ipex-llm 2.2.0b20250106__py3-none-manylinux2010_x86_64.whl → 2.2.0b20250106.post1__py3-none-manylinux2010_x86_64.whl

ipex_llm/transformers/models/utils.py

@@ -19,7 +19,7 @@ import torch
 import warnings
 from ipex_llm.utils.common import invalidInputError
 from ipex_llm.ggml.quantize import ggml_tensor_qtype
-from ipex_llm.transformers.utils import get_ipex_version, get_xpu_device_type
+from ipex_llm.transformers.utils import get_ipex_version, get_xpu_device_name
 from ipex_llm.transformers.low_bit_linear import SYM_INT4, SYM_INT8, FP8E5, IQ2_XXS, FP4, FP8E4,\
     FP6, ASYM_INT4
 
@@ -85,16 +85,14 @@ def use_quantize_kv_cache(linear: torch.nn.Module, x: torch.Tensor, kv_group: in
         return os.environ["IPEX_LLM_QUANTIZE_KV_CACHE"] == "1"
     elif os.environ.get("IPEX_LLM_LOW_MEM", None) is not None:
         return os.environ["IPEX_LLM_LOW_MEM"] == "1"
+    elif linear.qtype in [ggml_tensor_qtype["fp16"], ggml_tensor_qtype["bf16"]]:
+        return False
     else:
-        return x.device.type == 'xpu' and kv_cache_device_check(x, kv_group) \
-            and hasattr(linear, "qtype") and \
-            linear.qtype != ggml_tensor_qtype["fp16"] and linear.qtype != ggml_tensor_qtype["bf16"]
-
-
-def kv_cache_device_check(x: torch.Tensor, kv_group: int) -> bool:
-    return (get_xpu_device_type(x) in ["mtl", "lnl"] and kv_group <= 1) or \
-        ((get_xpu_device_type(x) == "arc" or get_xpu_device_type(x) == "flex") and
-            1 < x.size(0) and x.size(0) <= 8)
+        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
+        )
 
 
 def init_fp8_kv_cache(batch_size, num_heads, current_length, head_dim, device):
@@ -170,7 +168,7 @@ def should_use_fuse_rope(hidden_states, position_ids, training):
 
 def apply_rotary_pos_emb(q, k, cos, sin, position_ids, model_family):
     if model_family in ["llama", "baichuan", "internlm", "aquila", "gpt_neox", "mistral",
-                        "mixtral", "qwen2", "yuan", "stablelm", "qwen2_moe"]:
+                        "qwen2", "yuan", "stablelm", "qwen2_moe"]:
         # The first two dimensions of cos and sin are always 1, so we can `squeeze` them.
         cos = cos.squeeze(1).squeeze(0)  # [seq_len, dim]
         sin = sin.squeeze(1).squeeze(0)  # [seq_len, dim]
@@ -185,7 +183,7 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids, model_family):
         q_embed = (q * cos) + (rotate_half(q) * sin)
         k_embed = (k * cos) + (rotate_half(k) * sin)
         return q_embed, k_embed
-    elif model_family in ["gptj", "chatglm"]:
+    elif model_family in ["chatglm"]:
         q_embed = (q * cos) + (rotate_every_two(q) * sin)
         k_embed = (k * cos) + (rotate_every_two(k) * sin)
         return q_embed, k_embed
@@ -194,19 +192,6 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids, model_family):
                           f"{model_family} is not supported.")
 
 
-def apply_ipex_rotate_every_two(q, k, cos, sin):
-    # ipex's apply_rotary_embedding_two_qk can change the origin storage,
-    # so q/k will get the result directly.
-    from ipex_llm.transformers.utils import get_ipex_version
-    if get_ipex_version() >= "2.1.10+xpu":
-        torch.ops.torch_ipex.apply_rotary_embedding_two_qk(
-            q, k, sin, cos, q, k
-        )
-    else:
-        torch.ops.torch_ipex.apply_rotary_embedding(q, sin, cos, q)
-        torch.ops.torch_ipex.apply_rotary_embedding(k, sin, cos, k)
-
-
 def is_enough_kv_cache_room_4_36(past_key_value, idx, seq_len=1):
     # to determinate if is enough kv cache room in transformers==4.36
     # seq_len for current seq len
@@ -226,57 +211,6 @@ def is_enough_kv_cache_room_4_31(past_key_value, seq_len=1):
         (past_key_value[0].size(2) + seq_len) * past_key_value[0].size(3)
 
