ipex-llm 2.2.0b20250326__py3-none-win_amd64.whl → 2.2.0b20250328__py3-none-win_amd64.whl

ipex_llm/transformers/convert.py

@@ -1074,6 +1074,9 @@ def _optimize_pre(model, qtype=None):
     elif model.config.model_type == "deepseek_v3" and model.config.hidden_size == 2048:
         from ipex_llm.transformers.models.deepseek import padding_mla_v_hd
         model.apply(padding_mla_v_hd)
+    elif model.config.model_type == "qwen2_5_omni":
+        from ipex_llm.transformers.models.qwen2_5_omni import merge_qkv
+        model.apply(merge_qkv)
     return model
 
 
@@ -2043,7 +2046,38 @@ def _optimize_post(model):
         convert_forward(model, module.DeepseekV3Model, deepseek_model_forward)
         convert_forward(model, module.DeepseekV3Attention, deepseek_attention_forward)
         convert_forward(model, module.DeepseekV3MoE, deepseek_moe_forward)
+    elif model.config.model_type == "qwen2_5_omni":
+        modeling_module_name = model.__class__.__module__
+        module = importlib.import_module(modeling_module_name)
 
+        # llm opt
+        from ipex_llm.transformers.models.qwen2_5_omni import qwen2_5_omni_attention_forward
+        from ipex_llm.transformers.models.qwen2_5_omni import qwen2_5_omni_thinker_model_forward
+        from ipex_llm.transformers.models.qwen2 import qwen2_mlp_forward
+        from ipex_llm.transformers.models.common import rms_norm_forward
+        convert_forward(model.thinker.model, module.Qwen2_5OmniAttention,
+                        qwen2_5_omni_attention_forward)
+        convert_forward(model.thinker.model, module.Qwen2_5OmniSdpaAttention,
+                        qwen2_5_omni_attention_forward)
+        convert_forward(model.thinker.model, module.Qwen2_5OmniThinkerModel,
+                        qwen2_5_omni_thinker_model_forward)
+        convert_forward(model.thinker.model, module.Qwen2MLP, qwen2_mlp_forward)
+        convert_forward(model, module.Qwen2RMSNorm, rms_norm_forward)
+
+        # vision opt
+        from ipex_llm.transformers.models.qwen2_vl import qwen2_vision_get_dtype
+        from ipex_llm.transformers.models.qwen2_5_omni import qwen2_5_omni_vision_attention_forward
+        convert_forward(model.thinker.visual, module.Qwen2_5OmniVisionAttention,
+                        qwen2_5_omni_vision_attention_forward)
+        convert_forward(model.thinker.visual, module.Qwen2_5OmniVisionSdpaAttention,
+                        qwen2_5_omni_vision_attention_forward)
+
+        # tts opt
+        if hasattr(model, "talker"):
+            convert_forward(model.talker, module.Qwen2_5OmniAttention,
+                            qwen2_5_omni_attention_forward)
+            convert_forward(model.talker, module.Qwen2_5OmniThinkerModel,
+                            qwen2_5_omni_thinker_model_forward)
     return model
 
 

ipex_llm/transformers/models/qwen2_5_omni.py

@@ -0,0 +1,286 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+# Some parts of this file is adapted from
+# https://github.com/huggingface/transformers/blob/3a1ead0aabed473eafe527915eea8c197d424356/src/transformers/models/qwen2_5_omni/modeling_qwen2_5_omni.py
+# which is licensed under Apache License 2.0
+
+import math
+import torch
+from typing import Optional, Tuple, List, Union
+from transformers.cache_utils import Cache
+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_vision
+from transformers.models.qwen2_5_omni.modeling_qwen2_5_omni import apply_multimodal_rotary_pos_emb
+
+from ipex_llm.utils.common import invalidInputError
+from ipex_llm.transformers.kv import DynamicNormalCache
+from ipex_llm.transformers.models.common import merge_qkv_base
+from ipex_llm.transformers.models.common import attention_softmax
+from ipex_llm.transformers.models.common import scaled_dot_product_attention
+from ipex_llm.transformers.models.utils import use_sdp_non_causal
+
+
+def merge_qkv(module: torch.nn.Module):
