ipex-llm 2.2.0b20250109__py3-none-manylinux2010_x86_64.whl → 2.2.0b20250111__py3-none-manylinux2010_x86_64.whl

ipex_llm/transformers/convert.py

@@ -1590,6 +1590,9 @@ def _optimize_post(model):
         convert_forward(model,
                         module.Qwen2ForCausalLM,
                         qwen2_causal_lm_forward)
+        convert_forward(model,
+                        module.Qwen2Model,
+                        qwen2_model_forward)
         convert_forward(model,
                         module.Qwen2RMSNorm,
                         rms_norm_forward)
@@ -1602,12 +1605,6 @@ def _optimize_post(model):
         convert_forward(model,
                         module.Qwen2SdpaAttention,
                         qwen2_attention_forward)
-        if version.parse(trans_version) >= version.parse("4.42"):
-            from ipex_llm.transformers.models.qwen2 import qwen2_model_forward_4_42
-            convert_forward(model, module.Qwen2Model, qwen2_model_forward_4_42)
-        else:
-            from ipex_llm.transformers.models.qwen2 import qwen2_model_forward
-            convert_forward(model, module.Qwen2Model, qwen2_model_forward)
     elif model.config.model_type == "qwen2_moe":
         # for Qwen1.5-MOE-A2.7B
         modeling_module_name = model.__class__.__module__
@@ -1819,9 +1816,7 @@ def _optimize_post(model):
         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
+        convert_forward(model, module.Phi3MLP, mlp_silu_forward)
         convert_forward(model, module.Phi3RMSNorm, rms_norm_forward)
         if model.config.model_type == "phi3":
             from ipex_llm.transformers.models.phi3 import phi3_model_forward_wrapper

ipex_llm/transformers/convert_ipex.py

@@ -52,7 +52,14 @@ import os
 
 
 def _ipex_optimize_rmsnorm(_model, supported_classes, is_tpp=False, is_woq=False):
-    from intel_extension_for_pytorch.transformers.models.cpu.fusions.mha_fusion import _IPEXRMSNorm
+    try:
+        # old version use name `_IPEXRMSNorm`
+        from intel_extension_for_pytorch.transformers.models.cpu.fusions.mha_fusion \
+            import _IPEXRMSNorm
+    except ImportError:
+        # new version use name `_IPEXRMSNormCPU`
+        from intel_extension_for_pytorch.transformers.models.cpu.fusions.mha_fusion \
+            import _IPEXRMSNormCPU as _IPEXRMSNorm
     for supported_class in supported_classes:
         lowering_class_cpu(
             _model,

ipex_llm/transformers/low_bit_linear.py

@@ -47,7 +47,7 @@ import os
 import torch
 import torch.distributed
 import torch.nn.functional as F
-from torch import Tensor, device, dtype, nn
+from torch import Tensor, dtype, nn
 from operator import mul
 from functools import reduce
 from ipex_llm.transformers.xpu_customize_fwd import custom_fwd, custom_bwd
@@ -294,10 +294,10 @@ def use_batch_forward(x: torch.Tensor, qtype: int, output_len: int):
     if hard_condition:
         return (
             batch_size > 1
-            or (device in ["arc"] and qtype in [SYM_INT8, FP4])
-            or (device in ["arc", "mtl"] and qtype in [FP8E4])
-            or (device in ["lnl"] and qtype in [SYM_INT4] and x.shape[1] % 512 == 0)
-            or (device in ["bmg"] and qtype in [SYM_INT4, FP8E5])
+            or (device_name in ["arc"] and qtype in [SYM_INT8, FP4])
+            or (device_name in ["arc", "mtl"] and qtype in [FP8E4])
+            or (device_name in ["lnl"] and qtype in [SYM_INT4] and x.shape[1] % 512 == 0)
+            or (device_name in ["bmg"] and qtype in [SYM_INT4, FP8E5])
         )
     return False
 

ipex_llm/transformers/models/baichuan.py

@@ -30,8 +30,7 @@ from ipex_llm.transformers.models.utils import use_quantize_kv_cache, restore_fp
 from ipex_llm.transformers.models.utils import update_past_key_value
 from ipex_llm.transformers.models.utils import should_use_fuse_rope
 from ipex_llm.transformers.models.utils import use_sdp
-from ipex_llm.transformers.models.utils import apply_rotary_pos_emb, SILU
-from ipex_llm.transformers.models.utils import mlp_fusion_check
+from ipex_llm.transformers.models.utils import apply_rotary_pos_emb
 from ipex_llm.transformers.models.utils import is_enough_kv_cache_room_4_36
 from ipex_llm.transformers.kv import DynamicCompressFp8Cache, DynamicCompressCache
 import warnings

ipex_llm/transformers/models/internlm.py

@@ -113,21 +113,6 @@ def internlm_attention_forward(
     return attn_output, attn_weights, past_key_value
 
