ipex-llm 2.2.0b20250106.post1__py3-none-manylinux2010_x86_64.whl → 2.2.0b20250108__py3-none-manylinux2010_x86_64.whl

ipex_llm/transformers/convert.py

@@ -847,18 +847,9 @@ def replace_with_low_bit_linear_for_module(model, qtype, module_name=None,
                         mp_group=mp_group,
                     )
                     device = module.weight.data.device
-                    from ipex_llm.transformers.utils import get_ipex_version
-                    if get_ipex_version() < "2.1.10+xpu":
-                        new_linear._parameters['weight'] = nn.Parameter(module.weight)
-                    else:
-                        # only from 2.1, ipex provides matmul_bias_out
-                        # so we need to transpose weight
-                        new_weight = module.weight.transpose(0, 1).contiguous()
-                        new_linear._parameters['weight'] = nn.Parameter(new_weight)
-                        new_linear.weight_type = 2
+                    new_linear._parameters['weight'] = nn.Parameter(module.weight)
                     if module.bias is not None:
-                        new_linear._parameters['bias'] = nn.Parameter(module.bias.data)\
-                            .to(device)
+                        new_linear._parameters['bias'] = nn.Parameter(module.bias.data).to(device)
                 elif qtype == ggml_tensor_qtype["bf16"]:
                     module.to(torch.bfloat16)
                     new_linear = BF16Linear(
@@ -1420,6 +1411,7 @@ def _optimize_post(model):
         convert_forward(model, module.GlmRMSNorm, rms_norm_forward)
         convert_forward(model, module.GlmMLP, mlp_silu_forward)
         convert_forward(model, module.GlmAttention, glm_attention_forward)
+        convert_forward(model, module.GlmSdpaAttention, glm_attention_forward)
         glm_model_forward = glm_model_forward_wrapper(module.GlmModel.forward)
         convert_forward(model, module.GlmModel, glm_model_forward)
 
@@ -1428,10 +1420,12 @@ def _optimize_post(model):
             vision_module_name = model.model.vision.__class__.__module__
             vision_module = importlib.import_module(vision_module_name)
             from transformers.models.siglip.modeling_siglip import SiglipAttention
+            from transformers.models.siglip.modeling_siglip import SiglipSdpaAttention
             from ipex_llm.transformers.models.chatglm4v import vision_model_forward
             from ipex_llm.transformers.models.minicpmv import siglip_attention_forward
             convert_forward(model, vision_module.VisionModel, vision_model_forward)
             convert_forward(model, SiglipAttention, siglip_attention_forward)
+            convert_forward(model, SiglipSdpaAttention, siglip_attention_forward)
 
