ipex-llm 2.2.0b20250101__py3-none-win_amd64.whl → 2.2.0b20250102__py3-none-win_amd64.whl

ipex_llm/optimize.py

@@ -254,7 +254,9 @@ def optimize_model(model, low_bit='sym_int4', optimize_llm=True, modules_to_not_
                                  torch_dtype=torch_dtype,
                                  optimize_model=optimize_llm,
                                  modules_to_not_convert=modules_to_not_convert,
-                                 cpu_embedding=cpu_embedding)
+                                 cpu_embedding=cpu_embedding,
+                                 disable_optimize_pre=kwargs.pop("disable_optimize_pre",
+                                                                 False))
     # add save_low_bit to pretrained model dynamically
     import types
     model._bigdl_config = dict()

ipex_llm/transformers/convert.py

@@ -1081,7 +1081,8 @@ def ggml_convert_low_bit(model, qtype, optimize_model=True,
                          torch_dtype="auto",
                          imatrix_data=None,
                          embedding_qtype=None,
-                         mixed_precision=False):
+                         mixed_precision=False,
+                         disable_optimize_pre=False):
     if qtype in ggml_tensor_qtype.values():
         index = list(ggml_tensor_qtype.values()).index(qtype)
         logger.info(f"Converting the current model to "
@@ -1104,7 +1105,7 @@ def ggml_convert_low_bit(model, qtype, optimize_model=True,
         model = _optimize_ipex(model, qtype)
         return model
 
-    if optimize_model:
+    if optimize_model and not disable_optimize_pre:
         model = _optimize_pre(model, qtype)
 
     act_order = False

ipex_llm/transformers/low_bit_linear.py

@@ -764,6 +764,7 @@ class FP16Linear(nn.Linear):
         # weigh_type = 3 means weight has been transposed by esimd method
         self.weight_type = 1
         self.optimize_lm_head = optimize_lm_head
+        self.disable_fp16_opt = False
 
     def forward(self, x: torch.Tensor):
         # only work for GPU
@@ -779,8 +780,11 @@ class FP16Linear(nn.Linear):
             self.weight.data = self.weight.data.to(x.dtype)
 
         if not self.use_esimd_kernel(x):
-            if get_ipex_version() < "2.1.10+xpu" \
-                    or get_xpu_device_type(x) not in ["arc", "flex", "pvc"]:
+            if (
+                get_ipex_version() < "2.1.10+xpu"
+                or get_xpu_device_type(x) not in ["arc", "flex", "pvc"]
+                or self.disable_fp16_opt
+            ):
                 if self.weight_type == 2:
                     self.weight = torch.nn.Parameter(self.weight.transpose(0, 1).contiguous(),
                                                      requires_grad=False)
@@ -845,6 +849,8 @@ class FP16Linear(nn.Linear):
 
     def use_esimd_kernel(self, x):
         gpu_type = get_xpu_device_type(x)
+        if self.disable_fp16_opt:
+            return False
         # esimd kernel can only be used for Arc and Flex
         if gpu_type not in ["arc", "flex"]:
             return False

ipex_llm/transformers/model.py

@@ -445,6 +445,7 @@ class _BaseAutoModelClass:
         mixed_precision = kwargs.pop("mixed_precision", False)
         if embedding_qtype is not None:
             embedding_qtype = ggml_tensor_qtype[embedding_qtype]
+        disable_optimize_pre = kwargs.pop("disable_optimize_pre", False)
         _args = copy.deepcopy(args)
         _kwargs = copy.deepcopy(kwargs)
         awq_config = None
@@ -513,7 +514,8 @@ class _BaseAutoModelClass:
                                      torch_dtype=kwargs.get("torch_dtype", 'auto'),
                                      imatrix_data=imatrix_data,
                                      embedding_qtype=embedding_qtype,
-                                     mixed_precision=mixed_precision)
+                                     mixed_precision=mixed_precision,
+                                     disable_optimize_pre=disable_optimize_pre)
 
         if disk_embedding:
             from ipex_llm.transformers.embedding import DiskEmbedding

ipex_llm/transformers/models/baichuan.py

@@ -29,7 +29,7 @@ from ipex_llm.transformers.models.utils import use_quantize_kv_cache, restore_fp
     should_use_compresskv
 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_flash_attention, use_sdp
+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 is_enough_kv_cache_room_4_36
@@ -301,16 +301,10 @@ def baichuan_attention_forward_7b(
 
     # IPEX-LLM OPT: sdp
     attn_weights = None
-    if use_flash_attention(query_states, key_states, attention_mask):
-        attn_output = F.scaled_dot_product_attention(query_states.to(dtype=torch.float16),
-                                                     key_states.to(dtype=torch.float16),
-                                                     value_states.to(dtype=torch.float16),
-                                                     is_causal=True).to(hidden_states.dtype)
-    else:
-        attn_output = scaled_dot_product_attention(
-            query_states, key_states, value_states,
-            attention_mask, q_len == kv_seq_len
-        )
+    attn_output = scaled_dot_product_attention(
+        query_states, key_states, value_states,
+        attention_mask, q_len == kv_seq_len
+    )
 
     attn_output = attn_output.transpose(1, 2).contiguous()
     attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)

ipex_llm/transformers/models/chatglm.py

@@ -23,7 +23,7 @@ import torch.utils.checkpoint
 import torch.nn.functional as F
 from typing import Optional, Tuple
 from ipex_llm.transformers.models.utils import init_kv_cache, extend_kv_cache, append_kv_cache
-from ipex_llm.transformers.models.utils import use_flash_attention, use_sdp
+from ipex_llm.transformers.models.utils import use_sdp
 
 
 def rotate_half(x):
@@ -41,7 +41,7 @@ def apply_rotary_pos_emb_index(q, k, cos, sin, position_id):
 
