ipex-llm 2.2.0b20250227__py3-none-win_amd64.whl → 2.2.0b20250228__py3-none-win_amd64.whl

ipex_llm/transformers/models/chatglm4v.py

@@ -19,7 +19,7 @@
 
 import torch
 from typing import Optional, Tuple, Union
-from ipex_llm.transformers.models.common import merge_qkv_base
+from ipex_llm.transformers.models.common import merge_qkv_base, padding_qkv_hd
 from ipex_llm.transformers.models.common import scaled_dot_product_attention
 from ipex_llm.transformers.models.utils import update_past_key_value
 from ipex_llm.transformers.models.utils import use_quantize_kv_cache, use_sdp
@@ -265,26 +265,18 @@ def visual_attention_forward(self, x: "tensor(B, L, D)") -> "tensor(B, L, D)":
     q, k, v = qkv[0], qkv[1], qkv[2]
 
     bsz, q_len, kv_seq_len, head_dim = q.shape
-    if use_sdp(q_len, kv_seq_len, head_dim, q):
-        import xe_addons
-        out = xe_addons.sdp(q, k, v, None)
-    elif q.device.type == "cpu":
-        out = torch.nn.functional.scaled_dot_product_attention(q, k, v,
-                                                               attn_mask=None,
-                                                               dropout_p=0.,
-                                                               is_causal=False)
-    else:
-        attn_weights = torch.matmul(q / math.sqrt(head_dim),
-                                    k.transpose(2, 3)).to(v.dtype)
-        if kv_seq_len >= 2048 or bsz >= 64:
-            # for memory considerations, do not upcast attention to fp32
-            # for long sequences or large batches
-            attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1)
-        else:
-            # upcast attention to fp32
-            attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1,
-                                                       dtype=torch.float32).to(v.dtype)
-        out = torch.matmul(attn_weights, v)
+    q, k, v = padding_qkv_hd(
+        q, k, v,
+        head_dim, 128
+    )
+
+    attn_weights = None
+    attn_output = scaled_dot_product_attention(
+        q, k.contiguous(), v.contiguous(),
+        None, False, 1 / math.sqrt(head_dim)
+    )
+
+    out = attn_output[:, :, :, :head_dim]
     output = self.dense(out.transpose(1, 2).reshape(B, L, -1))
     output = self.output_dropout(output)
     return output

ipex_llm/transformers/models/deepseek.py

@@ -271,6 +271,25 @@ def deepseek_attention_forward(
     return attn_output, attn_weights, past_key_value
 
 
+def fuse_gate_forward(self, x: torch.Tensor):
+    if x.device.type == "xpu" and x.dtype in [torch.float, torch.half]:
+        x = x.view(-1, x.size(-1))
+        logits = torch.nn.functional.linear(
+            x.type(torch.float32), self.weight.type(torch.float32), None
+        )
+        scores = logits.sigmoid()
+
+        import xe_addons