 
-def use_flash_attention(query, key, attention_mask=None):
-    # here we support query's shape is always [batch_size, head_num, q_len, head_dim],
-    # key's shape is always [batch_size, head_num, k_len, head_dim]
-    invalidInputError(query.dim() == 4,
-                      "Here query input of use_flash_attention should be [batch_size, "
-                      "head_num, q_len, head_dim]")
-    invalidInputError(key.dim() == 4,
-                      "Here key input of use_flash_attention should be [batch_size, "
-                      "head_num, k_len, head_dim]")
-    bsz, _, q_len, _ = query.size()
-    k_len = key.size()[2]
-    # check whether ipex flash attention can be used
-    if q_len != k_len:
-        # now only use flash attention for first token
-        # as it seems have no performance benifit for rest token now
-        return False
-    if query.device.type != "xpu":
-        # ipex flash attention only support for xpu
-        return False
-    ipex_version = get_ipex_version()
-    if ipex_version <= "2.0.110+xpu":
-        # ipex flash attention is supported from ipex 2.1
-        return False
-    if not torch.xpu.has_xetla():
-        # ipex flash attention is only supported for xetla
-        # may update this later
-        return False
-    elif get_xpu_device_type(query) != "pvc":
-        return False
-    if query.dtype not in [torch.float32, torch.float16]:
-        # only use flash attention for fp32/fp16 input
-        return False
-    if bsz > 1:
-        # as flash attention doesn't support attn_mask in ipex 2.1,
-        # so it will cause output error for padded batch input
-        if attention_mask is None:
-            return True
-        else:
-            # TODO: below logic may change for different model
-            # attention mask shape : [bsz, 1, q_len, k_len]
-            if attention_mask[0].squeeze()[0, 0].item() != 0:
-                # first batch contains padding
-                # otherwise we suppose it should be a upper triangular matrix
-                # at the same time, the diagonal is also 0
-                return False
-            elif not attention_mask.equal(attention_mask[0].repeat(bsz, 1, 1, 1)):
-                # check whether mask of every batch is the same
-                return False
-    return True
-
-
 def use_sdp(q_len, kv_len, head_dim, query_states):
     return (
         query_states.device.type == "xpu"
@@ -315,38 +249,16 @@ def mlp_fusion_check(x, qtype, training):
     if training or x.requires_grad:
         return False
     if qtype == FP6:
-        device = get_xpu_device_type(x)
-        if device in ["mtl", "lnl"]:
+        device = get_xpu_device_name(x.device)
+        if device in ["mtl", "lnl", "arl"]:
             return False
     return True
 
 
-def use_decoding_fast_path(proj,
-                           use_fuse_rope,
-                           enough_kv_room,
-                           bs,
-                           qtype_check=decoding_fast_path_qtype_check):
-    if proj is None:
-        return False
-    device = get_xpu_device_type(proj.weight)
-    if not qtype_check(proj):
-        return False
-    if not use_fuse_rope:
-        return False
-    if not enough_kv_room:
-        return False
-    if bs != 1:
-        return False
-
-    if device in ["uhd"]:
-        return False
-    return True
-
-
 def use_xmx(x: torch.Tensor, qtype: int):
-    device = get_xpu_device_type(x)
+    device = get_xpu_device_name(x.device)
     return (
-        device in ["arc", "flex", "pvc"]
+        device in ["arc", "pvc"]
         and qtype in [SYM_INT4, SYM_INT8, FP8E4, FP8E5]
         and (
             (device == "pvc" and 1 < x.size(0) <= 16)
@@ -370,7 +282,7 @@ def fp16_fusion_check(proj, x, training):
         return False
     if x.requires_grad:
         return False
-    device_type = get_xpu_device_type(x)
+    device_type = get_xpu_device_name(x.device)
     if device_type != "pvc":
         return False
     return True
@@ -439,7 +351,7 @@ def should_use_compresskv(x: torch.Tensor, prompt_len: int):
     else:
         if use_compress_kv is None:
             return (
-                get_xpu_device_type(x) in ["mtl", "lnl"]
+                get_xpu_device_name(x.device) in ["mtl", "lnl", "arl"]
                 and prompt_len >= 1800
                 and prompt_len <= 4500
             )

ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py

@@ -473,10 +473,6 @@ def convert_llm_for_deploy(model: torch.nn.Module,
                        "n_splits_linear": n_splits_linear,
                        "n_splits_down_proj": n_splits_down_proj,
                        "lm_head_low_bit": lm_head_low_bit}
-        model.config.update(update_dict)
-        model.config.save_pretrained(save_directory)
-        if model.can_generate():
-            model.generation_config.save_pretrained(save_directory)
 
         from .qwen import convert_qwen_layer, convert_fused_qwen_layer
         from .qwen import convert_lm_head_and_embedding
@@ -537,8 +533,6 @@ def convert_llm_for_deploy(model: torch.nn.Module,
                        "n_splits_linear": n_splits_linear,
                        "n_splits_down_proj": n_splits_down_proj,
                        "lm_head_low_bit": lm_head_low_bit}
-        model.config.update(update_dict)
-        model.config.save_pretrained(save_directory)
 
         from .llama import convert_llama_layer, convert_fused_llama_layer
         from .llama import convert_lm_head_and_embedding
@@ -577,8 +571,6 @@ def convert_llm_for_deploy(model: torch.nn.Module,
                        "n_splits_linear": n_splits_linear,
                        "n_splits_down_proj": n_splits_down_proj,
                        "lm_head_low_bit": lm_head_low_bit}
-        model.config.update(update_dict)
-        model.config.save_pretrained(save_directory)
 