 
 def glm_sdpa(query, key, value, attention_mask=None, is_causal=False):
-    if use_flash_attention(query, key, attention_mask) or query.device.type == 'cpu':
+    if query.device.type == 'cpu':
         context_layer = F.scaled_dot_product_attention(query.to(key.dtype),
                                                        key,
                                                        value,

ipex_llm/transformers/models/qwen.py

@@ -33,7 +33,6 @@ from ipex_llm.transformers.models.utils import update_past_key_value, should_use
 from ipex_llm.transformers.models.utils import use_quantize_kv_cache
 from ipex_llm.transformers.models.utils import rotate_half, SILU
 from ipex_llm.transformers.models.utils import mlp_fusion_check
-from ipex_llm.transformers.models.utils import use_flash_attention
 from ipex_llm.utils.common import invalidInputError
 from transformers.modeling_outputs import BaseModelOutputWithPast
 
@@ -116,33 +115,28 @@ def qwen_attention_forward(
     past_key_value = (key_states.transpose(1, 2),
                       value_states.transpose(1, 2)) if use_cache else None
 
-    # IPEX-LLM OPT: sdp
+    # IPEX-LLM OPT: sdpa
     attn_weights = None
-    if use_flash_attention(query_states, key_states, attention_mask):
-        attn_output = F.scaled_dot_product_attention(query_states.to(dtype=torch.float16),
-                                                     key_states.to(dtype=torch.float16),
-                                                     value_states.to(dtype=torch.float16),
-                                                     is_causal=True).to(hidden_states.dtype)
+
+    if q_len > 1 and q_len != kv_seq_len:
+        causal_mask = torch.tril(
+            torch.ones((kv_seq_len, kv_seq_len), dtype=torch.bool, device=query_states.device)
+        ).view(1, 1, kv_seq_len, kv_seq_len)
+        causal_mask = causal_mask[
+            :, :, kv_seq_len - q_len:kv_seq_len, :kv_seq_len
+        ]
+        attention_mask = torch.zeros(causal_mask.shape, dtype=query_states.dtype,
+                                     device=query_states.device)
+        attention_mask.masked_fill_(causal_mask.logical_not(),
+                                    torch.finfo(attention_mask.dtype).min)
+        attention_mask = attention_mask.expand([bsz, -1, -1, -1])
     else:
-        if q_len > 1 and q_len != kv_seq_len:
-            causal_mask = torch.tril(
-                torch.ones((kv_seq_len, kv_seq_len), dtype=torch.bool, device=query_states.device)
-            ).view(1, 1, kv_seq_len, kv_seq_len)
-            causal_mask = causal_mask[
-                :, :, kv_seq_len - q_len:kv_seq_len, :kv_seq_len
-            ]
-            attention_mask = torch.zeros(causal_mask.shape, dtype=query_states.dtype,
-                                         device=query_states.device)
-            attention_mask.masked_fill_(causal_mask.logical_not(),
-                                        torch.finfo(attention_mask.dtype).min)
-            attention_mask = attention_mask.expand([bsz, -1, -1, -1])
-        else:
-            attention_mask = None
+        attention_mask = None
 
-        attn_output = scaled_dot_product_attention(
-            query_states, key_states, value_states,
-            attention_mask, q_len == kv_seq_len
-        )
+    attn_output = scaled_dot_product_attention(
+        query_states, key_states, value_states,
+        attention_mask, q_len == kv_seq_len
+    )
 
     attn_output = attn_output.transpose(1, 2).contiguous()
     attn_output = attn_output.view(bsz, q_len, self.hidden_size)
@@ -219,31 +213,25 @@ def qwen_attention_forward_registered(
     past_key_value = (key_states.transpose(1, 2),
                       value_states.transpose(1, 2)) if use_cache else None
 
-    # IPEX-LLM OPT: sdp
+    # IPEX-LLM OPT: sdpa
     attn_weights = None
 
-    if use_flash_attention(query_states, key_states, attention_mask):
-        attn_output = F.scaled_dot_product_attention(query_states.to(dtype=torch.float16),
-                                                     key_states.to(dtype=torch.float16),
-                                                     value_states.to(dtype=torch.float16),
-                                                     is_causal=True).to(hidden_states.dtype)
+    if q_len > 1 and q_len != kv_seq_len:
+        causal_mask = registered_causal_mask[
+            :, :, kv_seq_len - q_len:kv_seq_len, :kv_seq_len
+        ]
+        attention_mask = torch.zeros(causal_mask.shape, dtype=query_states.dtype,
+                                     device=query_states.device)
+        attention_mask.masked_fill_(causal_mask.logical_not(),
+                                    torch.finfo(attention_mask.dtype).min)
+        attention_mask = attention_mask.expand([bsz, -1, -1, -1])
     else:
-        if q_len > 1 and q_len != kv_seq_len:
-            causal_mask = registered_causal_mask[
-                :, :, kv_seq_len - q_len:kv_seq_len, :kv_seq_len
-            ]
-            attention_mask = torch.zeros(causal_mask.shape, dtype=query_states.dtype,
-                                         device=query_states.device)
-            attention_mask.masked_fill_(causal_mask.logical_not(),
-                                        torch.finfo(attention_mask.dtype).min)
-            attention_mask = attention_mask.expand([bsz, -1, -1, -1])
-        else:
-            attention_mask = None
+        attention_mask = None
 