+        topk_idx, topk_weight = xe_addons.moe_group_topk(
+            scores, self.e_score_correction_bias,
+            self.n_group, 2, self.topk_group, self.top_k,
+            self.top_k > 1 and self.norm_topk_prob, 1e-20, self.routed_scaling_factor
+        )
+    else:
+        topk_idx, topk_weight = self(x)
+    return topk_idx, topk_weight.to(x.dtype)
+
+
 def moe_infer_decode(self, x: torch.Tensor, topk_ids: torch.Tensor, topk_weight: torch.Tensor):
     if (
         x.device.type == "xpu"
@@ -301,7 +320,7 @@ def moe_infer_decode(self, x: torch.Tensor, topk_ids: torch.Tensor, topk_weight:
             expert_out = expert(x)
             outputs.append(expert_out)
         outs = torch.cat(outputs, dim=0)
-        reshaped_topk_weight = topk_weight.squeeze(0).unsqueeze(-1).to(outs.dtype)
+        reshaped_topk_weight = topk_weight.squeeze(0).unsqueeze(-1)
         final_out = (outs * reshaped_topk_weight).sum(dim=0, keepdim=True)
     return final_out
 
@@ -309,11 +328,13 @@ def moe_infer_decode(self, x: torch.Tensor, topk_ids: torch.Tensor, topk_weight:
 def deepseek_moe_forward(self, hidden_states: torch.Tensor):
     identity = hidden_states
     orig_shape = hidden_states.shape
-    topk_idx, topk_weight = self.gate(hidden_states)
+    # IPEX-LLM OPT start: fuse grouped topk in gate forward
+    topk_idx, topk_weight = fuse_gate_forward(self.gate, hidden_states)
+    # IPEX-LLM OPT end
     hidden_states = hidden_states.view(-1, hidden_states.shape[-1])
     flat_topk_idx = topk_idx.view(-1)
     if not self.training:
-        # IPEX-LLM OPT start : add special moe_infer implementation for decoding
+        # IPEX-LLM OPT start: add special moe_infer implementation for decoding
         if topk_idx.size(0) == 1 and self.ep_size == 1:
             y = moe_infer_decode(self, hidden_states, topk_idx, topk_weight)
         else:

{ipex_llm-2.2.0b20250227.dist-info → ipex_llm-2.2.0b20250228.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ipex-llm
-Version: 2.2.0b20250227
+Version: 2.2.0b20250228
 Summary: Large Language Model Develop Toolkit
 Home-page: https://github.com/intel-analytics/ipex-llm
 Author: BigDL Authors
@@ -27,7 +27,7 @@ Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine
 Requires-Dist: torch ==2.1.2+cpu ; (platform_system == "Linux") and extra == 'all'
 Requires-Dist: torch ==2.1.2 ; (platform_system == "Windows") and extra == 'all'
 Provides-Extra: cpp
-Requires-Dist: bigdl-core-cpp ==2.6.0b20250227 ; extra == 'cpp'
+Requires-Dist: bigdl-core-cpp ==2.6.0b20250228 ; extra == 'cpp'
 Requires-Dist: setuptools ; extra == 'cpp'
 Requires-Dist: onednn-devel ==2025.0.1 ; (platform_system == "Windows") and extra == 'cpp'
 Requires-Dist: onednn ==2025.0.1 ; (platform_system == "Windows") and extra == 'cpp'