     elif "mpt" in model.config.model_type:
         if model.config.architectures is not None:
@@ -1667,8 +1661,10 @@ def _optimize_post(model):
         convert_forward(model, module.Qwen2MLP, qwen2_mlp_forward)
         model.visual.get_dtype = MethodType(qwen2_vision_get_dtype, model.visual)
         convert_forward(model, module.VisionAttention, qwen2_vision_attention_forward)
+        convert_forward(model, module.VisionSdpaAttention, qwen2_vision_attention_forward)
         convert_forward(model, module.Qwen2VLModel, qwen2_vl_model_forward)
         convert_forward(model, module.Qwen2VLAttention, qwen2_vl_attention_forward)
+        convert_forward(model, module.Qwen2VLSdpaAttention, qwen2_vl_attention_forward)
     elif model.config.model_type == "aquila":
         modeling_module_name = model.__class__.__module__
         module = importlib.import_module(modeling_module_name)
@@ -1814,6 +1810,7 @@ def _optimize_post(model):
         from ipex_llm.transformers.models.starcoder2 import attention_forward
         from ipex_llm.transformers.models.starcoder2 import model_forward
         convert_forward(model, module.Starcoder2Attention, attention_forward)
+        convert_forward(model, module.Starcoder2SdpaAttention, attention_forward)
         convert_forward(model, module.Starcoder2Model, model_forward)
     elif model.config.model_type == "phi":
         # for phi-2
@@ -1829,6 +1826,7 @@ def _optimize_post(model):
         module = importlib.import_module(modeling_module_name)
         from ipex_llm.transformers.models.phi3 import attention_forward
         convert_forward(model, module.Phi3Attention, attention_forward)
+        convert_forward(model, module.Phi3SdpaAttention, attention_forward)
         from ipex_llm.transformers.models.phi3 import mlp_forward
         convert_forward(model, module.Phi3MLP, mlp_forward)
         from ipex_llm.transformers.models.common import rms_norm_forward
@@ -1872,6 +1870,8 @@ def _optimize_post(model):
                         module.StableLmAttention,
                         stablelm_attention_forward
                         )
+        if hasattr(module, "StableLmSdpaAttention"):
+            convert_forward(model, module.StableLmSdpaAttention, stablelm_attention_forward)
         convert_forward(model,
                         module.StableLmMLP,
                         mlp_silu_forward)
@@ -1886,6 +1886,7 @@ def _optimize_post(model):
         from ipex_llm.transformers.models.minicpm import minicpm_model_forward_wrapper
         from ipex_llm.transformers.models.minicpm import minicpm_decoder_layer_forward
         convert_forward(model, module.MiniCPMAttention, minicpm_attention_forward)
+        convert_forward(model, module.MiniCPMSdpaAttention, minicpm_attention_forward)
         convert_forward(model, module.MiniCPMMLP, mlp_silu_forward)
         convert_forward(model, module.MiniCPMRMSNorm, rms_norm_forward)
         convert_forward(model, module.MiniCPMDecoderLayer, minicpm_decoder_layer_forward)
@@ -1901,6 +1902,7 @@ def _optimize_post(model):
         convert_forward(model, module.MiniCPMRMSNorm, rms_norm_forward)
         convert_forward(model, module.MiniCPMMLP, mlp_silu_forward)
         convert_forward(model, module.MiniCPMAttention, minicpm3_attention_forward)
+        convert_forward(model, module.MiniCPMSdpaAttention, minicpm3_attention_forward)
         minicpm3_model_forward = minicpm3_model_forward_wrapper(module.MiniCPM3Model.forward)
         convert_forward(model, module.MiniCPM3Model, minicpm3_model_forward)
     elif model.config.model_type == "minicpmv":

ipex_llm/transformers/low_bit_linear.py

@@ -51,8 +51,7 @@ from torch import Tensor, device, dtype, nn
 from operator import mul
 from functools import reduce
 from ipex_llm.transformers.xpu_customize_fwd import custom_fwd, custom_bwd
-from ipex_llm.transformers.utils import get_autocast_dtype, get_xpu_device_name, \
-    get_ipex_version
+from ipex_llm.transformers.utils import get_autocast_dtype, get_xpu_device_name
 from ipex_llm.transformers.convert import is_deepspeed_available, get_use_vllm
 
 T = TypeVar("T", bound="torch.nn.Module")

ipex_llm/transformers/models/chatglm4v.py

@@ -301,6 +301,7 @@ def patch_embedding_forward(self, images: "tensor(B, C, H, W)") -> "tensor(B, L,
 
 def merge_qkv(module: torch.nn.Module):
     merge_qkv_base(module, "SiglipAttention")
+    merge_qkv_base(module, "SiglipSdpaAttention")
 
 
 def vision_model_forward(self: torch.nn.Module, image: torch.Tensor):

ipex_llm/transformers/models/glm.py

@@ -37,6 +37,7 @@ import torch
 
 from typing import Optional, Tuple
 from transformers.cache_utils import Cache
+from transformers.models.glm.modeling_glm import GlmAttention
 from transformers.models.glm.modeling_glm import apply_rotary_pos_emb
 from ipex_llm.transformers.kv import DynamicNormalCache, DynamicFp8Cache
 from ipex_llm.transformers.models.common import merge_qkv_base
@@ -46,8 +47,9 @@ from ipex_llm.transformers.models.utils import use_quantize_kv_cache
 