-        attn_output = scaled_dot_product_attention(
-            query_states, key_states, value_states,
-            attention_mask, q_len == kv_seq_len
-        )
+    attn_output = scaled_dot_product_attention(
+        query_states, key_states, value_states,
+        attention_mask, q_len == kv_seq_len
+    )
 
     attn_output = attn_output.transpose(1, 2).contiguous()
     attn_output = attn_output.view(bsz, q_len, self.hidden_size)

ipex_llm/transformers/models/qwen2.py

@@ -38,12 +38,10 @@
 #
 
 import os
-import math
 from typing import Optional, Tuple, Union, List
 
 import torch
 from torch.nn import CrossEntropyLoss
-from torch.nn.functional import scaled_dot_product_attention as sdpa
 
 from ipex_llm.transformers.models.common import merge_qkv_base
 from ipex_llm.transformers.models.common import scaled_dot_product_attention
@@ -51,13 +49,12 @@ from ipex_llm.transformers.models.utils import SILU, mlp_fusion_check
 from ipex_llm.transformers.models.utils import should_use_fuse_rope
 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.models.utils import use_flash_attention
 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 apply_rotary_pos_emb, repeat_kv
+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
@@ -580,21 +577,10 @@ def qwen2_attention_forward(
                                                              self.layer_idx, None)
 
     attn_weights = None
-    if use_flash_attention(query_states, key_states, attention_mask):
-        if attention_mask is not None:
-            attention_mask = attention_mask[:, :, :, :kv_seq_len]
-        # repeat k/v heads if n_kv_heads < n_heads
-        key_states = repeat_kv(key_states, self.num_key_value_groups)
-        value_states = repeat_kv(value_states, self.num_key_value_groups)
-        attn_output = sdpa(query_states.to(device, dtype=torch.float16),
-                           key_states.to(device, dtype=torch.float16),
-                           value_states.to(device, dtype=torch.float16),
-                           is_causal=True).to(hidden_states.dtype)
-    else:
-        attn_output = scaled_dot_product_attention(
-            query_states, key_states, value_states,
-            attention_mask, q_len == kv_seq_len
-        )
+    attn_output = scaled_dot_product_attention(
+        query_states, key_states, value_states,
+        attention_mask, q_len == kv_seq_len
+    )
 
     attn_output = attn_output.transpose(1, 2).contiguous()
     attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)

ipex_llm/transformers/npu_model.py

@@ -301,8 +301,7 @@ class _BaseAutoModelClass:
         model.share_memory()
 
         if not pipeline:
-            if (not hasattr(model, 'llm') and
-                    model.config.model_type in ["qwen2", "llama", "minicpm"]):
+            if model.config.model_type in ["qwen2", "llama", "minicpm"]:
                 from ipex_llm.transformers.npu_models.convert import optimize_llm_single_process
                 optimize_llm_single_process(
                     llm,
@@ -312,7 +311,8 @@ class _BaseAutoModelClass:
                     group_size=quantization_group_size,
                     qtype=qtype,
                     save_directory=save_directory,
-                    fuse_layers=fuse_layers
+                    fuse_layers=fuse_layers,
+                    has_llm=hasattr(model, "llm")
                 )
             else:
                 optimize_llm(

ipex_llm/transformers/npu_models/convert.py

@@ -449,7 +449,8 @@ def optimize_llm_single_process(
     group_size: int,
     qtype: str,
     save_directory: str,
-    fuse_layers: int=None
+    fuse_layers: int=None,
+    has_llm: bool=False
 ):
     from ipex_llm.transformers.npu_pipeline_model.convert_pipeline import convert_llm
     from .npu_llm_cpp import load_model_from_file
@@ -468,8 +469,13 @@ def optimize_llm_single_process(
         model.kv_len = kv_len
         model.model_ptr = model_ptr
         model.save_directory = save_directory
-        model.vocab_size = model.config.vocab_size
+        if model.config.vocab_size == 151666:
+            # for MiniCPM-V 2.6, 152064 is vocab_size of Qwen2-7B
+            model.vocab_size = 152064
+        else:
+            model.vocab_size = model.config.vocab_size
         model.logits_buffer = torch.empty(1, 1, model.vocab_size, dtype=torch.float32)
+        model.max_prompt_len = max_prompt_len
     except:
         invalidInputError(False,
                           "False to InitLLMPipeline.")
@@ -478,9 +484,10 @@ def optimize_llm_single_process(
     general_convert(model, PreTrainedModel, prepare_input_ids, "prepare_inputs_for_generation")
     general_convert(model, PreTrainedModel, causal_lm_forward)
     # patch generate function
-    import types
-    model.original_generate = model.generate
-    model.generate = types.MethodType(generate, model)
+    if not has_llm:
+        import types
+        model.original_generate = model.generate
+        model.generate = types.MethodType(generate, model)
     return model
 
 
@@ -491,9 +498,10 @@ def prepare_input_ids(
     else:  # prefill, reset the model here
         from .npu_llm_cpp import reset
         reset(self.model_ptr)
-    model_inputs = {
-        "input_ids": input_ids
-    }
+    if inputs_embeds is not None and past_key_values is None:
+        model_inputs = {"inputs_embeds": inputs_embeds}
+    else:
+        model_inputs = {"input_ids": input_ids}
     return model_inputs
 