         from .minicpm import convert_minicpm_layer, convert_fused_minicpm_layer
         from .minicpm import convert_lm_head_and_embedding
@@ -595,3 +587,8 @@ def convert_llm_for_deploy(model: torch.nn.Module,
                                       save_directory, weight_dir,
                                       convert_model=True,
                                       max_prompt_len=max_prompt_len)
+
+    model.config.update(update_dict)
+    model.config.save_pretrained(save_directory)
+    if model.can_generate():
+        model.generation_config.save_pretrained(save_directory)

ipex_llm/transformers/speculative.py

@@ -432,8 +432,7 @@ def _check_and_extend_kv_cache(past_key_values, max_step_draft, kv_alloc_block_l
     from ipex_llm.transformers.models.utils import is_enough_kv_cache_room_4_31, \
         extend_kv_cache
     enough_kv_room = True
-    if model_type not in ["chatglm", "qwen", "baichuan", "llama", "mistral",
-                          "gptj", "opt"]:
+    if model_type not in ["chatglm", "qwen", "baichuan", "llama", "mistral", "opt"]:
         return past_key_values, False
     cache_k = past_key_values[0][0]
     if model_type == "chatglm":
@@ -527,7 +526,7 @@ def _crop_past_key_values(self, past_key_values, new_cache_size, _enable_ipex=Fa
                         v[:-(new_cache_size), :, :, :])
                     for k, v in past_key_values
                 ]
-        elif self.config.model_type in ["baichuan", "gptj"]:
+        elif self.config.model_type in ["baichuan"]:
             past_key_values = [
                 (k[:, :, :-(new_cache_size), :],
                     v[:, :, :-(new_cache_size), :])
@@ -796,13 +795,6 @@ def _non_cpu_ipex_verify(self, verify_input_ids, past_key_values, cur_attention_
                                     device=verify_input_ids.device)
         position_ids = position_ids.unsqueeze(0).repeat(1, 1) + past_key_value_len
         forward_args["position_ids"] = position_ids
-    elif self.config.model_type == "gptj":
-        past_length = past_key_values[0][0].size(2)
-        input_len = verify_input_ids.shape[1]
-        position_ids = torch.arange(past_length, input_len + past_length,
-                                    dtype=torch.long, device=verify_input_ids.device)
-        position_ids = position_ids.unsqueeze(0).view(-1, input_len)
-        forward_args["position_ids"] = position_ids
 
     return self(**forward_args)
 
@@ -971,10 +963,6 @@ def speculative_generate(self,
                         past_key_value_len = past_key_values[0][0].shape[0]
                     position_ids = torch.Tensor([[past_key_value_len + step_draft]]).long()
                     forward_args["position_ids"] = position_ids
-                elif self.config.model_type == "gptj":
-                    past_length = draft_past_key_values[0][0].size(2)
-                    position_ids = torch.Tensor([[past_length]]).long().to(self.device)
-                    forward_args["position_ids"] = position_ids
 
                 if _enable_ipex:
                     if any(keyword in self.config.model_type

ipex_llm/transformers/utils.py

@@ -168,27 +168,14 @@ def get_ipex_version():
     return _ipex_version
 
 
-def get_xpu_device_type(x):
-    if x.device.type != "xpu":
-        return x.device.type
-    name = torch.xpu.get_device_name(x.device.index)
-    if name.startswith("Intel(R) Arc(TM) A"):
-        return "arc"
-    elif name.startswith("Intel(R) Graphics [0xe20b]"):
-        return "bmg"
-    elif name.startswith("Intel(R) Arc(TM)"):
-        if 'V' in name:
-            return "lnl"
-        else:
-            return "mtl"
-    elif name.startswith("Intel(R) Data Center GPU Flex"):
-        return "flex"
-    elif name.startswith("Intel(R) Data Center GPU Max"):
-        return "pvc"
-    elif name.startswith("Intel(R) UHD"):
-        return "uhd"
+def get_xpu_device_name(device: torch.device):
+    if device.type != "xpu":
+        return device.type
     else:
-        return "others"
+        # possiable device name:
+        # ["arc", "pvc", "mtl", "lnl", "bmg", "arl", "legacy", "unknown"]
+        import xe_linear
+        return xe_linear.get_xpu_device_name(device)
 
 
 def load_imatrix_data(imatrix_file):

