ipex-llm 2.2.0b20250204__py3-none-win_amd64.whl → 2.2.0b20250205__py3-none-win_amd64.whl

ipex_llm/transformers/low_bit_linear.py

@@ -500,16 +500,16 @@ class MatMulLowBit(torch.autograd.Function):
 
     @staticmethod
     @custom_fwd
-    def forward(ctx, A, weight, input_seq_size):
+    def forward(ctx, A, weight, output_size):
         ctx.is_empty = False
         import xe_linear
         if weight.qtype == NF4:
             result = xe_linear.forward_new(A,
                                            weight.data.view(torch.uint8),
                                            weight.qtype,
-                                           input_seq_size)
+                                           output_size)
         else:
-            result = xe_linear.forward_new(A, weight.data, weight.qtype, input_seq_size)
+            result = xe_linear.forward_new(A, weight.data, weight.qtype, output_size)
         if any(ctx.needs_input_grad[:2]):
             ctx.tensors = (A, weight)
         else:
@@ -627,89 +627,50 @@ class LowBitLinear(nn.Linear):
         if self.optimize_lm_head:
             x = reshape_lm_head_input(x)
 
-        # [batch, input_num, in_len]
-        # input_num == token num for Transformer
-        x_shape = x.shape
-        # Output shape, e.g., [batch, input_num, out_len]
-        new_shape = x_shape[:-1] + (self.out_len,)
+        # [batch, seq_len, in_len] -> [batch, seq_len, out_len]
+        new_shape = x.shape[:-1] + (self.out_len,)
+
         # Activation is empty tensor, e.g., [1, 0, 4096]
-        if 0 in x_shape:
+        if 0 in x.shape:
             # return empty tensor with output shape, x.dtype and x.device
             return torch.empty(new_shape, dtype=x.dtype, device=x.device)
 
-        x_2d = x.contiguous().view(-1, x_shape[-1])
-
         if self.act_order:
-            x_2d = x_2d[:, self.g_idx_map]
-        # x0 for weight
-        x0 = self.weight.data
-
-        if x0.device.type == "xpu":
-            # GPU logic
-            try:
-                import xe_linear
-                from ipex_llm.transformers.models.utils import use_xmx
-            except ModuleNotFoundError:
-                invalidInputError(False,
-                                  "Please `pip install bigdl_core_xe` first.")
+            x = x[..., self.g_idx_map]
 
-            if x_2d.is_contiguous() is False:
-                x_2d = x_2d.contiguous()
+        x_2d = x.contiguous().view(-1, x.shape[-1])
 
-            if len(x_shape) == 3:
-                input_seq_size = x_shape[1]
-            elif len(x_shape) < 3:
-                input_seq_size = 1
-
-            if is_training:
-                # training path
-                if x_2d.requires_grad:
-                    result = MatMulLowBit.apply(x_2d, self.weight, input_seq_size)
-                else:
-                    if self.weight.qtype == NF4:
-                        result = xe_linear.forward_new(x_2d,
-                                                       self.weight.data.view(torch.uint8),
-                                                       self.weight.qtype,
-                                                       input_seq_size)
-                    else:
-                        result = xe_linear.forward_new(x_2d,
-                                                       self.weight.data,
-                                                       self.weight.qtype,
-                                                       input_seq_size)
+        if self.weight.device.type == "xpu":
+            if is_training and x_2d.requires_grad:
+                result = MatMulLowBit.apply(x_2d, self.weight, self.out_len)
             else:
-                # inference path
-                # current workaround to reduce first token latency of fp32 input
-                # sometimes fp16 cause nan and training instability
-                # disable the conversion when training
-                # TODO: may modify the input length condition for empty cache.
                 do_empty_cache = self.low_memory_mode and x_2d.shape[0] >= 1024
                 if do_empty_cache:
                     torch.xpu.empty_cache()
 