 
 def merge_qkv(module: torch.nn.Module):
-    merge_qkv_base(module, "GlmAttention")
+    merge_qkv_base(module, GlmAttention)
     merge_qkv_base(module, "SiglipAttention")
+    merge_qkv_base(module, "SiglipSdpaAttention")
 
 
 def split_mlp(module: torch.nn.Module):

ipex_llm/transformers/models/llama.py

@@ -116,7 +116,7 @@ def llama_model_forward(
 
 
 def merge_qkv(module: torch.nn.Module):
-    return merge_qkv_base(module, LlamaAttention)
+    merge_qkv_base(module, LlamaAttention)
 
 
 def llama_attention_forward(

ipex_llm/transformers/models/minicpm.py

@@ -51,7 +51,8 @@ from transformers.cache_utils import Cache
 
 
 def merge_qkv(module: torch.nn.Module):
-    return merge_qkv_base(module, "MiniCPMAttention")
+    merge_qkv_base(module, "MiniCPMAttention")
+    merge_qkv_base(module, "MiniCPMSdpaAttention")
 
 
 def apply_residual_scale(module: torch.nn.Module):

ipex_llm/transformers/models/minicpmv.py

@@ -36,6 +36,7 @@ from transformers.generation.logits_process import RepetitionPenaltyLogitsProces
 # MiniCPM-V-2_5 and MiniCPM-V-2_6
 def merge_qkv(module: torch.nn.Module):
     merge_qkv_base(module, "SiglipAttention")
+    merge_qkv_base(module, "SiglipSdpaAttention")
     merge_qkv_base(module, "Idefics2VisionAttention")
 
 

ipex_llm/transformers/models/utils.py

@@ -19,7 +19,7 @@ import torch
 import warnings
 from ipex_llm.utils.common import invalidInputError
 from ipex_llm.ggml.quantize import ggml_tensor_qtype
-from ipex_llm.transformers.utils import get_ipex_version, get_xpu_device_name
+from ipex_llm.transformers.utils import get_xpu_device_name
 from ipex_llm.transformers.low_bit_linear import SYM_INT4, SYM_INT8, FP8E5, IQ2_XXS, FP4, FP8E4,\
     FP6, ASYM_INT4
 

ipex_llm/transformers/utils.py

@@ -154,20 +154,6 @@ def get_autocast_dtype(x):
                           f"Device {x.device} is not supported.")
 
 
-_ipex_version = None
-
-
-def get_ipex_version():
-
-    global _ipex_version
-    if _ipex_version is not None:
-        return _ipex_version
-
-    import intel_extension_for_pytorch as ipex
-    _ipex_version = ipex.__version__
-    return _ipex_version
-
-
 def get_xpu_device_name(device: torch.device):
     if device.type != "xpu":
         return device.type

ipex_llm/transformers/xpu_ops.py

@@ -20,9 +20,9 @@ import xe_batch
 import xe_addons
 
 
-@torch.library.register_fake("ipex_llm::forward_new")
-def _(x, weight, qtype, input_size):
-    return torch.empty_like(x)
+# @torch.library.register_fake("ipex_llm::forward_new")
+# def _(x, weight, qtype, input_size):
+#     return ???
 
 
 # @torch.library.register_fake("ipex_llm::dequant")
@@ -32,32 +32,38 @@ def _(x, weight, qtype, input_size):
 
 @torch.library.register_fake("ipex_llm::mlp_forward_xpu")
 def _(x, weight1, weight2, batch_size, state_size, output_size, act_type, qtype):
-    return torch.empty_like(x)
+    return torch.empty([batch_size, output_size],
+                       dtype=x.dtype, device=x.device)
 
 
-# @torch.library.register_fake("ipex_llm::rwkv_linear_attention_v4")
-# def _(time_decay, time_first, key, value, num_state, den_state, max_state)
-    # return ???
+@torch.library.register_fake("ipex_llm::rwkv_linear_attention_v4")
+def _(time_decay, time_first, key, value, num_state, den_state, max_state):
+    return torch.empty_like(key)
 