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

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250106
+Version: 2.2.0b20250106.post1
 Summary: Large Language Model Develop Toolkit
 Home-page: https://github.com/intel-analytics/ipex-llm
 Author: BigDL Authors
@@ -27,15 +27,17 @@ Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine
 Requires-Dist: torch ==2.1.2+cpu ; (platform_system == "Linux") and extra == 'all'
 Requires-Dist: torch ==2.1.2 ; (platform_system == "Windows") and extra == 'all'
 Provides-Extra: cpp
-Requires-Dist: bigdl-core-cpp ==2.6.0b20250106 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250106.post1 ; extra == 'cpp'
 Requires-Dist: setuptools ; extra == 'cpp'
 Provides-Extra: cpp-arl
-Requires-Dist: bigdl-core-cpp ==2.6.0b20250106 ; extra == 'cpp-arl'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250106.post1 ; 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'
 Requires-Dist: dpcpp-cpp-rt ==2024.2.1 ; (platform_system == "Windows") and extra == 'cpp-arl'
 Requires-Dist: mkl-dpcpp ==2024.2.1 ; (platform_system == "Windows") and extra == 'cpp-arl'
 Requires-Dist: onednn-devel ==2024.2.1 ; (platform_system == "Windows") and extra == 'cpp'
+Requires-Dist: onednn ==2024.2.1 ; (platform_system == "Windows") and extra == 'cpp'
 Requires-Dist: dpcpp-cpp-rt ==2024.2.1 ; (platform_system == "Windows") and extra == 'cpp'
 Requires-Dist: mkl-dpcpp ==2024.2.1 ; (platform_system == "Windows") and extra == 'cpp'
 Provides-Extra: llama-index
@@ -65,7 +67,7 @@ Requires-Dist: transformers ==4.40.0 ; extra == 'npu'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'npu'
 Requires-Dist: torch ==2.1.2+cpu ; (platform_system == "Linux") and extra == 'npu'
 Requires-Dist: torch ==2.1.2 ; (platform_system == "Windows") and extra == 'npu'
-Requires-Dist: bigdl-core-npu ==2.6.0b20250106 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.6.0b20250106.post1 ; (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'
@@ -85,9 +87,9 @@ Requires-Dist: setuptools <70.0.0 ; extra == 'xpu'
 Requires-Dist: torch ==2.1.0a0 ; extra == 'xpu'
 Requires-Dist: torchvision ==0.16.0a0 ; extra == 'xpu'
 Requires-Dist: intel-extension-for-pytorch ==2.1.10+xpu ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250106 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250106 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250106 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250106.post1 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250106.post1 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250106.post1 ; extra == 'xpu'
 Provides-Extra: xpu-2-1
 Requires-Dist: py-cpuinfo ; extra == 'xpu-2-1'
 Requires-Dist: protobuf ; extra == 'xpu-2-1'
@@ -102,12 +104,28 @@ Requires-Dist: setuptools <70.0.0 ; extra == 'xpu-2-1'
 Requires-Dist: torch ==2.1.0a0 ; extra == 'xpu-2-1'
 Requires-Dist: torchvision ==0.16.0a0 ; extra == 'xpu-2-1'
 Requires-Dist: intel-extension-for-pytorch ==2.1.10+xpu ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250106 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250106 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250106 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250106.post1 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250106.post1 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250106.post1 ; extra == 'xpu-2-1'
 Requires-Dist: 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'
+Provides-Extra: xpu-2-6
+Requires-Dist: py-cpuinfo ; extra == 'xpu-2-6'
+Requires-Dist: protobuf ; extra == 'xpu-2-6'
+Requires-Dist: mpmath ==1.3.0 ; extra == 'xpu-2-6'
+Requires-Dist: numpy ==1.26.4 ; extra == 'xpu-2-6'
+Requires-Dist: transformers ==4.37.0 ; extra == 'xpu-2-6'
+Requires-Dist: sentencepiece ; extra == 'xpu-2-6'
+Requires-Dist: tokenizers ==0.15.2 ; extra == 'xpu-2-6'
+Requires-Dist: accelerate ==0.23.0 ; extra == 'xpu-2-6'
+Requires-Dist: tabulate ; extra == 'xpu-2-6'
+Requires-Dist: setuptools ; extra == 'xpu-2-6'
+Requires-Dist: torch ==2.6.0+xpu ; extra == 'xpu-2-6'