@@ -60,7 +60,7 @@ Requires-Dist: transformers ==4.40.0 ; extra == 'npu'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'npu'
 Requires-Dist: torch ==2.1.2+cpu ; (platform_system == "Linux") and extra == 'npu'
 Requires-Dist: torch ==2.1.2 ; (platform_system == "Windows") and extra == 'npu'
-Requires-Dist: bigdl-core-npu ==2.6.0b20250227 ; (platform_system == "Windows") and extra == 'npu'
+Requires-Dist: bigdl-core-npu ==2.6.0b20250228 ; (platform_system == "Windows") and extra == 'npu'
 Provides-Extra: serving
 Requires-Dist: py-cpuinfo ; extra == 'serving'
 Requires-Dist: fschat[model_worker,webui] ==0.2.36 ; extra == 'serving'
@@ -80,9 +80,9 @@ Requires-Dist: setuptools <70.0.0 ; extra == 'xpu'
 Requires-Dist: torch ==2.1.0a0 ; extra == 'xpu'
 Requires-Dist: torchvision ==0.16.0a0 ; extra == 'xpu'
 Requires-Dist: intel-extension-for-pytorch ==2.1.10+xpu ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250227 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250227 ; extra == 'xpu'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250227 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250228 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250228 ; extra == 'xpu'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250228 ; extra == 'xpu'
 Provides-Extra: xpu-2-1
 Requires-Dist: py-cpuinfo ; extra == 'xpu-2-1'
 Requires-Dist: protobuf ; extra == 'xpu-2-1'
@@ -97,9 +97,9 @@ Requires-Dist: setuptools <70.0.0 ; extra == 'xpu-2-1'
 Requires-Dist: torch ==2.1.0a0 ; extra == 'xpu-2-1'
 Requires-Dist: torchvision ==0.16.0a0 ; extra == 'xpu-2-1'
 Requires-Dist: intel-extension-for-pytorch ==2.1.10+xpu ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250227 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250227 ; extra == 'xpu-2-1'
-Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250227 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-21 ==2.6.0b20250228 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-batch-21 ==2.6.0b20250228 ; extra == 'xpu-2-1'
+Requires-Dist: bigdl-core-xe-addons-21 ==2.6.0b20250228 ; extra == 'xpu-2-1'
 Requires-Dist: intel-openmp ; (platform_machine == "x86_64" or platform_machine == "AMD64") and extra == 'xpu-2-1'
 Requires-Dist: dpcpp-cpp-rt ==2024.0.2 ; (platform_system == "Windows") and extra == 'xpu-2-1'
 Requires-Dist: mkl-dpcpp ==2024.0.0 ; (platform_system == "Windows") and extra == 'xpu-2-1'
@@ -117,7 +117,7 @@ Requires-Dist: setuptools ; extra == 'xpu-2-6'
 Requires-Dist: torch ==2.6.0+xpu ; extra == 'xpu-2-6'
 Requires-Dist: torchvision ==0.21.0+xpu ; extra == 'xpu-2-6'
 Requires-Dist: torchaudio ==2.6.0+xpu ; extra == 'xpu-2-6'
-Requires-Dist: bigdl-core-xe-all ==2.6.0b20250227 ; extra == 'xpu-2-6'
+Requires-Dist: bigdl-core-xe-all ==2.6.0b20250228 ; extra == 'xpu-2-6'