 
-# @torch.library.register_fake("ipex_llm::rwkv_linear_attention_v5")
-# def _(time_decay, time_first, receptance, key, value, state)
-    # return ???
+@torch.library.register_fake("ipex_llm::rwkv_linear_attention_v5")
+def _(time_decay, time_first, receptance, key, value, state):
+    bsz, n_heads, seq_len, head_dim = key.shape
+    return torch.empty([bsz, seq_len, n_heads, head_dim],
+                       dtype=key.dtype, device=key.device)
 
 
-# @torch.library.register_fake("ipex_llm::rwkv_time_shift")
-# def _(hidden, shifted, mix):
-    # return ???
+@torch.library.register_fake("ipex_llm::rwkv_time_shift")
+def _(hidden, shifted, mix):
+    bsz, seq_len, hidden_size = hidden.shape
+    return torch.empty([mix.size(0), bsz, seq_len, hidden_size],
+                       dtype=hidden.dtype, device=hidden.device)
 
 
-# @torch.library.register_fake("ipex_llm::dequantize_rows")
-# def _(x, weight, qtype, state_size, output_size):
-    # return ???
+@torch.library.register_fake("ipex_llm::dequantize_rows")
+def _(x, weight, qtype, state_size, output_size):
+    return torch.empty([x.size(0), x.size(1), state_size],
+                       dtype=torch.float, device=weight.device)
 
 
-@torch.library.register_fake("ipex_llm::batch_forward")
-def _(x, weight, qtype):
-    return torch.empty_like(x)
+# @torch.library.register_fake("ipex_llm::batch_forward")
+# def _(x, weight, qtype):
+#     return ???
 
 
 @torch.library.register_fake("ipex_llm::sdp")

{ipex_llm-2.2.0b20250106.post1.dist-info → ipex_llm-2.2.0b20250108.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250106.post1
+Version: 2.2.0b20250108
 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 ==2.6.0b20250106.post1 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250108 ; extra == 'cpp'
 Requires-Dist: setuptools ; extra == 'cpp'
 Provides-Extra: cpp-arl
-Requires-Dist: bigdl-core-cpp ==2.6.0b20250106.post1 ; extra == 'cpp-arl'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250108 ; 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 ==2.6.0b20250106.post1 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.6.0b20250108 ; (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 ==2.6.0b20250106.post1 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250106.post1 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250106.post1 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250108 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250108 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250108 ; 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 ==2.6.0b20250106.post1 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250106.post1 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250106.post1 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250108 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250108 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250108 ; 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 ==2.6.0b20250106.post1 ; extra == 'xpu-2-6'
+Requires-Dist: bigdl-core-xe-all ==2.6.0b20250108 ; extra == 'xpu-2-6'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'xpu-2-6'
 Provides-Extra: xpu-arc
 Requires-Dist: py-cpuinfo ; extra == 'xpu-arc'
@@ -137,9 +137,9 @@ 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 ==2.6.0b20250106.post1 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250108 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250108 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250108 ; 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'
@@ -160,9 +160,9 @@ 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 ==2.6.0b20250106.post1 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250108 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250108 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250108 ; 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'
@@ -183,9 +183,9 @@ 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 ==2.6.0b20250106.post1 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250106.post1 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250106.post1 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250108 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250108 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250108 ; 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-2.2.0b20250106.post1.dist-info → ipex_llm-2.2.0b20250108.dist-info}/RECORD

