ipex-llm 2.2.0b20250405__py3-none-manylinux2010_x86_64.whl → 2.3.0b20250407__py3-none-manylinux2010_x86_64.whl

ipex_llm/transformers/convert.py

@@ -2072,12 +2072,31 @@ def _optimize_post(model):
         convert_forward(model.thinker.visual, module.Qwen2_5OmniVisionSdpaAttention,
                         qwen2_5_omni_vision_attention_forward)
 
+        # audio opt
+        from ipex_llm.transformers.models.qwen2_5_omni import qwen2_5_omni_audio_attention_forward
+        convert_forward(model.thinker.audio_tower, module.Qwen2_5OmniAudioAttention,
+                        qwen2_5_omni_audio_attention_forward)
+        convert_forward(model.thinker.audio_tower, module.Qwen2_5OmniAudioSdpaAttention,
+                        qwen2_5_omni_audio_attention_forward)
+
         # tts opt
-        if hasattr(model, "talker"):
-            convert_forward(model.talker, module.Qwen2_5OmniAttention,
+        if model.has_talker:
+            # talker part
+            convert_forward(model.talker.model, module.Qwen2_5OmniAttention,
+                            qwen2_5_omni_attention_forward)
+            convert_forward(model.talker.model, module.Qwen2_5OmniSdpaAttention,
                             qwen2_5_omni_attention_forward)
-            convert_forward(model.talker, module.Qwen2_5OmniThinkerModel,
+            convert_forward(model.talker.model, module.Qwen2_5OmniTalkerModel,
                             qwen2_5_omni_thinker_model_forward)
+            convert_forward(model.talker.model, module.Qwen2MLP, qwen2_mlp_forward)
+
+            # token2wav part
+            from ipex_llm.transformers.models.qwen2_5_omni import dit_attention_forward
+            from ipex_llm.transformers.models.qwen2_5_omni import _create_block_diff
+            convert_forward(model.token2wav, module.DiTAttention, dit_attention_forward)
+            dit_model = model.token2wav.code2wav_dit_model
+            dit_model._create_block_diff = MethodType(_create_block_diff, dit_model)
+
     return model
 
 

ipex_llm/transformers/models/qwen2_5_omni.py

@@ -20,9 +20,11 @@
 import math
 import torch
 from typing import Optional, Tuple, List, Union
-from transformers.cache_utils import Cache
+from transformers.cache_utils import Cache, EncoderDecoderCache
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS
 from transformers.modeling_outputs import BaseModelOutputWithPast
 from transformers.models.qwen2_5_omni.modeling_qwen2_5_omni import Qwen2_5OmniAttention
+from transformers.models.qwen2_5_omni.modeling_qwen2_5_omni import apply_rotary_pos_emb
 from transformers.models.qwen2_5_omni.modeling_qwen2_5_omni import apply_rotary_pos_emb_vision
 from transformers.models.qwen2_5_omni.modeling_qwen2_5_omni import apply_multimodal_rotary_pos_emb
 