 Requires-Dist: onednn-devel ==2025.0.1 ; extra == 'xpu-2-6'
 Requires-Dist: onednn ==2025.0.1 ; extra == 'xpu-2-6'
 Requires-Dist: dpcpp-cpp-rt ==2025.0.2 ; extra == 'xpu-2-6'
@@ -133,9 +133,9 @@ 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 ==2.6.0b20250227 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250227 ; extra == 'xpu-arc'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250227 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250228 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250228 ; extra == 'xpu-arc'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250228 ; 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'
@@ -156,9 +156,9 @@ 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 ==2.6.0b20250227 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250227 ; extra == 'xpu-arl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250227 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250228 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250228 ; extra == 'xpu-arl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250228 ; 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'
@@ -179,9 +179,9 @@ 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 ==2.6.0b20250227 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250227 ; extra == 'xpu-lnl'
-Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250227 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-23 ==2.6.0b20250228 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-batch-23 ==2.6.0b20250228 ; extra == 'xpu-lnl'
+Requires-Dist: bigdl-core-xe-addons-23 ==2.6.0b20250228 ; 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-2.2.0b20250227.dist-info → ipex_llm-2.2.0b20250228.dist-info}/RECORD

@@ -41,35 +41,35 @@ ipex_llm/langchain/llms/transformerspipelinellm.py,sha256=vm522YPPwWxxAPVvQBtxRf
 ipex_llm/langchain/vllm/__init__.py,sha256=T-EbRT6GJ_8RCu-iLmSzcftOimXSPQf2d5X72AUAy2Y,874
 ipex_llm/langchain/vllm/vllm.py,sha256=6dxc-ZISZQrJilEa_HA827l75Dv9rcHpY_G6FdJ8BVs,7793
 ipex_llm/libs/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ipex_llm/libs/bloom-api.dll,sha256=65MtrTb-0rDoLxFrTYjVZyLa-Ihm4qnfVxoj1kgGTRc,36352
-ipex_llm/libs/bloom.dll,sha256=OIwHSeyyJzVXGk9YC2-wRU3_NIIecUBWBJB6yDb3ev0,507904
-ipex_llm/libs/gptneox-api.dll,sha256=DuhoJWs6t3f1ipeC2SAcG3ijuapbPLYZamUuun35p5Y,24576
-ipex_llm/libs/gptneox.dll,sha256=zeOzWXzzbvvwvGktaQHuldzrbXWixGXn56zsEKyN-KU,568320
-ipex_llm/libs/libbloom_avx.dll,sha256=GFuVGKafurXEH3oQLENrLzUGEMUbimiwXG0fM0lFeJ4,536576
-ipex_llm/libs/libbloom_vnni.dll,sha256=4bb-gHHWI5p-iw8rKfcstQyMPxIKTiFscOLaUtKlWVQ,508416
-ipex_llm/libs/libgptneox_avx.dll,sha256=wra-GIZVmRxLd92_Ni1UE-aGQupXRv03rZYZorp-Te0,596992
-ipex_llm/libs/libgptneox_vnni.dll,sha256=EDYhgdm0DYtp2dBrM_5GHH7uQECet9e_Q4F6MoZze-I,568832