+                if self.qtype == NF4:
+                    w = self.weight.data.view(torch.uint8)
+                else:
+                    w = self.weight.data
+
                 if use_batch_forward(x_2d, self.weight.qtype, self.out_len):
                     import xe_batch
-                    result = xe_batch.batch_forward(x_2d, self.weight.data, self.weight.qtype)
-                elif (
-                    self.conver_to_half
-                    and x_2d.shape[0] > 1
-                    and x_2d.dtype == torch.float32
-                    and not use_xmx(x_2d, self.weight.qtype)
-                ):
+                    result = xe_batch.batch_forward(x_2d, w, self.qtype)
+                elif not is_training and self.conver_to_half \
+                        and x_2d.shape[0] > 1 and x_2d.dtype == torch.float:
+                    import xe_linear
                     x_2d = x_2d.half()
-                    result = xe_linear.forward_new(x_2d, self.weight.data,
-                                                   self.weight.qtype, input_seq_size)
+                    result = xe_linear.forward_new(x_2d, w, self.qtype, self.out_len)
                     result = result.to(x.dtype)
                 else:
-                    if self.weight.qtype == NF4:
-                        result = xe_linear.forward_new(x_2d, self.weight.data.view(torch.uint8),
-                                                       self.weight.qtype, input_seq_size)
-                    else:
-                        result = xe_linear.forward_new(x_2d, self.weight.data,
-                                                       self.weight.qtype, input_seq_size)
+                    import xe_linear
+                    result = xe_linear.forward_new(x_2d, w, self.qtype, self.out_len)
 
                 if do_empty_cache:
                     torch.xpu.empty_cache()
+
             result = result.view(new_shape)
+
             if self.mp_group is not None:
                 if get_use_vllm():
                     result = self.mp_group.all_reduce(result)
@@ -718,6 +679,7 @@ class LowBitLinear(nn.Linear):
                     dist.inference_all_reduce(result, group=self.mp_group)
                 else:
                     invalidInputError(False, "mp_group is not None, but no supported backend found")
+
             if self.bias is not None:
                 result += self.bias
         else:
@@ -731,7 +693,7 @@ class LowBitLinear(nn.Linear):
                 result = MatMulLowBitCPU.apply(x, self.weight)
             else:
                 from ipex_llm.utils.isa_checker import is_server, is_spr
-
+                x0 = self.weight.data
                 # convert if necessary, and compute a linear result
                 if is_server() and (not is_spr()) and \
                         self.qtype == SYM_INT4 and x_2d.shape[0] >= TORCH_LINEAR_THRESHOLD:

ipex_llm/transformers/models/utils.py

@@ -259,19 +259,6 @@ def mlp_fusion_check(x, qtype, training):
     return True
 
 
-def use_xmx(x: torch.Tensor, qtype: int):
-    device = get_xpu_device_name(x.device)
-    return (
-        device in ["arc", "pvc"]
-        and qtype in [SYM_INT4, SYM_INT8, FP8E4, FP8E5, WOQ_INT4]
-        and (
-            (device == "pvc" and 1 < x.size(0) <= 16)
-            or
-            (device != "pvc" and 1 < x.size(0) <= 64)
-        )
-    )
-
-
 def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
     batch, num_key_value_heads, slen, head_dim = hidden_states.shape
     if n_rep == 1:

ipex_llm/transformers/xpu_ops.py

@@ -20,9 +20,10 @@ import xe_batch
 import xe_addons
 
 
-# @torch.library.register_fake("ipex_llm::forward_new")
-# def _(x, weight, qtype, input_size):
-#     return ???
+@torch.library.register_fake("ipex_llm::forward_new")
+def _(x, weight, qtype, output_size):
+    return torch.empty([x.size(0), output_size],
+                       dtype=x.dtype, device=x.device)
 