+    merge_qkv_base(module, Qwen2_5OmniAttention)
+
+
+def qwen2_5_omni_attention_forward(
+    self,
+    hidden_states: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_value: Optional[Cache] = None,
+    output_attentions: bool = False,
+    use_cache: bool = False,
+    cache_position: Optional[torch.LongTensor]=None,
+    position_embeddings: Tuple[torch.Tensor, torch.Tensor]=None,
+):
+    bsz, q_len, _ = hidden_states.size()
+
+    qkv = self.qkv_proj(hidden_states)
+    qkv = qkv.view(bsz, q_len, self.num_heads + 2 * self.num_key_value_heads, self.head_dim)
+    qkv = qkv.transpose(1, 2)
+    query_states, key_states, value_states = qkv.split([self.num_heads,
+                                                        self.num_key_value_heads,
+                                                        self.num_key_value_heads], dim=1)
+
+    cos, sin = position_embeddings
+    if query_states.device.type == "xpu":
+        import xe_addons
+        xe_addons.rotary_half_with_cache_inplaced(query_states, key_states, cos, sin)
+    else:
+        query_states, key_states = apply_multimodal_rotary_pos_emb(
+            query_states, key_states, cos, sin, self.rope_scaling["mrope_section"]
+        )
+
+    key_states, value_states = past_key_value.update(key_states, value_states,
+                                                     self.layer_idx, None)
+
+    attn_weights = None
+    attn_output = scaled_dot_product_attention(
+        query_states, key_states, value_states,
+        attention_mask, q_len == key_states.size(2)
+    )
+
+    attn_output = attn_output.transpose(1, 2).contiguous()
+    attn_output = attn_output.reshape(bsz, q_len, -1)
+
+    attn_output = self.o_proj(attn_output)
+
+    if not output_attentions:
+        attn_weights = None
+    return attn_output, attn_weights, past_key_value
+
+
+def qwen2_5_omni_thinker_model_forward(
+    self,
+    input_ids: torch.LongTensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[List[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+) -> Union[Tuple, BaseModelOutputWithPast]:
+    output_attentions = (
+        output_attentions if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_hidden_states = (
+        output_hidden_states if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+
+    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+    invalidInputError((input_ids is None) ^ (inputs_embeds is None),
+                      "You must specify exactly one of input_ids or inputs_embeds")
+
+    if inputs_embeds is None:
+        inputs_embeds = self.embed_tokens(input_ids)
+
+    # ipex-llm changes start: kv cache
+    use_cache = use_cache if use_cache is not None else self.config.use_cache
+    use_cache = True if inputs_embeds.device.type == "xpu" else use_cache
+    # torch.jit.trace() doesn't support cache objects in the output
+    if use_cache and not torch.jit.is_tracing():
+        if not isinstance(past_key_values, DynamicNormalCache):
+            past_key_values = DynamicNormalCache.from_legacy_cache(past_key_values)
+    # ipex-llm changes end
+
+    if cache_position is None:
+        past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+        cache_position = torch.arange(
+            past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+        )
+
+    # the hard coded `3` is for temporal, height and width.
+    if position_ids is None:
+        position_ids = cache_position.view(1, 1, -1).expand(3, inputs_embeds.shape[0], -1)
+    elif position_ids.dim() == 2:
+        position_ids = position_ids[None, ...].expand(3, position_ids.shape[0], -1)
+
+    causal_mask = self._update_causal_mask(
+        attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions
+    )
+
+    hidden_states = inputs_embeds
+
+    # create position embeddings to be shared across the decoder layers