@@ -94,14 +94,14 @@ ipex_llm/serving/fastchat/tgi_api_protocol.py,sha256=brT3k3-V0NJrU4fRqUwWjC0O3iO
 ipex_llm/serving/fastchat/tgi_api_server.py,sha256=agNTAEiZPSuj3dEdIdYKwkoY0cXOUDX06DiM9VP2knQ,24418
 ipex_llm/serving/fastchat/vllm_worker.py,sha256=ZLz2Q9GxJO6r_LOiP6epgCRjBGk-K4EB1SNEWSJp5DA,11091
 ipex_llm/transformers/__init__.py,sha256=l4KkMkLe-pRC7b_kj6LCfeifgE-Uo33_Av_FwN9HnFA,1074
-ipex_llm/transformers/convert.py,sha256=TxWdTTOSvh-j5jqokQJVWykta4U4LHupE1QJ-9udzwc,98733
+ipex_llm/transformers/convert.py,sha256=APf2uHMgEeiAhsKm9dPgPWlyO0ADq2yHtZgovv9oczU,99101
 ipex_llm/transformers/convert_ipex.py,sha256=iKXo0n8fVFTOA2fNYYrByMFK0dovL-kLd2sVDk88AlQ,14334
 ipex_llm/transformers/embedding.py,sha256=bdgk59DvD4ZZyxRzewXOR7g56nThgO6uhIwk8QL7f-s,9299
 ipex_llm/transformers/kv.py,sha256=k4TU18LlA-Sbq9WNNQnfuzu3RSFBwFhmaV3BcGN5bAo,19191
 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=lPIvDuRoS0zusiJ6vw_fOTJgK5ylh4CuD3U-qs8ih4Y,40869
+ipex_llm/transformers/low_bit_linear.py,sha256=Obdd08D9dvuroS_6XWo4DXO_DrNRsbAqjz-mQAHmfxY,40845
 ipex_llm/transformers/model.py,sha256=fj7LBjrWtWwDJJYXnWiXsLGS4ayqqHfnh0p51dSDssE,40908
 ipex_llm/transformers/modelling_bigdl.py,sha256=7JpNVMuyq_OmtNUaMFMXdxPWZp2q0QHC02QeA-VTPOw,6709
 ipex_llm/transformers/npu_model.py,sha256=YW02GeVz-9ZGqxAeSz0AOvciS-17bo9eK5ZOBrICwSQ,39508
@@ -112,9 +112,9 @@ ipex_llm/transformers/relora.py,sha256=-dYzUV0P-IhO2jFdnzN9-v_sFzJpRj3ZwN9eCJzOo
 ipex_llm/transformers/speculative.py,sha256=0XNLgc9dGswJHVPrXo4iM7pPxkWwfFfJMECcivJSnIc,63368
 ipex_llm/transformers/streamer.py,sha256=RrVlLblzCOtABRUpaMXAyaMnCGgLUtAi_YesLumRbww,4842
 ipex_llm/transformers/training_patch.py,sha256=oxMkUtqyvqJiprw6dE3skkYfD1HOmUlH9N0hBkbn0G0,10799
-ipex_llm/transformers/utils.py,sha256=7syzq4jnEo-mWmS9E2VZ2GcFrjojWI8E7Hcx0tloifg,16996
+ipex_llm/transformers/utils.py,sha256=9IRSqfDokf8QFW9T47R--i3RL1E-_O31HO7IJf7H6pg,16748
 ipex_llm/transformers/xpu_customize_fwd.py,sha256=wFpIhs5F6tkNs8gBOrLxWdhLzO3EDHovVkERPIAoAvg,7611
-ipex_llm/transformers/xpu_ops.py,sha256=H46-69pMRQhekbAEoDfNacCInLWycMHDqrgMGLvFYfI,4362
+ipex_llm/transformers/xpu_ops.py,sha256=vw4cUwvqUqDr45d-WMIkCpM2oiHfjN-VjF0bjMSF4kY,4830
 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
@@ -150,21 +150,21 @@ ipex_llm/transformers/models/bloom.py,sha256=PxfzyYT-nFn3K5rZhTQjmcEjUUzAhUFzxIN