 
@@ -511,17 +519,31 @@ def causal_lm_forward(
     return_dict: Optional[bool] = None,
 ) -> Union[Tuple, CausalLMOutputWithPast]:
     from .npu_llm_cpp import run_prefill_with_logits, run_decode_with_logits
-    if isinstance(input_ids[0], torch.Tensor):
-        input_list = input_ids[0].flatten().tolist()
-    else:
-        input_list = input_ids[0]
-    input_length = len(input_list)
-    if input_length > 1:
-        logits = run_prefill_with_logits(self.model_ptr, input_list,
-                                         self.logits_buffer, self.vocab_size)
+    if input_ids is not None:
+        if isinstance(input_ids[0], torch.Tensor):
+            input_list = input_ids[0].flatten().tolist()
+        else:
+            input_list = input_ids[0]
+        input_length = len(input_list)
+        if input_length > 1:
+            logits = run_prefill_with_logits(self.model_ptr, input_list,
+                                             self.logits_buffer, self.vocab_size)
+        else:
+            logits = run_decode_with_logits(self.model_ptr, input_list[0],
+                                            self.logits_buffer, self.vocab_size)
+    elif inputs_embeds is not None:
+        seq_len = inputs_embeds.shape[1]
+        pad_len = self.max_prompt_len - seq_len
+        inputs_embeds = torch.nn.functional.pad(inputs_embeds.to(torch.float16),
+                                                (0, 0, 0, pad_len), value=0.0)
+        logits = run_prefill_with_logits(self.model_ptr, None, self.logits_buffer,
+                                         self.vocab_size, inputs_embeds, seq_len)
     else:
-        logits = run_decode_with_logits(self.model_ptr, input_list[0],
-                                        self.logits_buffer, self.vocab_size)
+        invalidInputError(False, "Please specify either input_ids or inputs_embeds.")
+
+    if self.config.vocab_size == 151666:
+        # for MiniCPM-V 2.6
+        logits = logits[:, :, :151666]
 
     return CausalLMOutputWithPast(
         loss=None,

ipex_llm/transformers/npu_models/npu_llm_cpp.py

@@ -48,8 +48,8 @@ _lib = ctypes.cdll.LoadLibrary(_lib_path)
 _lib.load_model_from_file.argtypes = [ctypes.c_char_p]
 _lib.load_model_from_file.restype = ctypes.c_void_p
 
-_lib.run_prefill.argtypes = [ctypes.c_void_p, ctypes.POINTER(ctypes.c_int), ctypes.c_int,
-                             ctypes.c_float]
+_lib.run_prefill.argtypes = [ctypes.c_void_p, ctypes.c_void_p, ctypes.c_int,
+                             ctypes.c_float, ctypes.c_bool]
 _lib.run_prefill.restype = ctypes.POINTER(ctypes.c_float)
 
 _lib.run_decode.argtypes = [ctypes.c_void_p, ctypes.c_int, ctypes.c_float]
@@ -61,8 +61,10 @@ _lib.llm_sample_token.restype = ctypes.c_int
 _lib.reset.argtypes = [ctypes.c_void_p]
 _lib.reset.restype = None
 
-_lib.run_prefill_with_logits.argtypes = [ctypes.c_void_p, ctypes.POINTER(ctypes.c_int),
-                                         ctypes.c_int, ctypes.POINTER(ctypes.c_float), ctypes.c_int]
+_lib.run_prefill_with_logits.argtypes = [ctypes.c_void_p, ctypes.c_void_p,
+                                         ctypes.c_int, ctypes.POINTER(ctypes.c_float),
+                                         ctypes.c_int, ctypes.c_bool]
+
 _lib.run_prefill_with_logits.restype = None
 
 _lib.run_decode_with_logits.argtypes = [ctypes.c_void_p, ctypes.c_int,
@@ -77,7 +79,7 @@ def load_model_from_file(model_dir: str):
 def run_prefill(model_ptr, input_ids, vocab_size, repetition_penalty=1.0):
     input_ptr = (ctypes.c_int32 * len(input_ids))(*input_ids)
     input_len = len(input_ids)
-    plogits = _lib.run_prefill(model_ptr, input_ptr, input_len, repetition_penalty)
+    plogits = _lib.run_prefill(model_ptr, input_ptr, input_len, repetition_penalty, False)
     new_token = _lib.llm_sample_token(plogits, True, vocab_size)
     return new_token
 
@@ -88,12 +90,19 @@ def run_decode(model_ptr, input_id, vocab_size, repetition_penalty=1.0):
     return new_token
 
 
-def run_prefill_with_logits(model_ptr, input_ids, logits, vocab_size):
-    input_ptr = (ctypes.c_int32 * len(input_ids))(*input_ids)
-    input_len = len(input_ids)
+def run_prefill_with_logits(model_ptr, input_ids, logits, vocab_size,
+                            inputs_embeds=None, seq_len=None):
+    if input_ids is not None:
+        input_ptr = (ctypes.c_int32 * len(input_ids))(*input_ids)
+        input_len = len(input_ids)
+    else:
+        input_ptr = inputs_embeds.contiguous().data.data_ptr()
+        input_ptr = ctypes.cast(input_ptr, ctypes.c_void_p)
+        input_len = seq_len
     logits_ptr = logits.data.data_ptr()
     logits_ptr = ctypes.cast(logits_ptr, ctypes.POINTER(ctypes.c_float))
-    _lib.run_prefill_with_logits(model_ptr, input_ptr, input_len, logits_ptr, vocab_size)
+    _lib.run_prefill_with_logits(model_ptr, input_ptr, input_len, logits_ptr,
+                                 vocab_size, (input_ids is None))
     return logits
 
 

ipex_llm/transformers/npu_pipeline_model/qwen.py

@@ -34,6 +34,10 @@ def convert_lm_head_and_embedding(model, temp_dir, weight_dir,
     lm_head_n_splits = 1
     asym = getattr(model.config, "asym", False)
 
+    if vocab_size == 151666:
+        # for MiniCPM-V 2.6 lm_head on NPU
+        vocab_size = 152064
+
     if not isinstance(lm_head, SlicedLMHead):
         asym = lm_head.qtype == "asym_int4_rtn"
         if asym:

{ipex_llm-2.2.0b20250101.dist-info → ipex_llm-2.2.0b20250102.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250101
+Version: 2.2.0b20250102
 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 ==2.6.0b20250101 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250102 ; extra == 'cpp'
 Requires-Dist: setuptools ; extra == 'cpp'
 Provides-Extra: cpp-arl
-Requires-Dist: bigdl-core-cpp ==2.6.0b20250101 ; extra == 'cpp-arl'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250102 ; 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: dpcpp-cpp-rt ==2024.2.1 ; (platform_system == "Windows") and extra == 'cpp-arl'
@@ -65,7 +65,7 @@ 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 ==2.6.0b20250101 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.6.0b20250102 ; (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 +85,9 @@ 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 ==2.6.0b20250101 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250101 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250101 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250102 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250102 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250102 ; extra == 'xpu'
 Provides-Extra: xpu-2-1
 Requires-Dist: py-cpuinfo ; extra == 'xpu-2-1'
 Requires-Dist: protobuf ; extra == 'xpu-2-1'
@@ -102,9 +102,9 @@ 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 ==2.6.0b20250101 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250101 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250101 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250102 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250102 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250102 ; 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'
@@ -119,9 +119,9 @@ 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 ==2.6.0b20250101 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250101 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250101 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250102 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250102 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250102 ; 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'
@@ -141,9 +141,9 @@ 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 ==2.6.0b20250101 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250101 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250101 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250102 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250102 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250102 ; 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'
@@ -163,9 +163,9 @@ 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 ==2.6.0b20250101 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250101 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250101 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250102 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250102 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250102 ; 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-2.2.0b20250101.dist-info → ipex_llm-2.2.0b20250102.dist-info}/RECORD