+    position_embeddings = self.rotary_emb(hidden_states, position_ids)
+    # ipex-llm changes start: rotary embedding
+    if inputs_embeds.device.type == "xpu":
+        cos, sin = position_embeddings
+        mrope_section = self.config.rope_scaling["mrope_section"] * 2
+        cos = torch.cat([m[i % 3] for i, m in
+                        enumerate(cos.split(mrope_section, dim=-1))], dim=-1).unsqueeze(1)
+        sin = torch.cat([m[i % 3] for i, m in
+                        enumerate(sin.split(mrope_section, dim=-1))], dim=-1).unsqueeze(1)
+        position_embeddings = cos.contiguous(), sin.contiguous()
+    # ipex-llm changes end
+
+    # decoder layers
+    all_hidden_states = () if output_hidden_states else None
+    all_self_attns = () if output_attentions else None
+    next_decoder_cache = None
+
+    for decoder_layer in self.layers:
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        layer_outputs = decoder_layer(
+            hidden_states,
+            attention_mask=causal_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_values,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+        )
+
+        hidden_states = layer_outputs[0]
+
+        if use_cache:
+            next_decoder_cache = layer_outputs[2 if output_attentions else 1]
+
+        if output_attentions:
+            all_self_attns += (layer_outputs[1],)
+
+    hidden_states = self.norm(hidden_states)
+
+    # add hidden states from the last decoder layer
+    if output_hidden_states:
+        all_hidden_states += (hidden_states,)
+
+    next_cache = next_decoder_cache if use_cache else None
+
+    if not return_dict:
+        return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
+                     if v is not None)
+    return BaseModelOutputWithPast(
+        last_hidden_state=hidden_states,
+        past_key_values=next_cache,
+        hidden_states=all_hidden_states,
+        attentions=all_self_attns,
+    )
+
+
+def qwen2_5_omni_vision_attention_forward(
+    self,
+    hidden_states: torch.Tensor,
+    cu_seqlens: torch.Tensor,
+    rotary_pos_emb: torch.Tensor = None
+) -> torch.Tensor:
+    seq_length = hidden_states.shape[0]
+    q = self.q(hidden_states).reshape(seq_length, self.num_heads, -1)
+    k = self.k(hidden_states).reshape(seq_length, self.num_heads, -1)
+    v = self.v(hidden_states).reshape(seq_length, self.num_heads, -1)
+    q = apply_rotary_pos_emb_vision(q.unsqueeze(0), rotary_pos_emb).squeeze(0)
+    k = apply_rotary_pos_emb_vision(k.unsqueeze(0), rotary_pos_emb).squeeze(0)
+    # q, k, v: [seq_length, num_heads, head_dim]
+
+    seq_lens = cu_seqlens.tolist()
+    invalidInputError(seq_lens[0] == 0 and seq_lens[-1] == seq_length,
+                      "unexpected input")
+
+    head_dim = q.size(-1)
+    if use_sdp_non_causal(head_dim, q.device, q.dtype):
+        image_num = len(seq_lens) - 1
+        image_size = seq_lens[1] - seq_lens[0]
+        guessed_seq_lens = torch.arange(0, (image_num + 1) * image_size, image_size,
+                                        dtype=cu_seqlens.dtype, device=cu_seqlens.device)
+        if (guessed_seq_lens == cu_seqlens).all():
+            q = q.view(image_num, image_size, self.num_heads, head_dim).permute(0, 2, 1, 3)
+            k = k.view(image_num, image_size, self.num_heads, head_dim).permute(0, 2, 1, 3)
+            v = v.view(image_num, image_size, self.num_heads, head_dim).permute(0, 2, 1, 3)
+            # q, k, v: [image_num, num_heads, image_size, head_dim]
+
+            attn_output = scaled_dot_product_attention(
+                q, k.contiguous(), v.contiguous(),
+                None, False
+            )
+            attn_output = attn_output.permute(0, 2, 1, 3).contiguous()
+            attn_output = attn_output.view(seq_length, self.num_heads, head_dim)
+            # attn_output: [seq_length, num_heads, head_dim]
+        else:
+            q = q.transpose(0, 1).unsqueeze(0)
+            k = k.transpose(0, 1).unsqueeze(0).contiguous()
+            v = v.transpose(0, 1).unsqueeze(0).contiguous()
+            # q, k, v: [1, num_heads, seq_length, head_dim]
+
+            attn_outputs = []
+            for i in range(image_num):
+                start_idx = seq_lens[i]
+                end_idx = seq_lens[i + 1]
+                tmp_q = q[:, :, start_idx:end_idx, :]
+                tmp_k = k[:, :, start_idx:end_idx, :]
+                tmp_v = v[:, :, start_idx:end_idx, :]
+                attn_output = scaled_dot_product_attention(
+                    tmp_q, tmp_k, tmp_v,
+                    None, False
+                )
+                attn_output = attn_output.permute(0, 2, 1, 3)
+                # attn_output: [1, seq_length, num_heads, head_dim]
+                attn_outputs.append(attn_output)
+            attn_output = torch.cat(attn_outputs, dim=1).squeeze(0)
+            # attn_output: [seq_length, num_heads, head_dim]
+    else:
+        attention_mask = torch.full(
+            [1, seq_length, seq_length], torch.finfo(q.dtype).min, device=q.device, dtype=q.dtype
+        )
+        for i in range(1, len(seq_lens)):
+            attention_mask[..., seq_lens[i - 1]:seq_lens[i], seq_lens[i - 1]:seq_lens[i]] = 0
+
+        q = q.transpose(0, 1)
+        k = k.transpose(0, 1)
+        v = v.transpose(0, 1)
+        # q, k, v: [num_heads, seq_length, head_dim]
+
+        attn_weights = torch.matmul(q, k.transpose(1, 2)) / math.sqrt(head_dim)
+        attn_weights = attn_weights + attention_mask
+        attn_weights = attention_softmax(attn_weights)
+        attn_output = torch.matmul(attn_weights, v)
+        attn_output = attn_output.transpose(0, 1)
+        # attn_output: [seq_length, num_heads, head_dim]
+
+    attn_output = attn_output.reshape(seq_length, -1)
+    attn_output = self.proj(attn_output)
+    return attn_output

{ipex_llm-2.2.0b20250326.dist-info → ipex_llm-2.2.0b20250328.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250326
+Version: 2.2.0b20250328
 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 ==2.6.0b20250326 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250328 ; 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 ==2.6.0b20250326 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.6.0b20250328 ; (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 ==2.6.0b20250326 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250326 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250326 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250328 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250328 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250328 ; 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 ==2.6.0b20250326 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250326 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250326 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250328 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250328 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250328 ; 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 ==2.6.0b20250326 ; extra == 'xpu-2-6'
+Requires-Dist: bigdl-core-xe-all ==2.6.0b20250328 ; 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 ==2.6.0b20250326 ; extra == 'xpu-2-6-arl'