@@ -284,3 +286,160 @@ def qwen2_5_omni_vision_attention_forward(
     attn_output = attn_output.reshape(seq_length, -1)
     attn_output = self.proj(attn_output)
     return attn_output
+
+
+def qwen2_5_omni_audio_attention_forward(
+    self,
+    hidden_states: torch.Tensor,
+    key_value_states: Optional[torch.Tensor] = None,
+    past_key_value: Optional[EncoderDecoderCache] = None,
+    cu_seqlens: Optional[torch.Tensor] = None,
+    layer_head_mask: Optional[torch.Tensor] = None,
+    output_attentions: bool = False,
+    cache_position: Optional[torch.LongTensor] = None,
+) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+    """Input shape: Batch x Time x Channel"""
+
+    # if key_value_states are provided this layer is used as a cross-attention layer
+    # for the decoder
+    is_cross_attention = key_value_states is not None
+    seq_length, _ = hidden_states.size()
+
+    # get query proj
+    query_states = self.q_proj(hidden_states)
+    query_states = query_states.reshape(seq_length, self.num_heads, -1)
+
+    seq_lens = cu_seqlens.tolist()
+    invalidInputError(seq_lens[0] == 0 and seq_lens[-1] == seq_length,
+                      "unexpected input")
+
+    if past_key_value is not None:
+        is_updated = past_key_value.is_updated.get(self.layer_idx)
+        if is_cross_attention:
+            # after the first generated id,
+            # we can subsequently re-use all key/value_states from cache
+            past_key_value.is_updated[self.layer_idx] = True
+            past_key_value = past_key_value.cross_attention_cache
+        else:
+            past_key_value = past_key_value.self_attention_cache
+
+    # use key_value_states if cross attention
+    current_states = key_value_states if key_value_states is not None else hidden_states
+    if is_cross_attention and past_key_value and is_updated:
+        # reuse k,v, cross_attentions
+        key_states = past_key_value.key_cache[self.layer_idx]
+        value_states = past_key_value.value_cache[self.layer_idx]
+    else:
+        key_states = self.k_proj(current_states).reshape(seq_length, self.num_heads, -1)
+        value_states = self.v_proj(current_states).reshape(seq_length, self.num_heads, -1)
+        if past_key_value is not None:
+            # save all key/value_states to cache to be re-used for fast auto-regressive generation
+            cache_position = cache_position if not is_cross_attention else None
+            key_states, value_states = past_key_value.update(
+                key_states, value_states, self.layer_idx, {"cache_position": cache_position}
+            )
+
+    if layer_head_mask is None and use_sdp_non_causal(query_states.size(-1),
+                                                      query_states.device, query_states.dtype):
+        kv_length = key_states.size(0)
+        padding_kv_length = (kv_length + 128 - 1) // 128 * 128
+        attention_mask = torch.full(
+            [1, 1, seq_length, padding_kv_length], torch.finfo(query_states.dtype).min,
+            device=query_states.device, dtype=query_states.dtype,
+        )
+        for i in range(1, len(cu_seqlens)):
+            attention_mask[..., seq_lens[i - 1]:seq_lens[i], seq_lens[i - 1]:seq_lens[i]] = 0
+
+        q = query_states.transpose(0, 1).unsqueeze(0)
+        k = key_states.transpose(0, 1).unsqueeze(0).contiguous()
+        v = value_states.transpose(0, 1).unsqueeze(0).contiguous()
+        # q, k, v: [1, num_heads, seq_length, head_dim]
+
+        attn_weights = None
+        attn_output = scaled_dot_product_attention(q, k, v, attention_mask, False)
+        attn_output = attn_output.permute(0, 2, 1, 3).squeeze(0)
+        # attn_output: [seq_length, num_heads, head_dim]
+    else:
+        attention_mask = torch.full(
+            [1, seq_length, key_states.size(0)], torch.finfo(query_states.dtype).min,
+            device=query_states.device, dtype=query_states.dtype,
+        )
+        for i in range(1, len(cu_seqlens)):
+            attention_mask[..., seq_lens[i - 1]:seq_lens[i], seq_lens[i - 1]:seq_lens[i]] = 0
+
+        query_states = query_states.transpose(0, 1)
+        key_states = key_states.transpose(0, 1)
+        value_states = value_states.transpose(0, 1)
+
+        attn_weights = torch.matmul(query_states,
+                                    key_states.transpose(1, 2)) / math.sqrt(self.head_dim)
+        attn_weights = attn_weights + attention_mask
+        attn_weights = attention_softmax(attn_weights)
+
+        if layer_head_mask is not None:
+            attn_weights = layer_head_mask.view(1, -1, 1, 1) * attn_weights
+
+        attn_output = torch.matmul(attn_weights, value_states).transpose(0, 1)
+
+    # Use the `embed_dim` from the config (stored in the class) rather than `hidden_state`s
+    # because `attn_output` can be partitioned across GPUs when using tensor-parallelism.
+    attn_output = attn_output.reshape(seq_length, self.embed_dim)
+    attn_output = self.out_proj(attn_output)
+
+    return attn_output, attn_weights, past_key_value
+
+
+def dit_attention_forward(
+    self,
+    x,
+    rope=None,
+    mask=None,
+) -> torch.Tensor:
+    batch_size = x.shape[0]
+
+    # `sample` projections.
+    query = self.to_q(x)
+    key = self.to_k(x)
+    value = self.to_v(x)
+
+    # attention
+    inner_dim = key.shape[-1]
+    head_dim = inner_dim // self.heads
+    query = query.view(batch_size, -1, self.heads, head_dim).transpose(1, 2)
+    key = key.view(batch_size, -1, self.heads, head_dim).transpose(1, 2)
+    value = value.view(batch_size, -1, self.heads, head_dim).transpose(1, 2)
+
+    # apply rotary position embedding
+    # Due to training process, only first head is applied with RoPE, will be fixed at next release
+    cos, sin = rope
+    query[:, :1], key[:, :1] = apply_rotary_pos_emb(query[:, :1], key[:, :1], cos, sin)
+
+    if use_sdp_non_causal(head_dim, query.device, query.dtype):
+        mask = torch.where(mask, 0, torch.finfo(query.dtype).min)
+        x = scaled_dot_product_attention(query, key.contiguous(), value.contiguous(), mask, False)
+        x = x.transpose(1, 2)
+    else:
+        attention_interface = ALL_ATTENTION_FUNCTIONS[self._attn_implementation]
+        x, _ = attention_interface(self, query, key, value, attention_mask=mask, is_causal=False)
+
+    # mask
+    x = x.reshape(batch_size, -1, self.heads * head_dim)
+    x = x.to(query.dtype)
+
+    # linear proj
+    x = self.to_out[0](x)
+    # dropout
+    x = self.to_out[1](x)
+
+    return x
+
+
+def _create_block_diff(self, x):
+    batch, seq_len = x.shape[0], x.shape[1]
+    block_indices = torch.arange(seq_len, device=x.device) // self.block_size
+
+    block_i = block_indices.unsqueeze(1)  # [seq_length, 1]
+    block_j = block_indices.unsqueeze(0)  # [1, seq_length]
+
+    block_diff = block_j - block_i  # (n, n)
+    return block_diff.unsqueeze(0).unsqueeze(0)