+Requires-Dist: torchvision ==0.21.0+xpu ; extra == 'xpu-2-6'
+Requires-Dist: torchaudio ==2.6.0+xpu ; extra == 'xpu-2-6'
+Requires-Dist: bigdl-core-xe-all ==2.6.0b20250106.post1 ; extra == 'xpu-2-6'
+Requires-Dist: 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'
 Requires-Dist: protobuf ; extra == 'xpu-arc'
@@ -119,9 +137,9 @@ Requires-Dist: tokenizers ==0.15.2 ; extra == 'xpu-arc'
 Requires-Dist: accelerate ==0.23.0 ; extra == 'xpu-arc'
 Requires-Dist: tabulate ; extra == 'xpu-arc'
 Requires-Dist: setuptools ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250106 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250106 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250106 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arc'
 Requires-Dist: 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'
@@ -130,6 +148,7 @@ Requires-Dist: torch ==2.3.1.post0+cxx11.abi ; (platform_system == "Windows") an
 Requires-Dist: torchvision ==0.18.1.post0+cxx11.abi ; (platform_system == "Windows") and extra == 'xpu-arc'
 Requires-Dist: intel-extension-for-pytorch ==2.3.110.post0+xpu ; (platform_system == "Windows") and extra == 'xpu-arc'
 Requires-Dist: onednn-devel ==2024.1.1 ; (platform_system == "Windows") and extra == 'xpu-arc'
+Requires-Dist: onednn ==2024.1.1 ; (platform_system == "Windows") and extra == 'xpu-arc'
 Provides-Extra: xpu-arl
 Requires-Dist: py-cpuinfo ; extra == 'xpu-arl'
 Requires-Dist: protobuf ; extra == 'xpu-arl'
@@ -141,9 +160,9 @@ Requires-Dist: tokenizers ==0.15.2 ; extra == 'xpu-arl'
 Requires-Dist: accelerate ==0.23.0 ; extra == 'xpu-arl'
 Requires-Dist: tabulate ; extra == 'xpu-arl'
 Requires-Dist: setuptools ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250106 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250106 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250106 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250106.post1 ; extra == 'xpu-arl'
 Requires-Dist: 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'
@@ -152,6 +171,7 @@ Requires-Dist: torch ==2.3.1.post0+cxx11.abi ; (platform_system == "Windows") an
 Requires-Dist: torchvision ==0.18.1.post0+cxx11.abi ; (platform_system == "Windows") and extra == 'xpu-arl'
 Requires-Dist: intel-extension-for-pytorch ==2.3.110.post0+xpu ; (platform_system == "Windows") and extra == 'xpu-arl'
 Requires-Dist: onednn-devel ==2024.1.1 ; (platform_system == "Windows") and extra == 'xpu-arl'
+Requires-Dist: onednn ==2024.1.1 ; (platform_system == "Windows") and extra == 'xpu-arl'
 Provides-Extra: xpu-lnl
 Requires-Dist: py-cpuinfo ; extra == 'xpu-lnl'
 Requires-Dist: protobuf ; extra == 'xpu-lnl'
@@ -163,9 +183,9 @@ Requires-Dist: tokenizers ==0.15.2 ; extra == 'xpu-lnl'
 Requires-Dist: accelerate ==0.23.0 ; extra == 'xpu-lnl'
 Requires-Dist: tabulate ; extra == 'xpu-lnl'
 Requires-Dist: setuptools ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250106 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250106 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250106 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250106.post1 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250106.post1 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250106.post1 ; extra == 'xpu-lnl'
 Requires-Dist: 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'
@@ -174,6 +194,7 @@ Requires-Dist: torch ==2.3.1.post0+cxx11.abi ; (platform_system == "Windows") an
 Requires-Dist: torchvision ==0.18.1.post0+cxx11.abi ; (platform_system == "Windows") and extra == 'xpu-lnl'
 Requires-Dist: intel-extension-for-pytorch ==2.3.110.post0+xpu ; (platform_system == "Windows") and extra == 'xpu-lnl'
 Requires-Dist: onednn-devel ==2024.1.1 ; (platform_system == "Windows") and extra == 'xpu-lnl'
+Requires-Dist: onednn ==2024.1.1 ; (platform_system == "Windows") and extra == 'xpu-lnl'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'xpu'
 Requires-Dist: dpcpp-cpp-rt ==2024.0.2 ; (platform_system == "Windows") and extra == 'xpu'
 Requires-Dist: mkl-dpcpp ==2024.0.0 ; (platform_system == "Windows") and extra == 'xpu'