@@ -2,7 +2,7 @@ ipex_llm/__init__.py,sha256=kSA9JjVLPlpN4YchWtfOybRh4XiP6d_VTYvzbAouPSU,2118
 ipex_llm/convert_model.py,sha256=jopEe6wu88ZPZfNFhgnQUu7807iciiWW_EMyTsVni5A,6816
 ipex_llm/llm_patching.py,sha256=becMYcawtR8lgl2yeRQhvvX6CLaq09WZGm9dDmLJWL0,3232
 ipex_llm/models.py,sha256=XROP6GLLrGQDlogGXpXZENbV143YNi6j0VPJeOdQ3Cg,1063
-ipex_llm/optimize.py,sha256=4VYz8vgxSnrqBJhz__eB7hCJSwrkNx_t_wvTLxPlPyI,12253
+ipex_llm/optimize.py,sha256=ml-qEpzsrWGcd-Wia6IxPBC1PhqT5pi_lp3VTOH_ns0,12415
 ipex_llm/cli/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
 ipex_llm/cli/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
 ipex_llm/cli/prompts/chat-with-llm.txt,sha256=PpSyd4FQQd-T7ptfXL9jZp7dgstevu1fsxWFa0IQ5Oc,216
@@ -41,35 +41,35 @@ ipex_llm/langchain/llms/transformerspipelinellm.py,sha256=vm522YPPwWxxAPVvQBtxRf
 ipex_llm/langchain/vllm/__init__.py,sha256=T-EbRT6GJ_8RCu-iLmSzcftOimXSPQf2d5X72AUAy2Y,874
 ipex_llm/langchain/vllm/vllm.py,sha256=6dxc-ZISZQrJilEa_HA827l75Dv9rcHpY_G6FdJ8BVs,7793
 ipex_llm/libs/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ipex_llm/libs/bloom-api.dll,sha256=2IyCxPjsahzooryTFYEq-7R6nkrliaZm8h3-Y6n35C4,36352
-ipex_llm/libs/bloom.dll,sha256=E2AQed3DUyh7u36R8_vi0Om92aDzkO-MP_aK4vehb58,506880
-ipex_llm/libs/gptneox-api.dll,sha256=dCc4_7Xl-Isu_xlGe9iZQfbSabN2V8WOByWyDIZLp_4,24576
-ipex_llm/libs/gptneox.dll,sha256=q3M6r7NEOL_CRooJqq3AZ2icota9FwlWsr4nab6izd0,567296
-ipex_llm/libs/libbloom_avx.dll,sha256=3KbBuBjRFd1k9fSaXzkIbXNB4zPuyQVEZt-9bx_NJOc,535040
-ipex_llm/libs/libbloom_vnni.dll,sha256=x_rnRwlhmUs0d7E_mkn77VBFd-_MtiwfoWCb6TOyKCc,506880
-ipex_llm/libs/libgptneox_avx.dll,sha256=VG1twX6up-CqWVgH_bhYXXAg6MhL8oP8CZL6oiaM8pg,595456
-ipex_llm/libs/libgptneox_vnni.dll,sha256=r6Pdq6XjvouI0NispOQco-CM4xSewQQThOBRmijgwyQ,567808
-ipex_llm/libs/libllama_avx.dll,sha256=mUHLFVirXIR2viOeKkugcuHEBJpSLmsc3d6V9Y0zDz4,589824
-ipex_llm/libs/libllama_vnni.dll,sha256=FXWgzdFnPw4o_UAAktlEZNDev8CT7SCdwzs3zV8mlMY,561664
-ipex_llm/libs/libstarcoder_avx.dll,sha256=ErlXweTeHBexsQQLwoaiPspJjRYu7R22GTYLMidXQCw,626688
-ipex_llm/libs/libstarcoder_vnni.dll,sha256=l54B9SqZDjmLRQlbsZqXauYfGo58xEiOctAaidVqnJw,598528
-ipex_llm/libs/llama-api.dll,sha256=cw4PwQE9d5eYEYi-7vY0aG83a28pFPfzlprW_YXzBg8,25600
-ipex_llm/libs/llama.dll,sha256=AiSnrnpJRvZrB3HIMTlecXpCblR0o7_fnHMdKclvCsg,561152
-ipex_llm/libs/main-bloom.exe,sha256=HfhuIi1jJa1gdkROK3yt97k8Q5noiYdfwXXKeI1d4XE,103424
-ipex_llm/libs/main-gptneox.exe,sha256=RX9FcsWS2oB9EKKT3DXZwkPWwSP9TMjysVxpk4e_FLQ,98816
-ipex_llm/libs/main-llama.exe,sha256=QQRN1avtgAnmrNLfrmGWL5BtOk9Id90nL8Y04OJzmMo,99840
-ipex_llm/libs/main-starcoder.exe,sha256=xVhlJ1m-H5W4c9YrdveFXumyedg23m4g0xZYqhaEbn4,157696
-ipex_llm/libs/pipeline.dll,sha256=XO75bZ1_OgLYxdPVeW3BNcd5d26YcO6sxVq0HHCueVU,72704
-ipex_llm/libs/quantize-bloom.exe,sha256=lU6EmSLCnlYi8vRHsXbq8sEGR8q6dPpyKS4cBQXNLyA,126464
-ipex_llm/libs/quantize-bloom_vnni.exe,sha256=rK2NcJzLcqd_I8_D_oX-VHsw8VCcLNYrF937D2LLaDU,127488
-ipex_llm/libs/quantize-gptneox.exe,sha256=UMzUlM4x5WT-J58IMhwTy2Hu0ZsNzC5UjjwtVapHQAQ,104448
-ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=WIAcb-27zWDc9jgaODw5mHa4YK7MU5QgGPVbdYcftPA,104960
-ipex_llm/libs/quantize-llama.exe,sha256=H_M_xrapNgrXuWfvSvvO5jPlGG45hn9nuceFj-hnhwA,109568
-ipex_llm/libs/quantize-llama_vnni.exe,sha256=8x-57l7IUfre4gHT-RM63-tsRayGCFHSIpAfYapP27E,110592
-ipex_llm/libs/quantize-starcoder.exe,sha256=U0-IhAZl2jRHLuf90PUSrw4d21CTNnWdoFtoqfMNdq4,127488
-ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=zCC8EJs9KUmglJklqndbPNrGfo9IKzB5Pwqo7aBZSBU,128512
-ipex_llm/libs/starcoder-api.dll,sha256=aX-nqvXEzk6GxiDB4nC0Qy1KBftizAhV_h1u6YhVzrw,21504
-ipex_llm/libs/starcoder.dll,sha256=fSQg0niB930j4GxEpEAFFMPhPlxZ_DnbLUy4OrM08Q4,598016
+ipex_llm/libs/bloom-api.dll,sha256=D4QwdMzMWEpvWCt6Qhf3TNufORG3i1vGsvrxKX0UeG0,36352
+ipex_llm/libs/bloom.dll,sha256=HiUWN81LFRa4ylyTNLlGbIiufHTlHr60HSpuvlaNcoM,506880
+ipex_llm/libs/gptneox-api.dll,sha256=U9_GilTEOKSvrLXbDd3-iAUjjJUtP0Ud-KOsi-51Xjo,24576
+ipex_llm/libs/gptneox.dll,sha256=RL0yZh9g1sj98wA8Ekx7xr2S_PSad5Ll-uhdF05ev64,567296
+ipex_llm/libs/libbloom_avx.dll,sha256=2ouFAhOxkBYHMerLJHkefV1XbgRHDFqN-KBneiH9g-I,535040
+ipex_llm/libs/libbloom_vnni.dll,sha256=Ab9mZrSYvKE82yJ70VQDpl8qI_0GaIOOidEWLsJxbAo,506880
+ipex_llm/libs/libgptneox_avx.dll,sha256=3dV2YKFjy_u5tYZayi2URq7hpmDIqGafalP7WQEK-Tw,595456