{ipex_llm-2.2.0b20250405.dist-info → ipex_llm-2.3.0b20250407.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250405
+Version: 2.3.0b20250407
 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 ==2.6.0b20250405 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.7.0b20250407 ; 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 ==2.6.0b20250405 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.7.0b20250407 ; (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 ==2.6.0b20250405 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250405 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250405 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-21 ==2.7.0b20250407 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.7.0b20250407 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.7.0b20250407 ; 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 ==2.6.0b20250405 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250405 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250405 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-21 ==2.7.0b20250407 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.7.0b20250407 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.7.0b20250407 ; 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 ==2.6.0b20250405 ; extra == 'xpu-2-6'
+Requires-Dist: bigdl-core-xe-all ==2.7.0b20250407 ; 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'
@@ -132,7 +132,7 @@ Requires-Dist: tokenizers ==0.15.2 ; extra == 'xpu-2-6-arl'
 Requires-Dist: accelerate ==0.23.0 ; extra == 'xpu-2-6-arl'
 Requires-Dist: tabulate ; extra == 'xpu-2-6-arl'
 Requires-Dist: setuptools ; extra == 'xpu-2-6-arl'
-Requires-Dist: bigdl-core-xe-all ==2.6.0b20250405 ; extra == 'xpu-2-6-arl'
+Requires-Dist: bigdl-core-xe-all ==2.7.0b20250407 ; extra == 'xpu-2-6-arl'
 Requires-Dist: onednn-devel ==2025.0.1 ; extra == 'xpu-2-6-arl'
 Requires-Dist: onednn ==2025.0.1 ; extra == 'xpu-2-6-arl'
 Requires-Dist: dpcpp-cpp-rt ==2025.0.2 ; extra == 'xpu-2-6-arl'
@@ -155,9 +155,9 @@ 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 ==2.6.0b20250405 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250405 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250405 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-23 ==2.7.0b20250407 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.7.0b20250407 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.7.0b20250407 ; 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'
@@ -178,9 +178,9 @@ 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 ==2.6.0b20250405 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250405 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250405 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-23 ==2.7.0b20250407 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.7.0b20250407 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.7.0b20250407 ; 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'
@@ -201,9 +201,9 @@ 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 ==2.6.0b20250405 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250405 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250405 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-23 ==2.7.0b20250407 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.7.0b20250407 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.7.0b20250407 ; 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-2.2.0b20250405.dist-info → ipex_llm-2.3.0b20250407.dist-info}/RECORD