 
 # @torch.library.register_fake("ipex_llm::dequant")

{ipex_llm-2.2.0b20250204.dist-info → ipex_llm-2.2.0b20250205.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250204
+Version: 2.2.0b20250205
 Summary: Large Language Model Develop Toolkit
 Home-page: https://github.com/intel-analytics/ipex-llm
 Author: BigDL Authors
@@ -27,10 +27,10 @@ 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 ==2.6.0b20250204 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250205 ; extra == 'cpp'
 Requires-Dist: setuptools ; extra == 'cpp'
 Provides-Extra: cpp-arl
-Requires-Dist: bigdl-core-cpp ==2.6.0b20250204 ; extra == 'cpp-arl'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250205 ; extra == 'cpp-arl'
 Requires-Dist: setuptools ; extra == 'cpp-arl'
 Requires-Dist: onednn-devel ==2024.1.1 ; (platform_system == "Windows") and extra == 'cpp-arl'
 Requires-Dist: onednn ==2024.1.1 ; (platform_system == "Windows") and extra == 'cpp-arl'
@@ -67,7 +67,7 @@ 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 ==2.6.0b20250204 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.6.0b20250205 ; (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'
@@ -87,9 +87,9 @@ 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 ==2.6.0b20250204 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250204 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250204 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250205 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250205 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250205 ; extra == 'xpu'
 Provides-Extra: xpu-2-1
 Requires-Dist: py-cpuinfo ; extra == 'xpu-2-1'
 Requires-Dist: protobuf ; extra == 'xpu-2-1'
@@ -104,9 +104,9 @@ 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 ==2.6.0b20250204 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250204 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250204 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250205 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250205 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250205 ; 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'
@@ -124,7 +124,7 @@ 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 ==2.6.0b20250204 ; extra == 'xpu-2-6'
+Requires-Dist: bigdl-core-xe-all ==2.6.0b20250205 ; 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'
@@ -140,9 +140,9 @@ 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 ==2.6.0b20250204 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250204 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250204 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250205 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250205 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250205 ; 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'
@@ -163,9 +163,9 @@ 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 ==2.6.0b20250204 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250204 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250204 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250205 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250205 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250205 ; 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'