+Requires-Dist: bigdl-core-xe-all ==2.6.0b20250328 ; 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 ==2.6.0b20250326 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250326 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250326 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250328 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250328 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250328 ; 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 ==2.6.0b20250326 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250326 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250326 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250328 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250328 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250328 ; 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 ==2.6.0b20250326 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250326 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250326 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250328 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250328 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250328 ; 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-2.2.0b20250326.dist-info → ipex_llm-2.2.0b20250328.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=pqTVajDI-O28V6_tMMU5ylAZz6KnhkNU1hZF0qcQmXc,36352
-ipex_llm/libs/bloom.dll,sha256=wdaa-Xyja6Cxd-JvmkqWQCFHsIJ_J2d1RZG0kqQRSJI,507904
-ipex_llm/libs/gptneox-api.dll,sha256=84H-AtLwz-L1zKmGOkkUF91tg31RyB9ReRxOf9tjhd0,24576
-ipex_llm/libs/gptneox.dll,sha256=S-ekc_gepDq_w905_X0wVecKy1Su0bQ17Osb2wmyqWU,568320
-ipex_llm/libs/libbloom_avx.dll,sha256=aJMe0Vf9FkZE-oh5m0xuAPh6Td8OPUbWMbhmkPQ4G-k,536576
-ipex_llm/libs/libbloom_vnni.dll,sha256=kfZOWQHNDrpnBIyAf59vR1VCpMmHkoxR7Qqrmg7TeGk,508416
-ipex_llm/libs/libgptneox_avx.dll,sha256=TGGRYlBnWRh8B6UERRZaxsF3aKT4eG1bwZXstU_AukM,596992
-ipex_llm/libs/libgptneox_vnni.dll,sha256=xdYuHYCTFcvIzpbybOcoMl1R4C942aTIZLO08rP0uX8,568832
-ipex_llm/libs/libllama_avx.dll,sha256=idhrt9J7wVzgNe3VV6TzwecsQlPJiCcQoXOwg9lgFcQ,591360
-ipex_llm/libs/libllama_vnni.dll,sha256=1_zWBiKs3IqO4WXF7tFwi_x4roh-MhOntf0FAwI1rh0,563200
-ipex_llm/libs/libstarcoder_avx.dll,sha256=8aFkmm6I-ui50-2KxF-CRCpqvp9gPK05dNnVZ23Ka-A,627712
-ipex_llm/libs/libstarcoder_vnni.dll,sha256=_IcUGJmh3vAdlyybVZi3TGevTCjWKWo0xm3lh-a8kIs,599552
-ipex_llm/libs/llama-api.dll,sha256=DMn2bTbxybghbfbhi4A30l1SwnPm-1yUQhfHKKUA2zI,25600
-ipex_llm/libs/llama.dll,sha256=9KDpAN-a92Ei93xwyZ6q0vlQvgw0Vtx6Du6xKbIOU4k,562688
-ipex_llm/libs/main-bloom.exe,sha256=TE0o7kdnDHBQ-TdFtFOHNRuW6iZLOaQkkMgODnLmfdo,103424
-ipex_llm/libs/main-gptneox.exe,sha256=AIz6Pud8TGqjeCyqt7PpgIkSoJkfrHVFEk_MADVL-XQ,98816
-ipex_llm/libs/main-llama.exe,sha256=UXD50b7iFZBMFNNutn8he0b1WCrVZMLQyls-UZeWDvQ,99840
-ipex_llm/libs/main-starcoder.exe,sha256=vfLp8wknjBhAyk5dEnd_SxZyc35hCw-UTBIQTSgxRYg,157696
-ipex_llm/libs/pipeline.dll,sha256=KSAXxye5J9pn3DM_jegYio4cKpCu42cByqyJyD735zQ,73216
-ipex_llm/libs/quantize-bloom.exe,sha256=lv21nEsnw5JKCyYvS311zpBW4siXoi0d11tmJMDDO9I,126464
-ipex_llm/libs/quantize-bloom_vnni.exe,sha256=pW0feE2Egqo8hdnE8BgYYH6ulZS17oGEoRAaWdh94ro,128000
-ipex_llm/libs/quantize-gptneox.exe,sha256=Dux1lopaaO1L76CJaXKdAyl4pT3hv9M2gkOuysrRHiQ,104448
-ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=fsBLc0pl7r7VbuYipbKNfZ73uTO3V_Re-_2SvNW1bn8,104960
-ipex_llm/libs/quantize-llama.exe,sha256=eoWV68mg5peDopzK5FwKH3E8hQwu13LgL76inE2A2rA,110080
-ipex_llm/libs/quantize-llama_vnni.exe,sha256=C_q5J5vMWKoqjF529FwL6kUbOBgjZzmPXk_QJcFnUZo,110592
-ipex_llm/libs/quantize-starcoder.exe,sha256=zXb514xNlefTjlyOGYLmFjGiAFEaHtYb4Vjvzv4vcyA,127488
-ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=EgLdrKE7R-EH7NKL9iRP5-5BP6c6fNia8mqSYgCZynU,128512
-ipex_llm/libs/starcoder-api.dll,sha256=TozlAJewwHxElGpYiew2naU7HLHZPTFu0pgUpOwecik,21504
-ipex_llm/libs/starcoder.dll,sha256=H9fiXMXtmwIgUZjeN9g71q0Ab8rxQKU9_Yn3Tejh7g8,599040
+ipex_llm/libs/bloom-api.dll,sha256=Ts41ETMTULbFvlg9mMqw084zBha5hP5T1gHw9HZkf1E,36352
+ipex_llm/libs/bloom.dll,sha256=BMXbtcBpiyARbOh9fZb4pfPEZLqAvtqCzSPhbQsxCdA,507904
+ipex_llm/libs/gptneox-api.dll,sha256=r_qjHto49nA4Ik2LZUsF2zpcQduIbgociD6UaH1HtAU,24576
+ipex_llm/libs/gptneox.dll,sha256=q1_5H7bxVe0dC4NmnpLMHAtuNdQuQQIBUKGrG2sgx2k,568320
+ipex_llm/libs/libbloom_avx.dll,sha256=Hop7_52kTcnPquuH1MvUb-c1halOLK0F56NNdNjyMOo,536576
+ipex_llm/libs/libbloom_vnni.dll,sha256=L-5Gd0Derh-6jffrFd-qfIQkst9SVfLGYwXUAVyQlbg,508416