 ipex_llm/transformers/models/chatglm.py,sha256=UHai1t2AUtGmF765_eHF8LUMVQzp_oCBx8TJB21WrHk,12597
 ipex_llm/transformers/models/chatglm2.py,sha256=SGCABJdYQLW0zDarEoWrEQLuWlbq9iQhYU8ZeR1-ptQ,15957
 ipex_llm/transformers/models/chatglm4.py,sha256=AAhAFFDDas5DBQPfh2Mwl7a2v7taKf6xphoeeNNFaBI,16593
-ipex_llm/transformers/models/chatglm4v.py,sha256=YRfuf9g1E0MQ_7wbHAOMvadFnO-j3LqI_k1SaRkDs0M,14055
+ipex_llm/transformers/models/chatglm4v.py,sha256=tyjDDyF6FEgLAT24EG3i4-auxZvkwmeLIy0Hds4K5Yo,14105
 ipex_llm/transformers/models/common.py,sha256=4obQMGF02FCiXrHnFle9Fsx7C33b1FDt37qJJ4YgxRc,11578
 ipex_llm/transformers/models/decilm.py,sha256=P-PBuDPf07GvKggLwJx_wPwIn6esN3rX8ai2JxRuZmE,5246
 ipex_llm/transformers/models/gemma.py,sha256=_E3Yw8Y45xyNVeLqyVKcpr8kjuICtETeL82cJ-bWJuU,9424
 ipex_llm/transformers/models/gemma2.py,sha256=2WZuv-FLzJyTJFaYxOuzJt47QE64M0lHnzAiO5T6ozI,8049
-ipex_llm/transformers/models/glm.py,sha256=gHYgfn20jPRL-ElXy-rUqMh6_LQcc5x7DEXSZuRA4E0,7094
+ipex_llm/transformers/models/glm.py,sha256=PE43uks9lojndBBHFVXK1VWisHhbY-kuCmhq0CwmD4s,7204
 ipex_llm/transformers/models/gpt2.py,sha256=YSaNgK1uLCFDuIFqnKO0Mi-AsOZsYav-7pNf_NpKGdM,3445
 ipex_llm/transformers/models/gptbigcode.py,sha256=cP1_qGWoa43R2WacAMblShjku4QupcCZiLaPPAoOUs4,9101
 ipex_llm/transformers/models/gptneox.py,sha256=loRh1x_5S6BCeOr_s5xr-N_1SQHL3Y5IiUBAEyoMUqQ,6172
 ipex_llm/transformers/models/internlm.py,sha256=ZbIUMDwNRcrCeduXfbA_uq1AUEWawEt6CJRvQl3LkAg,17832
 ipex_llm/transformers/models/internvl.py,sha256=Vx0vENIEQLX2M6P398mw5TOhpks0U8xf8rtRQvy94go,8154
-ipex_llm/transformers/models/llama.py,sha256=ozwtdQ0MbanJEtW4LBFGxqs_QAq82EonhL2dL6tGyw0,8567
-ipex_llm/transformers/models/minicpm.py,sha256=ib2rJTN7Tf7znBCtVrtXsF-_Uuk2aA7KVg02xzatLiI,10103
+ipex_llm/transformers/models/llama.py,sha256=n1JG1uElMB8t3Hpae94S6YTO_5q2N5BUAhb7mncvA6E,8560
+ipex_llm/transformers/models/minicpm.py,sha256=_eYBYafQxnuqKo9ENNkua73KU5goU2z-dkaLlF5uHnA,10147
 ipex_llm/transformers/models/minicpm3.py,sha256=FhNS6mi2rg7dSdF_QQGrao3g9EC6XLn1MTKd-kd0wF0,9191
-ipex_llm/transformers/models/minicpmv.py,sha256=ZV4s48WNIyRoEkvENnlmopnx3ojZANBer0LI6bRtxrY,9826
+ipex_llm/transformers/models/minicpmv.py,sha256=PP05b5iTnrMpiseCn8iJcxKJDnfq7WqXp9Mrch0kKZ0,9876
 ipex_llm/transformers/models/mistral.py,sha256=rE1GWQxXvF6aG-buPHDR13zeynDZEDIubPF4PiVhZbM,7451
 ipex_llm/transformers/models/mllama.py,sha256=ogpLmmN_OwcFUyjYB-oDC-l3uw8urFvUEc5edkjWHAk,10939