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

@@ -94,25 +94,25 @@ 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=2kcaxal7x1ltv_d-qE1r0FXlYNEwlQ2LwAMXLuz-X9Y,106252
+ipex_llm/transformers/convert.py,sha256=TxWdTTOSvh-j5jqokQJVWykta4U4LHupE1QJ-9udzwc,98733
 ipex_llm/transformers/convert_ipex.py,sha256=iKXo0n8fVFTOA2fNYYrByMFK0dovL-kLd2sVDk88AlQ,14334
 ipex_llm/transformers/embedding.py,sha256=bdgk59DvD4ZZyxRzewXOR7g56nThgO6uhIwk8QL7f-s,9299
 ipex_llm/transformers/kv.py,sha256=k4TU18LlA-Sbq9WNNQnfuzu3RSFBwFhmaV3BcGN5bAo,19191
 ipex_llm/transformers/lisa.py,sha256=F5WxbtXQ7RdKulj83h_2DnEIgKiKGZf7zvOmg6QBl2s,3289
-ipex_llm/transformers/loader.py,sha256=cOgX93xOC-4dt01GTJ5wyd7PjZ8S43r4mctkR2YxVuw,6893
-ipex_llm/transformers/lookup.py,sha256=c4ETIha6ZLbWvhcclSKRDdi5Ipuet4mfUnOkBa0E8kk,19607
-ipex_llm/transformers/low_bit_linear.py,sha256=dyyYyCqw0GK8hzaUGanrg-uIhU1HTLEEbvbxXMlm-80,41668
-ipex_llm/transformers/model.py,sha256=KcRjkauGg48BYrUBoUZaVMpg7Piuz5JrfIpVZd3EIjs,41105
+ipex_llm/transformers/loader.py,sha256=AwjV5RpI2t2bedlv7ZhLm8cfd-QJZm5hny-XyjIvdnk,6876
+ipex_llm/transformers/lookup.py,sha256=b6OlZ9OV10R9qeWw8mVryVpDxszkjwLkldvi7GPMJY8,19614
+ipex_llm/transformers/low_bit_linear.py,sha256=lPIvDuRoS0zusiJ6vw_fOTJgK5ylh4CuD3U-qs8ih4Y,40869
+ipex_llm/transformers/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/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
 ipex_llm/transformers/relora.py,sha256=-dYzUV0P-IhO2jFdnzN9-v_sFzJpRj3ZwN9eCJzOoCw,16567
-ipex_llm/transformers/speculative.py,sha256=Zf1nQb5GXpJQrUHBTL-H4RUBfdv3lGhfehzudHimhYk,64109
+ipex_llm/transformers/speculative.py,sha256=0XNLgc9dGswJHVPrXo4iM7pPxkWwfFfJMECcivJSnIc,63368
 ipex_llm/transformers/streamer.py,sha256=RrVlLblzCOtABRUpaMXAyaMnCGgLUtAi_YesLumRbww,4842
 ipex_llm/transformers/training_patch.py,sha256=oxMkUtqyvqJiprw6dE3skkYfD1HOmUlH9N0hBkbn0G0,10799
-ipex_llm/transformers/utils.py,sha256=IRzmoRwgCwnOaiDroSyqc2pvlj-ipNvcW8RubvVT_rY,17374
+ipex_llm/transformers/utils.py,sha256=7syzq4jnEo-mWmS9E2VZ2GcFrjojWI8E7Hcx0tloifg,16996
 ipex_llm/transformers/xpu_customize_fwd.py,sha256=wFpIhs5F6tkNs8gBOrLxWdhLzO3EDHovVkERPIAoAvg,7611
 ipex_llm/transformers/xpu_ops.py,sha256=H46-69pMRQhekbAEoDfNacCInLWycMHDqrgMGLvFYfI,4362
 ipex_llm/transformers/awq/__init__.py,sha256=Du5gu3-eeAkeDO_dEMBTzrDBA66DSN3uL3-rn8WGXQw,875
@@ -148,28 +148,24 @@ ipex_llm/transformers/models/baichuan.py,sha256=oJCAEENSG8oQhJ-QPN2SiapARjAGdOM6
 ipex_llm/transformers/models/bert.py,sha256=bJNic2pt1kph0kBwdK5MRGyWupFfx2Ts0V3D1L-5kWo,6085
 ipex_llm/transformers/models/bloom.py,sha256=PxfzyYT-nFn3K5rZhTQjmcEjUUzAhUFzxIN4kzRlCuc,8103
 ipex_llm/transformers/models/chatglm.py,sha256=UHai1t2AUtGmF765_eHF8LUMVQzp_oCBx8TJB21WrHk,12597
-ipex_llm/transformers/models/chatglm2.py,sha256=kfJThuKYb3unAB1XCzfop1iDW1gOkyFOjSr-lEjUdS0,24781
+ipex_llm/transformers/models/chatglm2.py,sha256=SGCABJdYQLW0zDarEoWrEQLuWlbq9iQhYU8ZeR1-ptQ,15957
 ipex_llm/transformers/models/chatglm4.py,sha256=AAhAFFDDas5DBQPfh2Mwl7a2v7taKf6xphoeeNNFaBI,16593
 ipex_llm/transformers/models/chatglm4v.py,sha256=YRfuf9g1E0MQ_7wbHAOMvadFnO-j3LqI_k1SaRkDs0M,14055