+ipex_llm/libs/libgptneox_vnni.dll,sha256=-xfG7NBOcISipqhp8qOhtxwpCKdHtWxQai3FMoLQLhQ,567808
+ipex_llm/libs/libllama_avx.dll,sha256=gKtT8c3qklcvigz1POAbRtBwAdvsPFyXS7yZVipOTBM,589824
+ipex_llm/libs/libllama_vnni.dll,sha256=i9wyOe9xqShLO_DlYbo7cn1b47uMJ4pwPwVx88EyGzY,561664
+ipex_llm/libs/libstarcoder_avx.dll,sha256=RQNEbNTNCSv3qgDcKJCx87GB3eBMysL-UEPQdYj3WTE,626688
+ipex_llm/libs/libstarcoder_vnni.dll,sha256=7luUJRSUDj9AyEEemYTxJzs4bcx8MqM0sLukiZfnVxc,598528
+ipex_llm/libs/llama-api.dll,sha256=CwGyWmsT8XHnRiuZ4OzBIIlHwRlkiu_IaovirJqwbRY,25600
+ipex_llm/libs/llama.dll,sha256=LAyZioZRUs_CCv4vo5cqzUWpz1MQMc_c4n0pLaygFvA,561152
+ipex_llm/libs/main-bloom.exe,sha256=qMBUiZQZcr2RLc0U2cnnGUig9S3F70bqHRjwUsKV_Ns,103424
+ipex_llm/libs/main-gptneox.exe,sha256=wsi7_iK9sTA4PNBW5_y_Jpa1gqcE1eK97DFv_lB8RgY,98816
+ipex_llm/libs/main-llama.exe,sha256=BSBjl3RX7QzMq7cy5BVrXEzGcXiauSsKZ9VL_y8w-qI,99840
+ipex_llm/libs/main-starcoder.exe,sha256=qzk78keT_JNdVi7cdf-ycpX2QD6N0ZkO0Z9GDUqZgqc,157696
+ipex_llm/libs/pipeline.dll,sha256=C5E1tWcpdFJZVVjr7CfbcJDULeQom9hIjX6Zgb_rbso,72704
+ipex_llm/libs/quantize-bloom.exe,sha256=35E6hdsDi_njtgEcAaSE9WwelkHd6yT9IV8fWROx-ts,126464
+ipex_llm/libs/quantize-bloom_vnni.exe,sha256=_ECFuoYcxXPDob77sTS6vZvjufPg2su3BhcUZtP__AU,127488
+ipex_llm/libs/quantize-gptneox.exe,sha256=dJsoz68WhbSOF0I5LD3D5AIypsAPJC-aJAqIG-XHjrA,104448
+ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=4rXbo9XEh8Ka_bYyXCgIW68qLSHGkYwkrwFhiNkLeXM,104960
+ipex_llm/libs/quantize-llama.exe,sha256=s1jEeBhAy4DBAUPQ7DpZuqavX5Thy1yff-U9ENcYZPA,109568
+ipex_llm/libs/quantize-llama_vnni.exe,sha256=a4C5W-pxItLAaiVYGrHXkHsEH_nB0sVmnd5fyQv19ec,110592
+ipex_llm/libs/quantize-starcoder.exe,sha256=Uhz8MG9ElW__KmkjR89QgHhqJvESste0iISLw0pvgXw,127488
+ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=lv3V22NpQs_H_6HgORxigYv47OzEECECrz4IzYmxcjs,128512
+ipex_llm/libs/starcoder-api.dll,sha256=-6pbKFCVrDXJ5N-KPAIN4tvmRlphbKxUOoYMR_1iFYc,21504
+ipex_llm/libs/starcoder.dll,sha256=e40-bxKji-RYl12eW2VwJB-m70A6bJQyR75TsBZGmT8,598016
 ipex_llm/llamaindex/__init__.py,sha256=T-EbRT6GJ_8RCu-iLmSzcftOimXSPQf2d5X72AUAy2Y,874
 ipex_llm/llamaindex/llms/__init__.py,sha256=KP1lEdGqDuxPoxL1ZSH25Pm2kKMPJBWUTLR0ckSLMIU,1139
 ipex_llm/llamaindex/llms/bigdlllm.py,sha256=FQBzq1KOjfc6uofTXAha3O7TqpJkNfOFepXQmOVlbnI,26314
@@ -87,17 +87,17 @@ 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=B4oI836JHEqg_qT3dcl2RaJdQs7rOyigMwj-racxhkc,106379
+ipex_llm/transformers/convert.py,sha256=V4KDyi-2FVWSYZAxe4PlAxGGZbauSbOCuqq56ME9yyQ,106461
 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=TJfEqNp6zB6YnNEUASga302WQXzNdrmU_miGCM0u-F8,41504
-ipex_llm/transformers/model.py,sha256=N-g9IQVvBiBhbL5Fo5DTWbmHPZY52sjfFuq0B8Qu6h4,40952
+ipex_llm/transformers/low_bit_linear.py,sha256=dyyYyCqw0GK8hzaUGanrg-uIhU1HTLEEbvbxXMlm-80,41668
+ipex_llm/transformers/model.py,sha256=KcRjkauGg48BYrUBoUZaVMpg7Piuz5JrfIpVZd3EIjs,41105
 ipex_llm/transformers/modelling_bigdl.py,sha256=7JpNVMuyq_OmtNUaMFMXdxPWZp2q0QHC02QeA-VTPOw,6709
-ipex_llm/transformers/npu_model.py,sha256=wPFEB4W1rYbpO_XqepREMef69dzo-zkFoqFRb_mqneA,37862
+ipex_llm/transformers/npu_model.py,sha256=a1mkyc6EqD7AJhqbYzokGhFubNpt5trIMuZT_dQKlTk,37861
 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
@@ -136,10 +136,10 @@ 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=0dkTSPqGPgSnwa8zSNKroam0pvSyQLSRpbb43-OgIlc,19815
+ipex_llm/transformers/models/baichuan.py,sha256=oJCAEENSG8oQhJ-QPN2SiapARjAGdOM6nEbyCcYOMCo,19334
 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=xCEhYzaXyTDBXqz111Uw4IW5x4TLbtBbYfmBT623gRI,12669
+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/chatglm4.py,sha256=AAhAFFDDas5DBQPfh2Mwl7a2v7taKf6xphoeeNNFaBI,16593
 ipex_llm/transformers/models/chatglm4v.py,sha256=YRfuf9g1E0MQ_7wbHAOMvadFnO-j3LqI_k1SaRkDs0M,14055
@@ -167,8 +167,8 @@ ipex_llm/transformers/models/mpt.py,sha256=z02NwHogJZVh-Mk4sYoIzR90SFIKhoNN_-ifs
 ipex_llm/transformers/models/phi.py,sha256=E6qz4EEuHIVGvaPo-wtLC5lz3iyMqTbAE_cRlcjQRKI,6670
 ipex_llm/transformers/models/phi3.py,sha256=jkiadJ85ToHpymY5GOM6orWlnx6LKN8_-v1MUcfGWPg,15159
 ipex_llm/transformers/models/phixtral.py,sha256=MDTMghcu7qAmZmRcUGqXXDXhSU3y_N59HRIXmlcjp5g,4890
-ipex_llm/transformers/models/qwen.py,sha256=iP4wcjdIZ0CvqbM8muM96y-rghpEnZSR3TgjiOBIq5k,20475