 ipex_llm/transformers/models/mpt.py,sha256=z02NwHogJZVh-Mk4sYoIzR90SFIKhoNN_-ifsD907TQ,9540
@@ -181,7 +181,7 @@ 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=RGQCYuQhYqtZ1j3RZkYi0_QvCRnUgUIPYxfBcLnElzg,6885
 ipex_llm/transformers/models/starcoder2.py,sha256=4P3mhRYf2Kreb1ESjrQGfy1puLMmZXgV35zf-Tksvao,6462
-ipex_llm/transformers/models/utils.py,sha256=85rGIzGZvWe3JjYpWcuc1nfzI_tn_zFcdZpWivxJkl0,15457
+ipex_llm/transformers/models/utils.py,sha256=isBCMMQP3j_opmda9XzD_dPk1ejvEXTztggbu1yIMSc,15439
 ipex_llm/transformers/models/yuan.py,sha256=1jRPebwAK2ENbyYokOmb4LSVo-szucWiygz9zTv-scs,7656
 ipex_llm/transformers/npu_models/__init__.py,sha256=ulEUGLjaP48LCrVeury3UxLjXxKzRi0UpSG4bYu-7f8,585
 ipex_llm/transformers/npu_models/baichuan.py,sha256=fJtd7fBrttySghRUgfZTAdxLjsSNC-XL08HISsXigLE,4685
@@ -250,11 +250,11 @@ 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-2.2.0b20250106.post1.data/scripts/ipex-llm-init,sha256=fLQsT2dRL6H5bThb4GuIWotAuqoLsIxFwA-0c2qmaO8,6672
-ipex_llm-2.2.0b20250106.post1.data/scripts/llm-chat,sha256=TdUnUmNapzuoe1c8IzrdVOQwWEg8IqsMSBRlOD3daZM,2249
-ipex_llm-2.2.0b20250106.post1.data/scripts/llm-cli,sha256=RXGPlLElHxcKzoUxljEMBIAXbzCDysXL-Nxw-xF-7LU,2457
-ipex_llm-2.2.0b20250106.post1.dist-info/METADATA,sha256=I0vPU5mDtPZR3wpY87fYHdn6r14U0T50NWGP7EsF5s8,12825
-ipex_llm-2.2.0b20250106.post1.dist-info/WHEEL,sha256=PPJcBMAZibF_2GFE9NmOJGqiaSMPiNFbJd6QaJjdA6Y,109
-ipex_llm-2.2.0b20250106.post1.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
-ipex_llm-2.2.0b20250106.post1.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
-ipex_llm-2.2.0b20250106.post1.dist-info/RECORD,,
+ipex_llm-2.2.0b20250108.data/scripts/ipex-llm-init,sha256=fLQsT2dRL6H5bThb4GuIWotAuqoLsIxFwA-0c2qmaO8,6672
+ipex_llm-2.2.0b20250108.data/scripts/llm-chat,sha256=TdUnUmNapzuoe1c8IzrdVOQwWEg8IqsMSBRlOD3daZM,2249
+ipex_llm-2.2.0b20250108.data/scripts/llm-cli,sha256=RXGPlLElHxcKzoUxljEMBIAXbzCDysXL-Nxw-xF-7LU,2457
+ipex_llm-2.2.0b20250108.dist-info/METADATA,sha256=NJp_uuPOJe8b5UQ8ASJbfzen2BGoc2DEM1ZInzr0X9E,12705
+ipex_llm-2.2.0b20250108.dist-info/WHEEL,sha256=PPJcBMAZibF_2GFE9NmOJGqiaSMPiNFbJd6QaJjdA6Y,109
+ipex_llm-2.2.0b20250108.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
+ipex_llm-2.2.0b20250108.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
+ipex_llm-2.2.0b20250108.dist-info/RECORD,,