ipex-llm 2.2.0b20250108__py3-none-manylinux2010_x86_64.whl → 2.2.0b20250110__py3-none-manylinux2010_x86_64.whl

ipex_llm/transformers/models/qwen2.py

@@ -51,217 +51,14 @@ from ipex_llm.transformers.models.utils import use_quantize_kv_cache, \
     should_use_compresskv, is_enough_kv_cache_room_4_36
 from ipex_llm.transformers.kv import DynamicFp8Cache, DynamicNormalCache, \
     DynamicCompressCache, DynamicCompressFp8Cache
-from ipex_llm.utils.common import invalidInputError
 
-from transformers.models.qwen2.modeling_qwen2 import Qwen2Attention, Qwen2MLP
+from transformers.models.qwen2.modeling_qwen2 import Qwen2Model, Qwen2Attention, Qwen2MLP
 from transformers.models.qwen2.modeling_qwen2 import apply_rotary_pos_emb
 from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
 from transformers.cache_utils import Cache
-from transformers import logging
-
-
-logger = logging.get_logger(__name__)
 
 
 def qwen2_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,  # for transformers >= 4.42
-) -> 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
-    )
-    use_cache = use_cache if use_cache is not None else self.config.use_cache
-
-    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-
-    # retrieve input_ids and inputs_embeds
-    if input_ids is not None and inputs_embeds is not None:
-        invalidInputError(False,
-                          "You cannot specify both input_ids and inputs_embeds at the same time")
-    elif input_ids is not None:
-        batch_size, seq_length = input_ids.shape
-    elif inputs_embeds is not None:
-        batch_size, seq_length, _ = inputs_embeds.shape
-    else:
-        invalidInputError(False,
-                          "You have to specify either decoder_input_ids or decoder_inputs_embeds")
-
-    if self.gradient_checkpointing and self.training:
-        if use_cache:
-            logger.warning_once(
-                "`use_cache=True` is incompatible with gradient checkpointing. "
-                "Setting `use_cache=False`..."
-            )
-            use_cache = False
-
-    past_key_values_length = 0
-
-    # ipex-llm changes start
-    # IPEX-LLM OPT: kv cache and quantize kv cache
-    inputs = input_ids if input_ids is not None else inputs_embeds
-    use_quantize_kv = (
-        self.config.hidden_size != 3584     # disable quantize kv in specific model
-        and use_quantize_kv_cache(self.layers[0].mlp.up_proj, inputs,
-                                  self.config.num_attention_heads//self.config.num_key_value_heads)
-    )
-    use_compress_kv = should_use_compresskv(inputs, inputs.shape[1]) or \
-        isinstance(past_key_values, DynamicCompressCache)
-
-    if use_cache:
-        if use_compress_kv and not isinstance(past_key_values, DynamicCompressCache):
-            if use_quantize_kv:
-                past_key_values = DynamicCompressFp8Cache.from_legacy_cache(past_key_values)
-            else:
-                past_key_values = DynamicCompressCache.from_legacy_cache(past_key_values)
-        elif use_quantize_kv and not use_compress_kv and not isinstance(past_key_values,
-                                                                        DynamicFp8Cache):
-            past_key_values = DynamicFp8Cache.from_legacy_cache(past_key_values)
-        if not use_quantize_kv and not use_compress_kv and not isinstance(past_key_values,
-                                                                          DynamicNormalCache):
-            past_key_values = DynamicNormalCache.from_legacy_cache(past_key_values)
-        past_key_values_length = past_key_values.get_usable_length(seq_length)
-    # ipex-llm changes end
-
-    if position_ids is None:
-        device = input_ids.device if input_ids is not None else inputs_embeds.device
-        position_ids = torch.arange(
-            past_key_values_length, seq_length + past_key_values_length,
-            dtype=torch.long, device=device
-        )
-        position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
-    else:
-        position_ids = position_ids.view(-1, seq_length).long()
-
-    if inputs_embeds is None:
-        inputs_embeds = self.embed_tokens(input_ids)
-
-    flash_attn_2 = self._attn_implementation == "flash_attention_2"
-    if attention_mask is not None and flash_attn_2 and use_cache:
-
-        is_padding_right = attention_mask[:, -1].sum().item() != batch_size
-        if is_padding_right:
-            invalidInputError(
-                False,
-                "You are attempting to perform batched generation with padding_side='right'"
-                " this may lead to unexpected behaviour for Flash Attention version of Qwen2."
-                " Make sure to  call `tokenizer.padding_side  = 'left'` before tokenizing "
-                "the input. "
-            )
-
-    from transformers.models.qwen2.modeling_qwen2 import _prepare_4d_causal_attention_mask_for_sdpa
-    from transformers.models.qwen2.modeling_qwen2 import _prepare_4d_causal_attention_mask
-
-    # ipex-llm changes start: don't generate `attention_mask` in decode phase
-    if seq_length == 1:
-        attention_mask = None
-    # ipex-llm changes end
-    elif self._attn_implementation == "flash_attention_2":
-        # 2d mask is passed through the layers
-        attention_mask = attention_mask if (attention_mask is not None and
-                                            0 in attention_mask) else None
-    elif self._attn_implementation == "sdpa" and not output_attentions:
-        # output_attentions=True can not be supported when using SDPA, and we fall back on
-        # the manual implementation that requires a 4D causal mask in all cases.
-        attention_mask = _prepare_4d_causal_attention_mask_for_sdpa(
-            attention_mask,
-            (batch_size, seq_length),
-            inputs_embeds,
-            past_key_values_length,
-        )
-    else:
-        # 4d mask is passed through the layers
-        attention_mask = _prepare_4d_causal_attention_mask(
-            attention_mask,
-            (batch_size, seq_length),
-            inputs_embeds,
-            past_key_values_length,
-            sliding_window=self.config.sliding_window,
-        )
-
-    hidden_states = inputs_embeds
-
-    # 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,)
-
-        if self.gradient_checkpointing and self.training:
-            layer_outputs = self._gradient_checkpointing_func(
-                decoder_layer.__call__,
-                hidden_states,
-                attention_mask,
-                position_ids,
-                past_key_values,
-                output_attentions,
-                use_cache,
-            )
-        else:
-            # ipex-llm changes
-            curr_device = decoder_layer.input_layernorm.weight.device
-            if attention_mask is not None:
-                attention_mask = attention_mask.to(curr_device)
-            if position_ids is not None:
-                position_ids = position_ids.to(curr_device)
-            # ipex-llm changes end
-            layer_outputs = decoder_layer(
-                hidden_states,
-                attention_mask=attention_mask,
-                position_ids=position_ids,
-                past_key_value=past_key_values,
-                output_attentions=output_attentions,
-                use_cache=use_cache,
-            )
-
-        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,)
-
-    # ipex-llm changes start: remove `to_legacy_cache`
-    next_cache = None
-    if use_cache:
-        next_cache = next_decoder_cache
-    # ipex-llm changes end
-
-    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_model_forward_4_42(
     self,
     input_ids: torch.LongTensor = None,
     attention_mask: Optional[torch.Tensor] = None,
@@ -274,43 +71,17 @@ def qwen2_model_forward_4_42(
     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
-    )
+    # IPEX-LLM OPT start: kv cache and quantize kv cache
+    inputs = input_ids if input_ids is not None else inputs_embeds
     use_cache = use_cache if use_cache is not None else self.config.use_cache
+    use_cache = True if inputs.device.type == "xpu" else use_cache
 