@@ -186,9 +186,9 @@ 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 ==2.6.0b20250204 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250204 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250204 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250205 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250205 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250205 ; 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-2.2.0b20250204.dist-info → ipex_llm-2.2.0b20250205.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=Liqh-LQ7dD6SJTbEYBvA359vCwptXW1ZkIDRBqV6HrQ,36352
-ipex_llm/libs/bloom.dll,sha256=z8mZQPHnJaTprtO4Li54v4AUUocHrhswbvABd_L6J1Y,507904
-ipex_llm/libs/gptneox-api.dll,sha256=hiet2igvNHeTG3cxpTx3YanuiBPbKQqu6XuahO8raBU,24576
-ipex_llm/libs/gptneox.dll,sha256=BMUCjypqgFHj2NYv4WtTSN5QmLLfwCVwdQjUEBbmtL0,568320
-ipex_llm/libs/libbloom_avx.dll,sha256=20YTK8QrzFsE_WxxDcBAkEV4GFJVFjqYaDXZJ2l7PYo,536576
-ipex_llm/libs/libbloom_vnni.dll,sha256=5kidOCZeFIelGHXIrbGg3UKmczMaB8kJHwv3k6ueJeI,508416
-ipex_llm/libs/libgptneox_avx.dll,sha256=XP0Z5yTD8GQVfoJxvuTRFOBMSWIdXTcEJ_Y-hASE3mQ,596992
-ipex_llm/libs/libgptneox_vnni.dll,sha256=jHxu_arkPjbXF32sMkfM56UGDFqzMAUoSBAj6zZrOrg,568832
-ipex_llm/libs/libllama_avx.dll,sha256=0nC8LsOxILe0iuPNVdJM9nahlerbWyETjD9OAh5A9qo,591360
-ipex_llm/libs/libllama_vnni.dll,sha256=AJd_Yrt-wvJltg2-NfUE0Sfyyhbz38OBvhJXVa31Wac,563200
-ipex_llm/libs/libstarcoder_avx.dll,sha256=2G9rZRohjoXR80p9aXhYd64tSVTWir3C-ZoUpknhAj4,627712
-ipex_llm/libs/libstarcoder_vnni.dll,sha256=lRO9clZiAJNXHLAOEPOsV-29qHyitp9TPI4g8Kolx_8,599552
-ipex_llm/libs/llama-api.dll,sha256=JFXvlgLKkieXTBFqxVT3A39DnRN1jWxopfbpNxztaJA,25600
-ipex_llm/libs/llama.dll,sha256=Tl5YtNoy6vP0pjEdOqzWyFj9n3uUCXn49ywtoZsq8QU,562688
-ipex_llm/libs/main-bloom.exe,sha256=MTmlkYHLdFmKvhZcJUyFre-J9dRJui2oqLMfoby3j2Y,103424
-ipex_llm/libs/main-gptneox.exe,sha256=tv02QCHWEJKwJx9M2H7pmfw2HDA_c1GK_QuSVHrFIzs,98816
-ipex_llm/libs/main-llama.exe,sha256=3SqMTCxXX4rFnq-hlMe587YjHQERtp82QEFnXILMo8A,99840
-ipex_llm/libs/main-starcoder.exe,sha256=aQqfyZ7SW2u6WZ4vCSb-KWVzGEljZNHtdXiL-ErMBW8,157696
-ipex_llm/libs/pipeline.dll,sha256=nRhyK2ZwGB8pDKijNZMCUxSMlD1mR9jVTFKxmxAR2v4,72704
-ipex_llm/libs/quantize-bloom.exe,sha256=vfmlpkCSjhiAJgTyRYLpKvEJySJg58h5SzXCilajiug,126464
-ipex_llm/libs/quantize-bloom_vnni.exe,sha256=9tBmQ8vNQQ_0Ux5YIIqwveeNqSlOjvqmMk12bz0R3gU,128000
-ipex_llm/libs/quantize-gptneox.exe,sha256=9bqrQpmqS6i4wcrPdDPrk0-EeuJsBOLpJm2WsCoDISA,104448
-ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=G-jfbjDWGREcfVpRR90JWkAuwfxVmY3Yji3B-OvlbbU,104960
-ipex_llm/libs/quantize-llama.exe,sha256=JydqEBFh_RkOU_nwHo4vqIL1t37Cy4FvuuivT9s2W4k,110080
-ipex_llm/libs/quantize-llama_vnni.exe,sha256=R00IZwHx1Q5pSnouN6HjhmyRWtkFdqtEb3cA2Qzi03I,110592
-ipex_llm/libs/quantize-starcoder.exe,sha256=ocAyh6uYU9AtjQZADxukNS2Z1PMl8hgim7A1ZUsd1XU,127488
-ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=q2zQhWiaBQPPtu3ceqRcOEGuGdYKL7EsuGtVgSBLt70,128512
-ipex_llm/libs/starcoder-api.dll,sha256=dbdN2Tu242FGudTJJm0eliVL-a5hxOAUtG0eQw9ITRY,21504
-ipex_llm/libs/starcoder.dll,sha256=R1aQDRSWcmp7P8iqJihXs1uXgjpHnQdHqPfQGXoOkdY,599040
+ipex_llm/libs/bloom-api.dll,sha256=TYlavqKZLm9cn6bP0_Or4CwB3Hjz8Yx4pp8S97-AHBs,36352
