm/libs/llama-api.dll,sha256=hg0_pZC8RjgFuTtc19NF5BMLTkchmj_X4FogUC_YeDA,25600
+ipex_llm/libs/llama.dll,sha256=3q1FkqwWE3Fo_-zZ74kWR-IOM7g-SSiSYg3Ud3wmkRo,562688
+ipex_llm/libs/main-bloom.exe,sha256=tdNzTM-5XqiWe1rLTV1qslHWUNi0VKcAWJdjaRKEAzg,103424
+ipex_llm/libs/main-gptneox.exe,sha256=XtmysEP3FoLLtJz5qTIoFPaG4MZhTAZ-0Gru65bk_7Y,98816
+ipex_llm/libs/main-llama.exe,sha256=r7ZrLzVb2_phMKmD50dz5xnrPC8H9H_IPK0T8js8x2E,99840
+ipex_llm/libs/main-starcoder.exe,sha256=udG6tPP_-3aCkFzfi8WxGD6lrauKPxwlaFhv3OF-gDg,157696
+ipex_llm/libs/pipeline.dll,sha256=_CoiiDG12fYEEeIJyqh-dAi9bK3am421KItj548mlek,73216
+ipex_llm/libs/quantize-bloom.exe,sha256=DU-zbDfk4ZZ5N8pQWDM39Xqoy_4rSsYGLhvJdZmunVQ,126464
+ipex_llm/libs/quantize-bloom_vnni.exe,sha256=trXZ2CtNiCkD1_mfvXWJf7IBgNaYRxk3sNzsYNbwBgQ,128000
+ipex_llm/libs/quantize-gptneox.exe,sha256=ZDLyAS9FaMXAZiQrb805iKHY-cR43Z-E453tS-C__bE,104448
+ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=alab9U5EUVg07XwQyNdXAbjp8p3IZ_0sP56IMkpXjTU,104960
+ipex_llm/libs/quantize-llama.exe,sha256=ttvTdoGd6pWVdgA4KR9A_B_Tdg1Gt8qN1mBQrZp7JIw,110080
+ipex_llm/libs/quantize-llama_vnni.exe,sha256=6ECTUDfKLzj-G_yvohaR8z5DSPt6wwnDq2uaRwXdCDc,110592
+ipex_llm/libs/quantize-starcoder.exe,sha256=ApQMB9yxcjqUomqgjnD23rzo3IidKgG_UHXwEAQ-4EA,127488
+ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=6U5MeFLPGtJrkJA6q4HE3NyDUzcmlpBz9WpwLBXS684,128512
+ipex_llm/libs/starcoder-api.dll,sha256=fVkYZ0zZ8mhERgN0hILyFt6cUpe__Y8oBW2Dq1ePinE,21504
+ipex_llm/libs/starcoder.dll,sha256=eOv12oBIIYpxBIoId8g_BuhlIlRZ32r29yDtNQQEMAg,599040
 ipex_llm/llamaindex/__init__.py,sha256=T-EbRT6GJ_8RCu-iLmSzcftOimXSPQf2d5X72AUAy2Y,874
 ipex_llm/llamaindex/llms/__init__.py,sha256=KP1lEdGqDuxPoxL1ZSH25Pm2kKMPJBWUTLR0ckSLMIU,1139
 ipex_llm/llamaindex/llms/bigdlllm.py,sha256=FQBzq1KOjfc6uofTXAha3O7TqpJkNfOFepXQmOVlbnI,26314
@@ -208,11 +208,11 @@ ipex_llm/transformers/npu_models/xlm_mp.py,sha256=sj8OVun8xJprM7ZJp0XzWa55rqlSIz
 ipex_llm/transformers/npu_pipeline_model/__init__.py,sha256=b2IXvVqQ5cItki021h8s3ymW12RPu8QNPprq4Mn3bDM,586
 ipex_llm/transformers/npu_pipeline_model/baichuan.py,sha256=ICxRzFQ4OIANDkkVi2_4xOeQXmfFXYMx3H52KuE1xR4,6208
 ipex_llm/transformers/npu_pipeline_model/common.py,sha256=faooJmM75qnVyZYuQLx9gJpVlotcVF4qXRCnOrknfk4,14776
-ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py,sha256=_l4RFmyBMbREo8vzKpHXAMtE202JVQ41Y2lPg1qCOMI,29846
-ipex_llm/transformers/npu_pipeline_model/llama.py,sha256=j2sipfFSrzV2VgLKPOClMHwWIDXqDsL1jIQJK25hneo,14397
-ipex_llm/transformers/npu_pipeline_model/minicpm.py,sha256=H7j_UaHj-IwEBriQ-bunle0-8s2NmvqnL9eYuixnmFc,21398
+ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py,sha256=IlvaZC9pi_ZJTWuO2dMSgCkc3V909lVhpAfktTEfSLI,29894
+ipex_llm/transformers/npu_pipeline_model/llama.py,sha256=ISJ84zj0Ce2f9vdgmoGfdxQz2LRaUMlZCEM3MV2VpoQ,14521
+ipex_llm/transformers/npu_pipeline_model/minicpm.py,sha256=Q-rUzXBlx5Ns3xemi7H6t8dnzu1q4e-MhUpJMzJmBRU,21522
 ipex_llm/transformers/npu_pipeline_model/pipeline_cpp.py,sha256=JNmodAMg_NQvDILug3E_fGXEh6cd3wsj4bvAzcd-vaU,2749
-ipex_llm/transformers/npu_pipeline_model/qwen.py,sha256=6MNtCL1CXoR19B4tKZSgv2e5gtma9bqDG7DOYMCnPt0,16013
+ipex_llm/transformers/npu_pipeline_model/qwen.py,sha256=WEMUdGZH3INyJm-1Hfv3o41BiP037n2ftS4qPM0jaiE,16221
 ipex_llm/utils/__init__.py,sha256=LlUgrD03rfw4iY8zWPtHH6p65Gw76waVOLHaqagETw0,1425
 ipex_llm/utils/benchmark_util_4_29.py,sha256=OU1W1quiaiJGsg1pd3HM9O6PmVSaPA0HHE7R8hNTfmQ,258653
 ipex_llm/utils/benchmark_util_4_42.py,sha256=HEiClCgKDp_T64HH8ulSTly8dvt6UwPDYZfrPVYvXcc,225383
@@ -248,11 +248,11 @@ ipex_llm/vllm/xpu/engine/__init__.py,sha256=pY_CpyuZd72fr6s32ejeKHKFW0K4vUU2rzZj
 ipex_llm/vllm/xpu/engine/engine.py,sha256=k4-D27WS_Gk3mA--w3HWAjPjb4Aiu043MVPi0ZoAUBc,5984
 ipex_llm/vllm/xpu/entrypoints/openai/api_server.py,sha256=GshTZFB8e4PWvqckfbmTOU6b0oLkNn7A-vzLuG9--j8,21544
 ipex_llm/vllm/xpu/entrypoints/openai/cli_args.py,sha256=2rENA2ucynMaIjiZBEh2ez1o5vR32GaP514t39CD7KM,8676
-ipex_llm-2.2.0b20250207.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
-ipex_llm-2.2.0b20250207.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
-ipex_llm-2.2.0b20250207.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
-ipex_llm-2.2.0b20250207.dist-info/METADATA,sha256=d1hx5hE5Xeb3lHGWqeF35SK9GZOX6syXJ_Syu5b35IU,12369
-ipex_llm-2.2.0b20250207.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
-ipex_llm-2.2.0b20250207.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
-ipex_llm-2.2.0b20250207.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
-ipex_llm-2.2.0b20250207.dist-info/RECORD,,
+ipex_llm-2.2.0b20250208.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
+ipex_llm-2.2.0b20250208.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
+ipex_llm-2.2.0b20250208.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
+ipex_llm-2.2.0b20250208.dist-info/METADATA,sha256=lvf2U7DribCbFo-qabprOhWsiYrX4yLUPfykdJZ91Dk,12369
+ipex_llm-2.2.0b20250208.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
+ipex_llm-2.2.0b20250208.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
+ipex_llm-2.2.0b20250208.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
+ipex_llm-2.2.0b20250208.dist-info/RECORD,,