 
-def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
-    """
-    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep).
-    The hidden states go from (batch,
-    num_key_value_heads, seqlen, head_dim) to
-    (batch, num_attention_heads, seqlen, head_dim)
-    """
-    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
-    if n_rep == 1:
-        return hidden_states
-    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads,
-                                                           n_rep, slen, head_dim)
-    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
-
-
 def internlm2_attention_forward(
     self,
     hidden_states: torch.Tensor,

ipex_llm/transformers/models/phi3.py

@@ -39,7 +39,6 @@ import warnings
 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 should_use_fuse_rope, rotate_half
-from ipex_llm.transformers.models.utils import mlp_fusion_check, SILU
 from ipex_llm.transformers.models.utils import use_sdp, use_sdp_causal
 from ipex_llm.transformers.models.utils import use_quantize_kv_cache, restore_fp8_kv_cache
 from ipex_llm.transformers.models.utils import should_use_compresskv, is_enough_kv_cache_room_4_36
@@ -213,24 +212,8 @@ def split_mlp(module: torch.nn.Module):
 
         del module.gate_up_proj
 
-
-def mlp_forward(
-    self,
-    hidden_states: torch.FloatTensor
-) -> torch.FloatTensor:
-    x_2d = hidden_states.view(-1, hidden_states.shape[-1])
-    qtype = getattr(self.gate_proj, "qtype", None)
-    if mlp_fusion_check(x_2d, qtype, self.training):
-        x_2d = x_2d.contiguous()
-        import xe_linear
-        return self.down_proj(xe_linear.mlp_forward_xpu(
-            x_2d, self.gate_proj.weight.data, self.up_proj.weight.data,
-            x_2d.shape[0], x_2d.shape[1], self.gate_proj.out_features,
-            SILU, qtype
-        ))
-    return self.down_proj(
-        self.activation_fn(self.gate_proj(hidden_states)) * self.up_proj(hidden_states)
-    )
+        # rename activation function
+        module.act_fn = module.activation_fn
 
 
 def phi3_model_forward_wrapper(origin_model_forward):

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
-    num_heads, num_kv_heads = self.config.num_attention_heads, self.config.num_key_value_heads
-    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, num_heads, num_kv_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,
@@ -274,44 +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
-    num_heads, num_kv_heads = self.config.num_attention_heads, self.config.num_key_value_heads
-    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,
-                                  num_heads, num_kv_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:
@@ -328,79 +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/utils.py

@@ -272,26 +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/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-2.2.0b20250109.dist-info → ipex_llm-2.2.0b20250111.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250109
+Version: 2.2.0b20250111
 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 ==2.6.0b20250109 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250111 ; extra == 'cpp'
 Requires-Dist: setuptools ; extra == 'cpp'
 Provides-Extra: cpp-arl
-Requires-Dist: bigdl-core-cpp ==2.6.0b20250109 ; extra == 'cpp-arl'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250111 ; 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 ==2.6.0b20250109 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.6.0b20250111 ; (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 ==2.6.0b20250109 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250109 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250109 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250111 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250111 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250111 ; 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 ==2.6.0b20250109 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250109 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250109 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250111 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250111 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250111 ; 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 ==2.6.0b20250109 ; extra == 'xpu-2-6'
+Requires-Dist: bigdl-core-xe-all ==2.6.0b20250111 ; 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 ==2.6.0b20250109 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250109 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250109 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250111 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250111 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250111 ; 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 ==2.6.0b20250109 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250109 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250109 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250111 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250111 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250111 ; 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 ==2.6.0b20250109 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250109 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250109 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250111 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250111 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250111 ; 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-2.2.0b20250109.dist-info → ipex_llm-2.2.0b20250111.dist-info}/RECORD