+ipex_llm/libs/bloom.dll,sha256=wSBD5w8Jerz1w_YZcXgXkuo2Zvb2MAxwfP93Kf07vkU,507904
+ipex_llm/libs/gptneox-api.dll,sha256=faHoG2Cjjw_FGzdzB_yLSJaJkbOr6xh7YVNnF9_Qu0o,24576
+ipex_llm/libs/gptneox.dll,sha256=0vyvMBJ4C0BWzdDiLxVLLAam4E8CSgNKCSnZFM_To6s,568320
+ipex_llm/libs/libbloom_avx.dll,sha256=5lA3evSyudxU5-i5-SfShwIxCAVL0ssHnwLUKxI-vDI,536576
+ipex_llm/libs/libbloom_vnni.dll,sha256=sI5rtSX7dBFVWzJieXKvtlHyDaLXD4cWmhZSTidAgvI,508416
+ipex_llm/libs/libgptneox_avx.dll,sha256=B21Em9BYrn1HTz2O2B5FTY02w0-p9qpevh5N7X6S4QU,596992
+ipex_llm/libs/libgptneox_vnni.dll,sha256=cq3bT4fI_m_bKh67LxQ6lclPX02z2EpYVKAmrmL4_zs,568832
+ipex_llm/libs/libllama_avx.dll,sha256=KlyAxUy8bWTOtyFfzKGJ_AM3JkCYTHMi9fhKNtT_uUU,591360
+ipex_llm/libs/libllama_vnni.dll,sha256=75HWr7FpY3Xs-QtPq5gVkcvJ57KdjK5kTIdQk6xSA1E,563200
+ipex_llm/libs/libstarcoder_avx.dll,sha256=sjyGRSsjjc_GcyZpvB89UhighhQ9wU80zldZsglRjwA,627712
+ipex_llm/libs/libstarcoder_vnni.dll,sha256=F97jMsjqaIjnePj60vxqApcUrKp3-SZxXz4COBGUzvM,599552
+ipex_llm/libs/llama-api.dll,sha256=OnpoPUxtof1EROi67pOWUuxSy6PriboSZalzEiUlCg8,25600
+ipex_llm/libs/llama.dll,sha256=K2aQszJnyEO-JSxwYrxrFmcZV07P1QmTRfRAvXwM2Z8,562688
+ipex_llm/libs/main-bloom.exe,sha256=BGIcVdDx4iqv1RyoeUy84fxmKyD0GqE4LqlVMSMcQg8,103424
+ipex_llm/libs/main-gptneox.exe,sha256=qPpFcf6SJBH-RENZKMUdne3CrXWqfuhuriKmdvmmIso,98816
+ipex_llm/libs/main-llama.exe,sha256=4BO76bHr0rfxHd_aMOsEwrHJWZaOmYnPplmjbx2yxtA,99840
+ipex_llm/libs/main-starcoder.exe,sha256=eEw_9oi2MN1b3Pqz07-ZWj2iT23_B8E7_kpZXNf1E80,157696
+ipex_llm/libs/pipeline.dll,sha256=Zgvo_nVUIniV1LIFPTTOomYivrqVBKteML0Ho8uzuwA,72704
+ipex_llm/libs/quantize-bloom.exe,sha256=xTuVYryKBNIvU-GPakVyx_B1fkST7AgG1al6db1NPrI,126464
+ipex_llm/libs/quantize-bloom_vnni.exe,sha256=s0oXCglgy-t9unR0LUDBPlafr4kuQSHbwg3-1RdekFY,128000
+ipex_llm/libs/quantize-gptneox.exe,sha256=HZHwFQTn_GvK-t-BRNogifb5YsfqaUtTsJSYA2OPP2M,104448
+ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=uBZd6QtfKaoTuwdtr_dnP_8gq_a-3aFbXpS_hT-_okc,104960
+ipex_llm/libs/quantize-llama.exe,sha256=E37KzQH13mrrtw42I1k_SUI9ineOnuarayVLc-PLCMs,110080
+ipex_llm/libs/quantize-llama_vnni.exe,sha256=E8xUhe9Z-YHHbIS4bv9q_6KaJyHLZ0wIavtdNKoBpjQ,110592
+ipex_llm/libs/quantize-starcoder.exe,sha256=zM_szw58gFqVlG-asnXct8c0acWYkrfUegmx0HxHB0w,127488
+ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=ZlbP1fHeLMMBNrZr-ZMa142ctlDWK5eO6i3yE92Xy8I,128512
+ipex_llm/libs/starcoder-api.dll,sha256=UkGT5bPDpDOjdeD3wugMOrTt8cY8dEyUHa7_noZ6M2M,21504
+ipex_llm/libs/starcoder.dll,sha256=n3aruSfoeRcYLia9OREio5bvI2y9JLhhZaoPmjscVw8,599040
 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
@@ -94,7 +94,7 @@ ipex_llm/transformers/kv.py,sha256=k4TU18LlA-Sbq9WNNQnfuzu3RSFBwFhmaV3BcGN5bAo,1
 ipex_llm/transformers/lisa.py,sha256=F5WxbtXQ7RdKulj83h_2DnEIgKiKGZf7zvOmg6QBl2s,3289
 ipex_llm/transformers/loader.py,sha256=AwjV5RpI2t2bedlv7ZhLm8cfd-QJZm5hny-XyjIvdnk,6876
 ipex_llm/transformers/lookup.py,sha256=b6OlZ9OV10R9qeWw8mVryVpDxszkjwLkldvi7GPMJY8,19614
-ipex_llm/transformers/low_bit_linear.py,sha256=mFJRKU60ZVHm-V7gDsJYIz-ryntZi15XhS0eqSUPag4,41136
+ipex_llm/transformers/low_bit_linear.py,sha256=9Z3qQc8j5Kkq4vqrY4gqz8_b2V8hFIhvRm-vdIT8N_4,39123