-ipex_llm/transformers/models/cohere.py,sha256=RCUBfhI_eNuZ1NyQC6vm28eYqPmgZn6pHSvQMwFyT2A,25884
 ipex_llm/transformers/models/common.py,sha256=4obQMGF02FCiXrHnFle9Fsx7C33b1FDt37qJJ4YgxRc,11578
 ipex_llm/transformers/models/decilm.py,sha256=P-PBuDPf07GvKggLwJx_wPwIn6esN3rX8ai2JxRuZmE,5246
-ipex_llm/transformers/models/falcon.py,sha256=f5BzMbv4E-R5Pete8zBscbgiueXGIaWGs-5RbcMlUo4,33549
 ipex_llm/transformers/models/gemma.py,sha256=_E3Yw8Y45xyNVeLqyVKcpr8kjuICtETeL82cJ-bWJuU,9424
 ipex_llm/transformers/models/gemma2.py,sha256=2WZuv-FLzJyTJFaYxOuzJt47QE64M0lHnzAiO5T6ozI,8049
 ipex_llm/transformers/models/glm.py,sha256=gHYgfn20jPRL-ElXy-rUqMh6_LQcc5x7DEXSZuRA4E0,7094
 ipex_llm/transformers/models/gpt2.py,sha256=YSaNgK1uLCFDuIFqnKO0Mi-AsOZsYav-7pNf_NpKGdM,3445
 ipex_llm/transformers/models/gptbigcode.py,sha256=cP1_qGWoa43R2WacAMblShjku4QupcCZiLaPPAoOUs4,9101
-ipex_llm/transformers/models/gptj.py,sha256=TTIx461X2nOcIkrAcZhEf7d7mjJ3yvEC9KLVc1-hrpc,17973
 ipex_llm/transformers/models/gptneox.py,sha256=loRh1x_5S6BCeOr_s5xr-N_1SQHL3Y5IiUBAEyoMUqQ,6172
 ipex_llm/transformers/models/internlm.py,sha256=ZbIUMDwNRcrCeduXfbA_uq1AUEWawEt6CJRvQl3LkAg,17832
 ipex_llm/transformers/models/internvl.py,sha256=Vx0vENIEQLX2M6P398mw5TOhpks0U8xf8rtRQvy94go,8154
 ipex_llm/transformers/models/llama.py,sha256=ozwtdQ0MbanJEtW4LBFGxqs_QAq82EonhL2dL6tGyw0,8567
 ipex_llm/transformers/models/minicpm.py,sha256=ib2rJTN7Tf7znBCtVrtXsF-_Uuk2aA7KVg02xzatLiI,10103
 ipex_llm/transformers/models/minicpm3.py,sha256=FhNS6mi2rg7dSdF_QQGrao3g9EC6XLn1MTKd-kd0wF0,9191
-ipex_llm/transformers/models/minicpmv.py,sha256=igfugwyP3Nu3DcLhYXpe9W9Sk3q2YSrzjj7ukAwuJkg,9829
+ipex_llm/transformers/models/minicpmv.py,sha256=ZV4s48WNIyRoEkvENnlmopnx3ojZANBer0LI6bRtxrY,9826
 ipex_llm/transformers/models/mistral.py,sha256=rE1GWQxXvF6aG-buPHDR13zeynDZEDIubPF4PiVhZbM,7451
-ipex_llm/transformers/models/mixtral.py,sha256=zShaxabIoQaL3cV0Rptf7VCN4QuJsV4KBhanff1TASY,26601
 ipex_llm/transformers/models/mllama.py,sha256=ogpLmmN_OwcFUyjYB-oDC-l3uw8urFvUEc5edkjWHAk,10939
 ipex_llm/transformers/models/mpt.py,sha256=z02NwHogJZVh-Mk4sYoIzR90SFIKhoNN_-ifsD907TQ,9540
 ipex_llm/transformers/models/phi.py,sha256=E6qz4EEuHIVGvaPo-wtLC5lz3iyMqTbAE_cRlcjQRKI,6670
@@ -182,10 +178,10 @@ ipex_llm/transformers/models/qwen2_vl.py,sha256=jIm4yZSd751BkRqgj3wR1QBkDIh-TMCL
 ipex_llm/transformers/models/qwen_vl.py,sha256=j7Nzzz2Qvynu9yrCXmoEfERjw43hXof5TbXIs7Ms-oY,17105
 ipex_llm/transformers/models/rwkv4.py,sha256=H4KMtxN0JA2ZTXnonHpsUUJ5xULemo-D1Jzl0ri_UY8,6123
 ipex_llm/transformers/models/rwkv5.py,sha256=OkRNj1pCAZg1z2Fw-I0DEnxLEdZyPeRSQ6msrkxLOCs,10710
-ipex_llm/transformers/models/sd.py,sha256=7qkti_5jf3KmB-W4hDVeYGJvipuwGrxPox_4scGmnIc,5991
+ipex_llm/transformers/models/sd.py,sha256=VvHV5u-0k2MgHu3NL9113hPj7DgfxqctuKzEEeNfRDU,5981
 ipex_llm/transformers/models/stablelm.py,sha256=RGQCYuQhYqtZ1j3RZkYi0_QvCRnUgUIPYxfBcLnElzg,6885
 ipex_llm/transformers/models/starcoder2.py,sha256=4P3mhRYf2Kreb1ESjrQGfy1puLMmZXgV35zf-Tksvao,6462
-ipex_llm/transformers/models/utils.py,sha256=hpTT9X5zCi2CVlFRR-nApn_1Bsh1S8SHlFJwc-tKxf0,18985
+ipex_llm/transformers/models/utils.py,sha256=85rGIzGZvWe3JjYpWcuc1nfzI_tn_zFcdZpWivxJkl0,15457