@@ -94,7 +94,7 @@ 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=BreA3EY6hkNq0rVixb_sUuTLzMrcWXTt3yRsshCPHHQ,1214
-ipex_llm/transformers/convert.py,sha256=D3QSoD48NzOKz9rKIsgrI0sNzdmJvb0sP4k_0f8Lcmo,105824
+ipex_llm/transformers/convert.py,sha256=xqqZFGcdDRko2IYgfSgDRs8ef4THUR25IAhSyDV0VUs,106933
 ipex_llm/transformers/convert_ipex.py,sha256=_nSnUTQy-yfkKaqGdqnBdWztZf3NGmnbZ0TKaDrF4X4,14617
 ipex_llm/transformers/embedding.py,sha256=bdgk59DvD4ZZyxRzewXOR7g56nThgO6uhIwk8QL7f-s,9299
 ipex_llm/transformers/kv.py,sha256=k4TU18LlA-Sbq9WNNQnfuzu3RSFBwFhmaV3BcGN5bAo,19191
@@ -178,7 +178,7 @@ ipex_llm/transformers/models/phi3.py,sha256=AaWB7TPQdrDYgpcVHglG0Q0480bxNOw1mFeP
 ipex_llm/transformers/models/phixtral.py,sha256=MDTMghcu7qAmZmRcUGqXXDXhSU3y_N59HRIXmlcjp5g,4890
 ipex_llm/transformers/models/qwen.py,sha256=A3WiVCzA7NLkcjp4zhFkZvKZzZWZlg0WFuVV_556TAI,19543
 ipex_llm/transformers/models/qwen2.py,sha256=zK-FpUaxEhjD4gZa1ZvArodAilz29T_cpeAqfCGosc0,14317
-ipex_llm/transformers/models/qwen2_5_omni.py,sha256=uAm_dZBAf53nYt3d1bFitFcIWJV80wqU4q4NJRbwxIE,12015
+ipex_llm/transformers/models/qwen2_5_omni.py,sha256=gNnWvGiPntn6GlyIUtCVYCSIMPU6FG9TCS7fhWbwerY,18779
 ipex_llm/transformers/models/qwen2_moe.py,sha256=a0gYo-ngf8SxaEnBdZUJDnPS6Mkn_poDd8xqhx50icI,19516
 ipex_llm/transformers/models/qwen2_vl.py,sha256=G-9e2oN4f5p5IWQ-zsBZuONxTura3BjlgyT2meigbHQ,13579
 ipex_llm/transformers/models/qwen_vl.py,sha256=j7Nzzz2Qvynu9yrCXmoEfERjw43hXof5TbXIs7Ms-oY,17105
@@ -263,11 +263,11 @@ ipex_llm/vllm/xpu/engine/__init__.py,sha256=pY_CpyuZd72fr6s32ejeKHKFW0K4vUU2rzZj
 ipex_llm/vllm/xpu/engine/engine.py,sha256=NvCMbp0X8NVrOqbwm4FTvXOptTRLzu9jQsy37ZHnTk8,9493
 ipex_llm/vllm/xpu/entrypoints/openai/api_server.py,sha256=IjiSze9vzBCAkLu_VwIcJwuO1jyFna7DLrj6aSL7RaQ,35220
 ipex_llm/vllm/xpu/entrypoints/openai/cli_args.py,sha256=hB398yYtKauASRzevctScdbFIjiiSGMAe1bwEuIHrhY,10893
-ipex_llm-2.2.0b20250405.data/scripts/ipex-llm-init,sha256=fLQsT2dRL6H5bThb4GuIWotAuqoLsIxFwA-0c2qmaO8,6672
-ipex_llm-2.2.0b20250405.data/scripts/llm-chat,sha256=TdUnUmNapzuoe1c8IzrdVOQwWEg8IqsMSBRlOD3daZM,2249
-ipex_llm-2.2.0b20250405.data/scripts/llm-cli,sha256=RXGPlLElHxcKzoUxljEMBIAXbzCDysXL-Nxw-xF-7LU,2457
-ipex_llm-2.2.0b20250405.dist-info/METADATA,sha256=PgVyv73hHB7drLqtbM1eeXXGVckTSeeiLcIIgJu7OLg,13917
-ipex_llm-2.2.0b20250405.dist-info/WHEEL,sha256=PPJcBMAZibF_2GFE9NmOJGqiaSMPiNFbJd6QaJjdA6Y,109
-ipex_llm-2.2.0b20250405.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
-ipex_llm-2.2.0b20250405.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
-ipex_llm-2.2.0b20250405.dist-info/RECORD,,
+ipex_llm-2.3.0b20250407.data/scripts/ipex-llm-init,sha256=fLQsT2dRL6H5bThb4GuIWotAuqoLsIxFwA-0c2qmaO8,6672
+ipex_llm-2.3.0b20250407.data/scripts/llm-chat,sha256=TdUnUmNapzuoe1c8IzrdVOQwWEg8IqsMSBRlOD3daZM,2249
+ipex_llm-2.3.0b20250407.data/scripts/llm-cli,sha256=RXGPlLElHxcKzoUxljEMBIAXbzCDysXL-Nxw-xF-7LU,2457
+ipex_llm-2.3.0b20250407.dist-info/METADATA,sha256=8oGf4R8c2G6AyNdw28cESK3wjRjCnyghRmKtGC8dTLI,13917
+ipex_llm-2.3.0b20250407.dist-info/WHEEL,sha256=PPJcBMAZibF_2GFE9NmOJGqiaSMPiNFbJd6QaJjdA6Y,109
+ipex_llm-2.3.0b20250407.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
+ipex_llm-2.3.0b20250407.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
+ipex_llm-2.3.0b20250407.dist-info/RECORD,,