-ipex_llm/libs/libllama_avx.dll,sha256=s01TJZeHHoYisRaW-1NUYcZg_ZbE6m35KwkAYMFqRs4,591360
-ipex_llm/libs/libllama_vnni.dll,sha256=PYJ1pTeKcu2F1FPW6Jgg74S3dVohlMBp1wR_20bCfws,563200
-ipex_llm/libs/libstarcoder_avx.dll,sha256=dGkEXQ_QMxpz9v4qOF-9RAIMulbaRXvH2Fe-66hukYs,627712
-ipex_llm/libs/libstarcoder_vnni.dll,sha256=A5J7GUshakFbuGCMeMKiOoXdA2NHCMop6iXyaBT0yD4,599552
-ipex_llm/libs/llama-api.dll,sha256=gfSMVyetJXlba_bc3IR5MDs9ghBSlO86jQgaoZf8m6U,25600
-ipex_llm/libs/llama.dll,sha256=mbjHEwy298ZxJINN01oIZ3SgDdKEau0667KuGNQVgjg,562688
-ipex_llm/libs/main-bloom.exe,sha256=CR3dd0gvPmMVbNVyD8wC5zazMvd6hfJsUS2zFV3ho_k,103424
-ipex_llm/libs/main-gptneox.exe,sha256=xgWT3UmysnlustVwGYOBPR89hdr0v_l7_o3RPHWe5sY,98816
-ipex_llm/libs/main-llama.exe,sha256=q6UwR3P9hs6YR-dWHpMOa4tA_RwBEKEgIqmelo6XTwE,99840
-ipex_llm/libs/main-starcoder.exe,sha256=vYHaNA-xMiAaEN4jGu4vftCNbDn01Osi86THsvjqGR8,157696
-ipex_llm/libs/pipeline.dll,sha256=pKdP3VCtXQ8N4-vJCqZBKjIsLvxh88vswPVJ33lTq8U,73216
-ipex_llm/libs/quantize-bloom.exe,sha256=BqcU5KAZZ9gCgVRoKE_6vRkIOEsm_gcZWKdtxQOjYco,126464
-ipex_llm/libs/quantize-bloom_vnni.exe,sha256=Kn_Tx6o7vgdjvgJDFgl0fQ3RO5ocgWPg-Bzld39-mKg,128000
-ipex_llm/libs/quantize-gptneox.exe,sha256=MMGf6-PI0QBxgtmRy6eTr8PNwwc34yrIUf7Jz8AGjtg,104448
-ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=dAGRCSISGIxZYQLBL7gJ7hWCp9WBhpVZhqhLqFz7cBc,104960
-ipex_llm/libs/quantize-llama.exe,sha256=mEBTmFgSiCNvVZdsjjZaSci4l2s2yKrjo2X6WIdGlV8,110080
-ipex_llm/libs/quantize-llama_vnni.exe,sha256=Bhk_MvFp_IsEp8eOE6A85vRhjRblA98bBnNycgiG258,110592
-ipex_llm/libs/quantize-starcoder.exe,sha256=l6CMfpkXu22cr8_KX52zUWHRFHPAQpARSPV_BgXW4Ks,127488
-ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=c1NuzFqHtTZlkecWe7Vq3D5-iCpuu_haJVLY75ZDkQ4,128512
-ipex_llm/libs/starcoder-api.dll,sha256=vxpPhx8UmBnfZtoX2uc05c2tB1hAy2W-MrUWXBSK8o4,21504
-ipex_llm/libs/starcoder.dll,sha256=bv-NzJaI2G6AsKZJZiMIkz75Bo7ntj02Q_DD99U6vq0,599040
+ipex_llm/libs/bloom-api.dll,sha256=ixnQ2ZnnykU4Glxe69JZDWtqX7AkobeaAcjVU7Yf3j4,36352
+ipex_llm/libs/bloom.dll,sha256=EcRvb83qOVppvSqPqpYkpT5konP7I98u-6jCiTJawVM,507904
+ipex_llm/libs/gptneox-api.dll,sha256=63SdwgYRLd7A-3JOKHQCBIsTTdi18Ub5hbarfvWfTk8,24576
+ipex_llm/libs/gptneox.dll,sha256=HbDq4kTtcbMUnGv3JhaajlBIHGTWnVaVKB1fQTNudi8,568320
+ipex_llm/libs/libbloom_avx.dll,sha256=yzPhBVNqPF20ZTM1rF3eU826JVXmKxrti5p7mCo8BrI,536576
+ipex_llm/libs/libbloom_vnni.dll,sha256=IPOUZTWTQ_lmAKOF3S2Hg75BMArLzsm_N5Aqe_zpnmU,508416
+ipex_llm/libs/libgptneox_avx.dll,sha256=G1EW8Y2Em4ZRZPPYJtK08O-16EYIj6tTWMcuquA_qQo,596992
+ipex_llm/libs/libgptneox_vnni.dll,sha256=Uv_eGnuOLB4UxAnVfwwfY0jzAxa5CGzBRwj4zyU3I6c,568832
+ipex_llm/libs/libllama_avx.dll,sha256=UWbKBXsRxETPqYXOcUUTbCaNsm6RJvjLwr1IdOD4a0g,591360
+ipex_llm/libs/libllama_vnni.dll,sha256=1SKB0NzGdFV308cxjfDoqK8tI1iDSlrm0S2F5J3axVg,563200