 ipex_llm/transformers/models/yuan.py,sha256=1jRPebwAK2ENbyYokOmb4LSVo-szucWiygz9zTv-scs,7656
 ipex_llm/transformers/npu_models/__init__.py,sha256=ulEUGLjaP48LCrVeury3UxLjXxKzRi0UpSG4bYu-7f8,585
 ipex_llm/transformers/npu_models/baichuan.py,sha256=fJtd7fBrttySghRUgfZTAdxLjsSNC-XL08HISsXigLE,4685
@@ -218,7 +214,7 @@ ipex_llm/transformers/npu_models/xlm_mp.py,sha256=sj8OVun8xJprM7ZJp0XzWa55rqlSIz
 ipex_llm/transformers/npu_pipeline_model/__init__.py,sha256=b2IXvVqQ5cItki021h8s3ymW12RPu8QNPprq4Mn3bDM,586
 ipex_llm/transformers/npu_pipeline_model/baichuan.py,sha256=ICxRzFQ4OIANDkkVi2_4xOeQXmfFXYMx3H52KuE1xR4,6208
 ipex_llm/transformers/npu_pipeline_model/common.py,sha256=QxJoJESpv0BpwO_FBeAT2wKA56wNFfen8iI37PrMKuA,7838
-ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py,sha256=953Gua2tFKLICpbmCBm-Lxnur85Ce7jNMeRTNT4DKZE,28715
+ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py,sha256=wrRgmNT13RVtQRp5gFRBxNEPJHxFMLeGqb8a58YodPQ,28512
 ipex_llm/transformers/npu_pipeline_model/llama.py,sha256=MnvHRytLt3oy5jIPUBe8AeEJ6PtPWLbhQ5a9WqjZ1TQ,19905
 ipex_llm/transformers/npu_pipeline_model/minicpm.py,sha256=MDMesYlVbECKdK0xxkt1LwHgpkJOO7ZwBExYAwMGQa0,20637
 ipex_llm/transformers/npu_pipeline_model/pipeline_cpp.py,sha256=JNmodAMg_NQvDILug3E_fGXEh6cd3wsj4bvAzcd-vaU,2749
@@ -254,11 +250,11 @@ ipex_llm/vllm/xpu/engine/__init__.py,sha256=pY_CpyuZd72fr6s32ejeKHKFW0K4vUU2rzZj
 ipex_llm/vllm/xpu/engine/engine.py,sha256=k4-D27WS_Gk3mA--w3HWAjPjb4Aiu043MVPi0ZoAUBc,5984
 ipex_llm/vllm/xpu/entrypoints/openai/api_server.py,sha256=GshTZFB8e4PWvqckfbmTOU6b0oLkNn7A-vzLuG9--j8,21544
 ipex_llm/vllm/xpu/entrypoints/openai/cli_args.py,sha256=2rENA2ucynMaIjiZBEh2ez1o5vR32GaP514t39CD7KM,8676
-ipex_llm-2.2.0b20250106.data/scripts/ipex-llm-init,sha256=fLQsT2dRL6H5bThb4GuIWotAuqoLsIxFwA-0c2qmaO8,6672
-ipex_llm-2.2.0b20250106.data/scripts/llm-chat,sha256=TdUnUmNapzuoe1c8IzrdVOQwWEg8IqsMSBRlOD3daZM,2249
-ipex_llm-2.2.0b20250106.data/scripts/llm-cli,sha256=RXGPlLElHxcKzoUxljEMBIAXbzCDysXL-Nxw-xF-7LU,2457
-ipex_llm-2.2.0b20250106.dist-info/METADATA,sha256=RVDr0pwoPE6J0yPUZ9k7t6_jQn01wTwAXkU5ViqE-c8,11374
-ipex_llm-2.2.0b20250106.dist-info/WHEEL,sha256=PPJcBMAZibF_2GFE9NmOJGqiaSMPiNFbJd6QaJjdA6Y,109
-ipex_llm-2.2.0b20250106.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
-ipex_llm-2.2.0b20250106.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
-ipex_llm-2.2.0b20250106.dist-info/RECORD,,
+ipex_llm-2.2.0b20250106.post1.data/scripts/ipex-llm-init,sha256=fLQsT2dRL6H5bThb4GuIWotAuqoLsIxFwA-0c2qmaO8,6672
+ipex_llm-2.2.0b20250106.post1.data/scripts/llm-chat,sha256=TdUnUmNapzuoe1c8IzrdVOQwWEg8IqsMSBRlOD3daZM,2249
+ipex_llm-2.2.0b20250106.post1.data/scripts/llm-cli,sha256=RXGPlLElHxcKzoUxljEMBIAXbzCDysXL-Nxw-xF-7LU,2457
+ipex_llm-2.2.0b20250106.post1.dist-info/METADATA,sha256=I0vPU5mDtPZR3wpY87fYHdn6r14U0T50NWGP7EsF5s8,12825
+ipex_llm-2.2.0b20250106.post1.dist-info/WHEEL,sha256=PPJcBMAZibF_2GFE9NmOJGqiaSMPiNFbJd6QaJjdA6Y,109
+ipex_llm-2.2.0b20250106.post1.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
+ipex_llm-2.2.0b20250106.post1.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
+ipex_llm-2.2.0b20250106.post1.dist-info/RECORD,,