-ipex_llm/transformers/models/qwen2.py,sha256=k_FcPfPAXIotRE03ULFBYA5hWE0M5CfktAhCWbEy8Yw,26369
+ipex_llm/transformers/models/qwen.py,sha256=XIJ_bLzediBURWU-OOS3H6WBIGXQue6jDdUHJsAabwY,19391
+ipex_llm/transformers/models/qwen2.py,sha256=b49HO4GSudwGJ3n6uHVno1oo3DgRt3jOjtQnLOB3cdY,25530
 ipex_llm/transformers/models/qwen2_moe.py,sha256=EA_OYxYAEgrvi7VpDW192AJXG9Fwe2aBtOAZPkOAJk4,19350
 ipex_llm/transformers/models/qwen2_vl.py,sha256=jIm4yZSd751BkRqgj3wR1QBkDIh-TMCLAMM8SZ8n6Qo,13419
 ipex_llm/transformers/models/qwen_vl.py,sha256=j7Nzzz2Qvynu9yrCXmoEfERjw43hXof5TbXIs7Ms-oY,17105
@@ -185,7 +185,7 @@ ipex_llm/transformers/npu_models/baichuan_mp.py,sha256=tHhO-0v5z6IhxsfzAPYWXVbLr
 ipex_llm/transformers/npu_models/chatglm.py,sha256=YzpGLZ7ORt6qkwW9mCwZ_xhOAI8uHSDHJrmqWgNM234,10511
 ipex_llm/transformers/npu_models/chatglm4.py,sha256=J4523DzhIzZxIvlf1V9qU4auzEGKvC80YqyxuCJygjw,9795
 ipex_llm/transformers/npu_models/common.py,sha256=tTUJL7IxVrJSnXle6nla35wTUrBf2sOEt7Ya1qyMezY,4853
-ipex_llm/transformers/npu_models/convert.py,sha256=cX10r7Q0a2qFcEIhRcvmw1eSmUtmeOcoNn8kVqv3Su8,24224
+ipex_llm/transformers/npu_models/convert.py,sha256=FILSGnoltcR9FMrCkw0eOKh6p3sbBI5i0Ms8AsJc04E,25342
 ipex_llm/transformers/npu_models/convert_mp.py,sha256=t7160V4MmYpnex2NfuLTcqoc1meGEXdYi4AAPotfbzk,24518
 ipex_llm/transformers/npu_models/glm_edge.py,sha256=VsJex-6530h4ZQk35TxRe1MnttAHT41omE8LV47LgBE,6723
 ipex_llm/transformers/npu_models/kv.py,sha256=2OSFO9Z6e4nGdVxXEM-Bq2qa_npYYbGmQt3lcCZxTlU,9201
@@ -198,7 +198,7 @@ ipex_llm/transformers/npu_models/minicpm_mp.py,sha256=0iCRWN9UIUQp5tSKyu-orpGCOx
 ipex_llm/transformers/npu_models/minicpmv_mp.py,sha256=m11WT6s_H5wkFtlz7aHMOL9b_CoL_G5MhoL5te4la_Q,20147
 ipex_llm/transformers/npu_models/mistral.py,sha256=iRdmIQI_bbbZxRCYRvnV4rWjX2t-6vkHNl1ICAsLoy4,10759
 ipex_llm/transformers/npu_models/mp_models_base.py,sha256=rY-5tq8DfxRsiaIITl0PQOTiPLJnUm_5L-oWzbK12N8,28429
-ipex_llm/transformers/npu_models/npu_llm_cpp.py,sha256=SL1p5UBjheufhum-GktNQQ4iXjRlZ9Hgatzb3NFr6Bs,3900
+ipex_llm/transformers/npu_models/npu_llm_cpp.py,sha256=B40sBujvy31ETFBgcYAf4CN23UuTCBEJVaxjIMaoEHk,4268
 ipex_llm/transformers/npu_models/paraformer_mp.py,sha256=lGEjmKHW_Pk3BE3nqa1ZVgJ3P5p4lNp7p6wMV7KrtCU,37871
 ipex_llm/transformers/npu_models/phi3.py,sha256=R-EuqHsTrPTX33HtCGAMFlRdXB_j5mH_7FDnj62JtNM,6555
 ipex_llm/transformers/npu_models/phi3_v.py,sha256=EMZuTPkGfuDVp9c5BU1HyzXHWKswHRQ8bvQjzocIyHA,7737
@@ -213,7 +213,7 @@ ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py,sha256=953Gua2tFKLI
 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
-ipex_llm/transformers/npu_pipeline_model/qwen.py,sha256=iVUNrcRLXE5eQGZIetgkLTINo8RW15RHM7SUetvJNRQ,14748
+ipex_llm/transformers/npu_pipeline_model/qwen.py,sha256=3paMXr1viuztybhmVLqQ9XvM3EZbxncDuNSNwLF8OI0,14849
 ipex_llm/utils/__init__.py,sha256=NdB_InYE65dNgW4ruEPUOlgKEO2ELcsJoqkP7O5kpog,1391
 ipex_llm/utils/benchmark_util_4_29.py,sha256=OU1W1quiaiJGsg1pd3HM9O6PmVSaPA0HHE7R8hNTfmQ,258653
 ipex_llm/utils/benchmark_util_4_42.py,sha256=HEiClCgKDp_T64HH8ulSTly8dvt6UwPDYZfrPVYvXcc,225383
@@ -244,11 +244,11 @@ 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-2.2.0b20250101.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
-ipex_llm-2.2.0b20250101.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
-ipex_llm-2.2.0b20250101.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
-ipex_llm-2.2.0b20250101.dist-info/METADATA,sha256=J_SK__bQX0TB_vOUTfpk9DV4DhTzVc-LzVzN_LmERV8,11374
-ipex_llm-2.2.0b20250101.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
-ipex_llm-2.2.0b20250101.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
-ipex_llm-2.2.0b20250101.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
-ipex_llm-2.2.0b20250101.dist-info/RECORD,,
+ipex_llm-2.2.0b20250102.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
+ipex_llm-2.2.0b20250102.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
+ipex_llm-2.2.0b20250102.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
+ipex_llm-2.2.0b20250102.dist-info/METADATA,sha256=fF_EkmZQW5wODRZlaJEQgQnS6Xieiem4h1vZcvsRxRE,11374
+ipex_llm-2.2.0b20250102.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
+ipex_llm-2.2.0b20250102.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
+ipex_llm-2.2.0b20250102.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
+ipex_llm-2.2.0b20250102.dist-info/RECORD,,