 ipex_llm/transformers/model.py,sha256=cQJNlAkdfoWmVbWd-TS2hf-Do41mMO9orPvG3FO4Nns,40855
 ipex_llm/transformers/modelling_bigdl.py,sha256=7JpNVMuyq_OmtNUaMFMXdxPWZp2q0QHC02QeA-VTPOw,6709
 ipex_llm/transformers/npu_model.py,sha256=X8ZtvZJpzz64XrSPhUYXXZmdJcbZ9X6G3Vlzw-zgN1Q,39749
@@ -107,7 +107,7 @@ ipex_llm/transformers/streamer.py,sha256=RrVlLblzCOtABRUpaMXAyaMnCGgLUtAi_YesLum
 ipex_llm/transformers/training_patch.py,sha256=oxMkUtqyvqJiprw6dE3skkYfD1HOmUlH9N0hBkbn0G0,10799
 ipex_llm/transformers/utils.py,sha256=JBekwpPD-CyMxt1OzvVsp7tu26pSA4v2mjuaUbqrAgI,16995
 ipex_llm/transformers/xpu_customize_fwd.py,sha256=wFpIhs5F6tkNs8gBOrLxWdhLzO3EDHovVkERPIAoAvg,7611
-ipex_llm/transformers/xpu_ops.py,sha256=vw4cUwvqUqDr45d-WMIkCpM2oiHfjN-VjF0bjMSF4kY,4830
+ipex_llm/transformers/xpu_ops.py,sha256=z95iTtcDQvNyJOvB4A6B_ECTYjHp4A7x-FsssoETOMs,4914
 ipex_llm/transformers/awq/__init__.py,sha256=Du5gu3-eeAkeDO_dEMBTzrDBA66DSN3uL3-rn8WGXQw,875
 ipex_llm/transformers/awq/act.py,sha256=YwomJzOOKwkKtzGrm4L4kwBstBLO1Z8SK4CKi8PSYVQ,2172
 ipex_llm/transformers/awq/awq.py,sha256=cGyRQJWwAEJtOtdSbsBoQ33KX_Ie0pv5OJHC0ACEELE,8861
@@ -174,7 +174,7 @@ ipex_llm/transformers/models/rwkv5.py,sha256=OkRNj1pCAZg1z2Fw-I0DEnxLEdZyPeRSQ6m
 ipex_llm/transformers/models/sd.py,sha256=VvHV5u-0k2MgHu3NL9113hPj7DgfxqctuKzEEeNfRDU,5981
 ipex_llm/transformers/models/stablelm.py,sha256=fj-XtOnR6kggnFUQTMPCOOzolkPztN06WAv8QW-XRnI,7054
 ipex_llm/transformers/models/starcoder2.py,sha256=ONKvD7JCkRM0DI-R56x28QFBJ7CjD5hOZBQ_3WfOcNk,6626
-ipex_llm/transformers/models/utils.py,sha256=Rdn9T4zk6Hz8ybJp6kvlyfPwgHwdxEZ8R4zGtMeozWg,15105
+ipex_llm/transformers/models/utils.py,sha256=qI5ln8SQGTvR_IyxFkoZhefgOErnXUnJrifIyhiqT9c,14753
 ipex_llm/transformers/models/whisper.py,sha256=ju3WP8Eq-KvD7kb3Qy51r4FOfSX3NBxfp5RBcq__gzc,4241
 ipex_llm/transformers/models/yuan.py,sha256=JYAn_ZaSGK0NBJLEIxCACfAq084a66GFJkdd5NbpmMA,7732
 ipex_llm/transformers/npu_models/__init__.py,sha256=ulEUGLjaP48LCrVeury3UxLjXxKzRi0UpSG4bYu-7f8,585
@@ -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-2.2.0b20250204.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
-ipex_llm-2.2.0b20250204.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
-ipex_llm-2.2.0b20250204.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
-ipex_llm-2.2.0b20250204.dist-info/METADATA,sha256=axGzhPAqCjlvETBNA5ONJm7wqp1r6Q-Ac202bfz7lSs,12879
-ipex_llm-2.2.0b20250204.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
-ipex_llm-2.2.0b20250204.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
-ipex_llm-2.2.0b20250204.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
-ipex_llm-2.2.0b20250204.dist-info/RECORD,,
+ipex_llm-2.2.0b20250205.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
+ipex_llm-2.2.0b20250205.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
+ipex_llm-2.2.0b20250205.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
+ipex_llm-2.2.0b20250205.dist-info/METADATA,sha256=dcy4swE9fq7tfGndfg4yAhMDaIs4-0nrRpcDFuAaMFw,12879
+ipex_llm-2.2.0b20250205.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
+ipex_llm-2.2.0b20250205.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
+ipex_llm-2.2.0b20250205.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
+ipex_llm-2.2.0b20250205.dist-info/RECORD,,