@@ -94,17 +94,17 @@ 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=umI137wqV2d4itS0AJQoZcygeWBATpSJSDJ805cZ-SY,98499
-ipex_llm/transformers/convert_ipex.py,sha256=iKXo0n8fVFTOA2fNYYrByMFK0dovL-kLd2sVDk88AlQ,14334
+ipex_llm/transformers/convert.py,sha256=i2IOmDnQBKNtcfUL95l8w4rNBMiU4SqX_9uz_LtPHMI,98086
+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
 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=Obdd08D9dvuroS_6XWo4DXO_DrNRsbAqjz-mQAHmfxY,40845
+ipex_llm/transformers/low_bit_linear.py,sha256=QBHrAG7lgOgVO1LHPNlimn8Icm44kEpnWOLtVuHoHDA,40857
 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
+ipex_llm/transformers/npu_model.py,sha256=X8ZtvZJpzz64XrSPhUYXXZmdJcbZ9X6G3Vlzw-zgN1Q,39749
 ipex_llm/transformers/patches.py,sha256=halPWm__ORh2fRFSIFPiCNg3LQBfrRkTPtmtRpBJCZQ,1286
 ipex_llm/transformers/pipeline_parallel.py,sha256=uNZpOXljNmdoEYnP8U-VFiN4dRZb2piQbIf2bG9LQnE,49051
 ipex_llm/transformers/qlora.py,sha256=jtPGsvWFjbTUGzDBCdfftnCis_0nJQNRpACSwXUbbGU,14943
@@ -144,7 +144,7 @@ ipex_llm/transformers/gguf/models/model_implement/yuan2/configuration_yuan.py,sh
 ipex_llm/transformers/gguf/models/model_implement/yuan2/yuan_hf_model.py,sha256=_AOGMV65XHxgTxIib7lgs49InopcecTzRwgtYR8NTUg,51084
 ipex_llm/transformers/models/__init__.py,sha256=tp2DcVkKg1-QvdYk7DY7rZvQWCDQ4ZjU8NAQ7Fclrpg,584
 ipex_llm/transformers/models/aquila.py,sha256=VZb5Drpo_fTxwcExZ397LygnsNPX2sVbie9_JeFudZI,5252
-ipex_llm/transformers/models/baichuan.py,sha256=cAQLmVG-3R8CSTGTcDy2JOOzVe-Ej8AXjIEIjvZBGlo,18376
+ipex_llm/transformers/models/baichuan.py,sha256=8b43mBRZJEf_xLNoodhA4r9x1anqwC3Wt8awWel-aUo,18306
 ipex_llm/transformers/models/bert.py,sha256=0Mm9jkvkzBxtc_z_GE1TcZoPz-HOg2Z2973ZEWgSwJk,5601
 ipex_llm/transformers/models/bloom.py,sha256=PxfzyYT-nFn3K5rZhTQjmcEjUUzAhUFzxIN4kzRlCuc,8103
 ipex_llm/transformers/models/chatglm.py,sha256=UHai1t2AUtGmF765_eHF8LUMVQzp_oCBx8TJB21WrHk,12597
@@ -159,7 +159,7 @@ ipex_llm/transformers/models/glm.py,sha256=lmeEWd_W2O638VzVW4Gm6cJre5XZcg_QBmPs8
 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=OifyiobRligleyZLpLBSe44A6Sq0uMG-8-NOcRCcT4Q,18080
+ipex_llm/transformers/models/internlm.py,sha256=JZFrI2HXsIAfM-6pA2RO0wcXopOliC1FggLMzNzaDZ4,17404
 ipex_llm/transformers/models/internvl.py,sha256=Vx0vENIEQLX2M6P398mw5TOhpks0U8xf8rtRQvy94go,8154
 ipex_llm/transformers/models/llama.py,sha256=NzpyQve_RC9ez1W-jWPLGZ80k_S1I5Rx5saAzCsDIoI,8558
 ipex_llm/transformers/models/minicpm.py,sha256=eaPNVNrep0_xGoELhZd886ff0ceoKqB6cusdAhd52eE,10145
@@ -169,10 +169,10 @@ ipex_llm/transformers/models/mistral.py,sha256=uVhkdXaq15v1P3QY0emVsA7SxUbAWChHE
 ipex_llm/transformers/models/mllama.py,sha256=ZyRq9DTKsvk1AlRbr-z6ngjS3Sr_7YuGZ6-Yr1MBBAM,10937
 ipex_llm/transformers/models/mpt.py,sha256=z02NwHogJZVh-Mk4sYoIzR90SFIKhoNN_-ifsD907TQ,9540