-    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 cannot specify both input_ids and inputs_embeds at the same time, "
-        "and must specify either one"
+    use_quantize_kv = self.config.hidden_size != 3584 and use_quantize_kv_cache(
+        self.layers[0].mlp.down_proj, inputs,
+        self.config.num_attention_heads, self.config.num_key_value_heads
     )
 
-    if self.gradient_checkpointing and self.training:
-        if use_cache:
-            logger.warning_once(
-                "`use_cache=True` is incompatible with gradient checkpointing. "
-                "Setting `use_cache=False`..."
-            )
-            use_cache = False
-
-    if inputs_embeds is None:
-        inputs_embeds = self.embed_tokens(input_ids)
-
-    # ipex-llm changes start
-    # IPEX-LLM OPT: kv cache and quantize kv cache
-    use_quantize_kv = (
-        self.config.hidden_size != 3584     # disable quantize kv in specific model
-        and use_quantize_kv_cache(self.layers[0].mlp.up_proj, inputs_embeds,
-                                  self.config.num_attention_heads//self.config.num_key_value_heads)
-    )
-    use_compress_kv = should_use_compresskv(inputs_embeds, inputs_embeds.shape[1]) or \
+    use_compress_kv = should_use_compresskv(inputs, inputs.shape[1]) or \
         isinstance(past_key_values, DynamicCompressCache)
 
     if use_cache:
@@ -327,79 +98,24 @@ def qwen2_model_forward_4_42(
             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
-        )
-    if position_ids is None:
-        position_ids = cache_position.unsqueeze(0)
-
-    causal_mask = self._update_causal_mask(
-        attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions
-    )
-
-    hidden_states = inputs_embeds
-
-    # 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,)
-
-        if self.gradient_checkpointing and self.training:
-            layer_outputs = self._gradient_checkpointing_func(
-                decoder_layer.__call__,
-                hidden_states,
-                causal_mask,
-                position_ids,
-                past_key_values,
-                output_attentions,
-                use_cache,
-                cache_position,
-            )
-        else:
-            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,
-            )
-
-        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,)
-
-    # ipex-llm changes start: remove `to_legacy_cache`
-    next_cache = None
-    if use_cache:
-        next_cache = next_decoder_cache
-    # ipex-llm changes end
+    # `cache_position` is required after transformers 4.42
+    if cache_position is not None:
+        kwargs = {"cache_position": cache_position}
+    else:
+        kwargs = {}
 