+ipex_llm/libs/libgptneox_avx.dll,sha256=BKByiQHeKNZuRhx62VVgjq-aracBadJOUTEMz13onNw,596992
+ipex_llm/libs/libgptneox_vnni.dll,sha256=sNpCWU6uQp8fUDt0fljX6KDwSDnuUTdvGHhkxbHZOO8,568832
+ipex_llm/libs/libllama_avx.dll,sha256=K2CC5yda4M1NgobPCXZcJCUt8pgoJZraAzqIFUugmJ8,591360
+ipex_llm/libs/libllama_vnni.dll,sha256=klgeb_4tsCBxmt0cE1Yae-drQkacd8VsT-HmuUdJ5hk,563200
+ipex_llm/libs/libstarcoder_avx.dll,sha256=L2xrqzfee1-HNEnmbSHOKkG6p34CFiyJ8k3r5b3mMCQ,627712
+ipex_llm/libs/libstarcoder_vnni.dll,sha256=Z0k9K3VqMFqnr3vLhACbyRnLcJBNt1pghE6ZsbIwB7o,599552
+ipex_llm/libs/llama-api.dll,sha256=ZOtLxLBbxY5LmtMOIZijetNIMpX_1a-RaYj0NobdtFw,25600
+ipex_llm/libs/llama.dll,sha256=DvzfMBusPfpfOMJdyDu6iDJeKT-dBfuN1566Q6BCqwI,562688
+ipex_llm/libs/main-bloom.exe,sha256=dTu_LbEK4w6bXeFF__Dwpj76QNyS-DZRN-khVoqnKHE,103424
+ipex_llm/libs/main-gptneox.exe,sha256=v-uJaAaaL3qMR-r1eddj9JisqZeJSGXxK61HP01JITs,98816
+ipex_llm/libs/main-llama.exe,sha256=pv0qyPACjPpsA_MQ17oGVQc4F3q8fGP9ip866XKdKag,99840
+ipex_llm/libs/main-starcoder.exe,sha256=jds-PpbYmDHT8FcYDtf7p-KoC3y9JLjxQAZvqjCQEYI,157696
+ipex_llm/libs/pipeline.dll,sha256=wkTpoUoilbta0wEhCuOdwuntJ4Qyp-L6Qvj43WhdXnY,73216
+ipex_llm/libs/quantize-bloom.exe,sha256=d7tG88PlzBNQw87qvzbSaxM30iRJQghbH7bqPd-RoYU,126464
+ipex_llm/libs/quantize-bloom_vnni.exe,sha256=8uH4onmTtaS37KFnT9uO-077T5-OjQj1bswR4QjJgOc,128000
+ipex_llm/libs/quantize-gptneox.exe,sha256=UwwX002f7N6ucLI3rTtJGPzPjkuVJSN_OQnPzMGvYk8,104448
+ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=W826KPed7EFGamySWnFySY8Fn0yHvjx2ZJL5wXDbDGE,104960
+ipex_llm/libs/quantize-llama.exe,sha256=90S6ZcTtVANWWqxZq8ObYIaiYt8WdiOuGZZJbXK0tI0,110080
+ipex_llm/libs/quantize-llama_vnni.exe,sha256=4RfT-4FEytTcRPK7OAcKQmz24e-Ej3SWfb8SyXxVhxE,110592
+ipex_llm/libs/quantize-starcoder.exe,sha256=FyItO9Doa6qKNQPqM0AuTL8PhGyntdpcVddZU_b-q18,127488
+ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=2Ft41uF6zfOOCnrytztIvb-pw_i_7etfGIQ9jVD-qwY,128512
+ipex_llm/libs/starcoder-api.dll,sha256=Z0OH30QyHbQbKlohSLBNDgEj0GJbieTACkzSnHVDfno,21504
+ipex_llm/libs/starcoder.dll,sha256=9bCpmwwgIa6A61GlkEsRDaMfq8j4KMUmCFhR_p15kaU,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
@@ -87,7 +87,7 @@ 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=CfdQz1wDgjmAT4uT_y9ZRbOrFjmsXxFUgPeWEExskk8,103770
+ipex_llm/transformers/convert.py,sha256=D3QSoD48NzOKz9rKIsgrI0sNzdmJvb0sP4k_0f8Lcmo,105824
 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
@@ -171,6 +171,7 @@ 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_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
@@ -255,11 +256,11 @@ 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-2.2.0b20250326.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
-ipex_llm-2.2.0b20250326.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
-ipex_llm-2.2.0b20250326.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
-ipex_llm-2.2.0b20250326.dist-info/METADATA,sha256=fHSU2c7G6TDPCkVikxp0pUvb-oq-QgVvbFVE1Ea17BM,13917
-ipex_llm-2.2.0b20250326.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
-ipex_llm-2.2.0b20250326.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
-ipex_llm-2.2.0b20250326.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
-ipex_llm-2.2.0b20250326.dist-info/RECORD,,
+ipex_llm-2.2.0b20250328.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
+ipex_llm-2.2.0b20250328.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
+ipex_llm-2.2.0b20250328.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
+ipex_llm-2.2.0b20250328.dist-info/METADATA,sha256=94dtiTavz1mtEX8Bd4VuZtum6co-x8W9ZaITI8DqM7w,13917
+ipex_llm-2.2.0b20250328.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
+ipex_llm-2.2.0b20250328.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
+ipex_llm-2.2.0b20250328.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
+ipex_llm-2.2.0b20250328.dist-info/RECORD,,