 ipex_llm/transformers/models/phi.py,sha256=E6qz4EEuHIVGvaPo-wtLC5lz3iyMqTbAE_cRlcjQRKI,6670
-ipex_llm/transformers/models/phi3.py,sha256=Fo6PlZ24Gdm7eeeZOTMm1Bfh3U6P4rvq7-_2FHvp0vE,15503
+ipex_llm/transformers/models/phi3.py,sha256=AaWB7TPQdrDYgpcVHglG0Q0480bxNOw1mFePddlBEFk,14849
 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=JLaY9ZT7A22oO0G8K-nvjvKQDaIrKA5o-jEHvk_y3eI,25604
+ipex_llm/transformers/models/qwen2.py,sha256=zK-FpUaxEhjD4gZa1ZvArodAilz29T_cpeAqfCGosc0,14317
 ipex_llm/transformers/models/qwen2_moe.py,sha256=a0gYo-ngf8SxaEnBdZUJDnPS6Mkn_poDd8xqhx50icI,19516
 ipex_llm/transformers/models/qwen2_vl.py,sha256=NrhxlaPj7W-HUBmKc3CSTwZy1lkoZ9qDaxM4GvE0kHs,13583
 ipex_llm/transformers/models/qwen_vl.py,sha256=j7Nzzz2Qvynu9yrCXmoEfERjw43hXof5TbXIs7Ms-oY,17105
@@ -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=fj-XtOnR6kggnFUQTMPCOOzolkPztN06WAv8QW-XRnI,7054
 ipex_llm/transformers/models/starcoder2.py,sha256=ONKvD7JCkRM0DI-R56x28QFBJ7CjD5hOZBQ_3WfOcNk,6626
-ipex_llm/transformers/models/utils.py,sha256=ihbWS5kQK2KHDVPkMhgjik3nM8B2fWf-E-z4BWNUstk,15568
+ipex_llm/transformers/models/utils.py,sha256=WYBc26vSiy_CzV07z-eT5ts90Kko2yUmS3DDZtfGcRk,15065
 ipex_llm/transformers/models/yuan.py,sha256=JYAn_ZaSGK0NBJLEIxCACfAq084a66GFJkdd5NbpmMA,7732
 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-2.2.0b20250109.data/scripts/ipex-llm-init,sha256=fLQsT2dRL6H5bThb4GuIWotAuqoLsIxFwA-0c2qmaO8,6672
-ipex_llm-2.2.0b20250109.data/scripts/llm-chat,sha256=TdUnUmNapzuoe1c8IzrdVOQwWEg8IqsMSBRlOD3daZM,2249
-ipex_llm-2.2.0b20250109.data/scripts/llm-cli,sha256=RXGPlLElHxcKzoUxljEMBIAXbzCDysXL-Nxw-xF-7LU,2457
-ipex_llm-2.2.0b20250109.dist-info/METADATA,sha256=gPslIWSw_X5E5ULhQa8rOHeRo_UeBDXCAyPjBSPB-nU,12705
-ipex_llm-2.2.0b20250109.dist-info/WHEEL,sha256=PPJcBMAZibF_2GFE9NmOJGqiaSMPiNFbJd6QaJjdA6Y,109
-ipex_llm-2.2.0b20250109.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
-ipex_llm-2.2.0b20250109.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
-ipex_llm-2.2.0b20250109.dist-info/RECORD,,
+ipex_llm-2.2.0b20250111.data/scripts/ipex-llm-init,sha256=fLQsT2dRL6H5bThb4GuIWotAuqoLsIxFwA-0c2qmaO8,6672
+ipex_llm-2.2.0b20250111.data/scripts/llm-chat,sha256=TdUnUmNapzuoe1c8IzrdVOQwWEg8IqsMSBRlOD3daZM,2249
+ipex_llm-2.2.0b20250111.data/scripts/llm-cli,sha256=RXGPlLElHxcKzoUxljEMBIAXbzCDysXL-Nxw-xF-7LU,2457
+ipex_llm-2.2.0b20250111.dist-info/METADATA,sha256=8HtPWBsOYbGoboTzVcl5ygPez_bfjvEzqxWanbYhx_o,12705
+ipex_llm-2.2.0b20250111.dist-info/WHEEL,sha256=PPJcBMAZibF_2GFE9NmOJGqiaSMPiNFbJd6QaJjdA6Y,109
+ipex_llm-2.2.0b20250111.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
+ipex_llm-2.2.0b20250111.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
+ipex_llm-2.2.0b20250111.dist-info/RECORD,,