-    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,
+    return Qwen2Model.forward(
+        self=self,
+        input_ids=input_ids,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        return_dict=return_dict,
+        **kwargs
     )
 
 

ipex_llm/transformers/models/qwen2_moe.py

@@ -73,8 +73,10 @@ def qwen2moe_model_forward(
     return_dict: Optional[bool] = None,
 ):
     use_cache = use_cache if use_cache is not None else self.config.use_cache
-    input = input_ids if input_ids is not None else inputs_embeds
-    use_quantize_kv = use_quantize_kv_cache(self.layers[0].mlp.shared_expert.up_proj, input)
+    inputs = input_ids if input_ids is not None else inputs_embeds
+    num_heads, num_kv_heads = self.config.num_attention_heads, self.config.num_key_value_heads
+    use_quantize_kv = use_quantize_kv_cache(self.layers[0].mlp.shared_expert.up_proj, inputs,
+                                            num_heads, num_kv_heads)
     if use_cache:
         if use_quantize_kv and not isinstance(past_key_values, DynamicFp8Cache):
             past_key_values = DynamicFp8Cache.from_legacy_cache(past_key_values)

ipex_llm/transformers/models/qwen2_vl.py

@@ -88,7 +88,9 @@ def qwen2_vl_model_forward(
     # IPEX-LLM OPT start: kv cache and quantize kv cache
     inputs = input_ids if input_ids is not None else inputs_embeds
     use_cache = True if inputs.device.type == "xpu" else use_cache
-    use_quantize_kv = use_quantize_kv_cache(self.layers[0].mlp.down_proj, inputs)
+    num_heads, num_kv_heads = self.config.num_attention_heads, self.config.num_key_value_heads
+    use_quantize_kv = use_quantize_kv_cache(self.layers[0].mlp.down_proj, inputs,
+                                            num_heads, num_kv_heads)
     if use_cache:
         if use_quantize_kv and not isinstance(past_key_values, DynamicFp8Cache):
             past_key_values = DynamicFp8Cache.from_legacy_cache(past_key_values)

ipex_llm/transformers/models/stablelm.py

@@ -69,8 +69,10 @@ def stablelm_model_forward(
 ):
     # IPEX-LLM OPT: kv cache and quantize kv cache
     use_cache = use_cache if use_cache is not None else self.config.use_cache
+    num_heads, num_kv_heads = self.config.num_attention_heads, self.config.num_key_value_heads
     use_quantize_kv = (self.layers[0].self_attn.head_dim in [64, 80, 96, 128]
-                       and use_quantize_kv_cache(self.layers[0].mlp.up_proj, input_ids))
+                       and use_quantize_kv_cache(self.layers[0].mlp.up_proj, input_ids,
+                                                 num_heads, num_kv_heads))
     if use_cache:
         if use_quantize_kv and not isinstance(past_key_values, DynamicFp8Cache):
             past_key_values = DynamicFp8Cache.from_legacy_cache(past_key_values)

ipex_llm/transformers/models/starcoder2.py

@@ -132,7 +132,9 @@ def model_forward(
     return_dict: Optional[bool] = None,
 ):
     use_cache = use_cache if use_cache is not None else self.config.use_cache
-    use_quantize_kv = use_quantize_kv_cache(self.layers[0].mlp.c_fc, input_ids)
+    num_heads, num_kv_heads = self.config.num_attention_heads, self.config.num_key_value_heads
+    use_quantize_kv = use_quantize_kv_cache(self.layers[0].mlp.c_fc, input_ids,
+                                            num_heads, num_kv_heads)
     if use_cache:
         if use_quantize_kv and not isinstance(past_key_values, DynamicFp8Cache):
             past_key_values = DynamicFp8Cache.from_legacy_cache(past_key_values)

ipex_llm/transformers/models/utils.py

@@ -74,7 +74,8 @@ def append_kv_cache(cache_k, cache_v, key_states, value_states):
     return new_cache_k, new_cache_v
 
 
-def use_quantize_kv_cache(linear: torch.nn.Module, x: torch.Tensor, kv_group: int = 1) -> bool:
+def use_quantize_kv_cache(linear: torch.nn.Module, x: torch.Tensor,
+                          num_heads: int, num_kv_heads: int) -> bool:
     if os.environ.get("BIGDL_QUANTIZE_KV_CACHE", None) is not None:
         warnings.warn(
             "`BIGDL_QUANTIZE_KV_CACHE` is deprecated and will be removed in future releases. "
@@ -90,8 +91,11 @@ def use_quantize_kv_cache(linear: torch.nn.Module, x: torch.Tensor, kv_group: in
     else:
         device_name = get_xpu_device_name(x.device)
         return (
-            device_name in ["mtl", "lnl", "arl"] and kv_group == 1
-            or device_name in ["arc", "bmg"] and x.size(0) > 1
+            num_kv_heads >= 4
+            and (
+                device_name in ["mtl", "lnl", "arl"] and num_heads // num_kv_heads <= 4
+                or device_name in ["arc", "bmg"] and x.size(0) > 1
+            )
         )
 
 
@@ -268,26 +272,6 @@ def use_xmx(x: torch.Tensor, qtype: int):
     )
 
 
-def fp16_fusion_check(proj, x, training):
-    # only use fp16 fusion on PVC inference
-    if proj is None:
-        return False
-    if not hasattr(proj, "qtype"):
-        return False
-    if proj.qtype != ggml_tensor_qtype["fp16"]:
-        return False
-    if proj.weight_type != 2:
-        return False
-    if training:
-        return False
-    if x.requires_grad:
-        return False
-    device_type = get_xpu_device_name(x.device)
-    if device_type != "pvc":
-        return False
-    return True
-
-
 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/models/yuan.py

@@ -158,7 +158,8 @@ def yuan_attention_forward(
                                                         "yuan")
 
     # IPEX-LLM OPT: kv cache and quantzie kv cache
-    use_quantize_kv = use_quantize_kv_cache(self.qk_proj, hidden_states)
+    use_quantize_kv = use_quantize_kv_cache(self.qk_proj, hidden_states,
+                                            self.num_heads, self.num_heads)
     key_states, value_states = update_past_key_value(
         None if past_key_value is None else (past_key_value[0], past_key_value[1]),
         key_states, value_states,

ipex_llm/transformers/npu_model.py

@@ -182,13 +182,17 @@ class _BaseAutoModelClass:
         if hasattr(model, "config") and model.config.model_type == "glm":
             # convert to llama structure
             from .npu_models.glm_edge import convert_config, load_weights, convert_state_dict
-            import json
             original_path = model.config._name_or_path
             del model
 
-            with open(os.path.join(original_path, "config.json")) as f:
-                original_config = json.load(f)
+            original_config, _ = PretrainedConfig.get_config_dict(original_path)
             config = convert_config(original_config)
+
+            if not os.path.isdir(original_path):
+                # all model files are already cached
+                from transformers.utils.hub import cached_file
+                resolved_file = cached_file(original_path, "config.json")
+                original_path = os.path.dirname(resolved_file)
             original_state_dict = load_weights(original_path)
             new_dict, _ = convert_state_dict(original_state_dict, config,
                                              original_config.get("partial_rotary_factor", 1.0),

{ipex_llm-2.2.0b20250108.dist-info → ipex_llm-2.2.0b20250110.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250108
+Version: 2.2.0b20250110
 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 ==2.6.0b20250108 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250110 ; extra == 'cpp'
 Requires-Dist: setuptools ; extra == 'cpp'
 Provides-Extra: cpp-arl
-Requires-Dist: bigdl-core-cpp ==2.6.0b20250108 ; extra == 'cpp-arl'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250110 ; 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 ==2.6.0b20250108 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.6.0b20250110 ; (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 ==2.6.0b20250108 ; 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'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250110 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250110 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250110 ; 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 ==2.6.0b20250108 ; 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: bigdl-core-xe-21 ==2.6.0b20250110 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250110 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250110 ; 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 ==2.6.0b20250108 ; extra == 'xpu-2-6'
+Requires-Dist: bigdl-core-xe-all ==2.6.0b20250110 ; 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 ==2.6.0b20250108 ; 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: bigdl-core-xe-23 ==2.6.0b20250110 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250110 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250110 ; 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 ==2.6.0b20250108 ; 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: bigdl-core-xe-23 ==2.6.0b20250110 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250110 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250110 ; 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 ==2.6.0b20250108 ; 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: bigdl-core-xe-23 ==2.6.0b20250110 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250110 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250110 ; 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'