+ipex_llm/libs/libstarcoder_avx.dll,sha256=5CfhJXtIpcT45PlW3Yvjp04t4ScT8qlUliOk5nUFdRQ,627712
+ipex_llm/libs/libstarcoder_vnni.dll,sha256=4-x8ULcUmoi-7Y8xFH7sE4yuNnlLdkG5TeGsgsfN3MU,599552
+ipex_llm/libs/llama-api.dll,sha256=44QKnI8wImM5KYR_ronbP6C3FwuSSs1Q9QLympftxaM,25600
+ipex_llm/libs/llama.dll,sha256=RijZAbUriRq8bL97DDpzc4BZ3XwfvjKzv3z-Rb9fjRI,562688
+ipex_llm/libs/main-bloom.exe,sha256=dxZ7ho-JuUXNd_DPqDqc3AOkIMMPgexUnd2ba49WlTo,103424
+ipex_llm/libs/main-gptneox.exe,sha256=iBps-o17U9TxtAR2099D_PwxfMmCr6t9yg2IpbDKnsU,98816
+ipex_llm/libs/main-llama.exe,sha256=xw9FcrmoyDyLrTnkfIJfr5_gD2TXP3lTP6tBNY0JrVE,99840
+ipex_llm/libs/main-starcoder.exe,sha256=fsqH16NcWWHf5AvzssxXgeweIKgFliHX8osVs03NUsw,157696
+ipex_llm/libs/pipeline.dll,sha256=LStM4D6Ta2qcV5Z6wKfoFZtAlp1nYs9tDXU57UlJjzo,73216
+ipex_llm/libs/quantize-bloom.exe,sha256=-q9aLdewOTuBhkaeMmQ9iNDphCT24e_NzIRvOQcEe5Y,126464
+ipex_llm/libs/quantize-bloom_vnni.exe,sha256=MHrN0-FOvjxZGIE4qLhrWzegCnNgAaK09X1VA8DPKFk,128000
+ipex_llm/libs/quantize-gptneox.exe,sha256=icJ6XQ0x-04FNfQsjaGQ9F1N2zYN0am3Tu7UousXuq4,104448
+ipex_llm/libs/quantize-gptneox_vnni.exe,sha256=ZAJ6HwreLXw1ZqxDjDAQrzB3b7VBNhDoF_mOpfmdzms,104960
+ipex_llm/libs/quantize-llama.exe,sha256=905BqROD6AlPRSRTH9n3oFh26Nf15DG1r1UTyv4yp1o,110080
+ipex_llm/libs/quantize-llama_vnni.exe,sha256=XC2ftT1GJ9osplZsX9i_IvDSkDzeu7kG9p-SAt6XjfM,110592
+ipex_llm/libs/quantize-starcoder.exe,sha256=_vPwU5OR1Ufp-GcbAhwViLyEMskqoJHU1EEzS1p0Qew,127488
+ipex_llm/libs/quantize-starcoder_vnni.exe,sha256=z_oEjL8CYNiFBWY4DPuwhk2lIiMm_757raqBMY7HrGk,128512
+ipex_llm/libs/starcoder-api.dll,sha256=ODp02d-lbYMW3IIxZ8ivbbQXPsINaH_lYfD5ige9T9Y,21504
+ipex_llm/libs/starcoder.dll,sha256=7eVBYqa4hQ1-kUmBZeRP6Lb7Mu_xJ2NROcGnVTTLOXY,599040
 ipex_llm/llamaindex/__init__.py,sha256=T-EbRT6GJ_8RCu-iLmSzcftOimXSPQf2d5X72AUAy2Y,874
 ipex_llm/llamaindex/llms/__init__.py,sha256=KP1lEdGqDuxPoxL1ZSH25Pm2kKMPJBWUTLR0ckSLMIU,1139
 ipex_llm/llamaindex/llms/bigdlllm.py,sha256=FQBzq1KOjfc6uofTXAha3O7TqpJkNfOFepXQmOVlbnI,26314
@@ -145,10 +145,10 @@ ipex_llm/transformers/models/bloom.py,sha256=PxfzyYT-nFn3K5rZhTQjmcEjUUzAhUFzxIN
 ipex_llm/transformers/models/chatglm.py,sha256=UHai1t2AUtGmF765_eHF8LUMVQzp_oCBx8TJB21WrHk,12597
 ipex_llm/transformers/models/chatglm2.py,sha256=KyAIX7zGVQDQuwwM3QMBNWZbTeMHEzKUIgAryT0voHc,14933
 ipex_llm/transformers/models/chatglm4.py,sha256=QvUehdaCePB3MNHyWg3dneDxmjtBdxYeKUyQUVcsgfM,16886
-ipex_llm/transformers/models/chatglm4v.py,sha256=L6y45M_wjS2_HqchmCUxRlQZUNuSNCGOiynAQrGh918,14124
+ipex_llm/transformers/models/chatglm4v.py,sha256=Ba9Xtzwtzk_rzg5khGqDrlHfJsDwc5YcM5_yPoord7o,13324
 ipex_llm/transformers/models/common.py,sha256=0OTRaXekOPApRdQ8UKl5Du8DOtKJ6awnQIStvYvFQOI,13018
 ipex_llm/transformers/models/decilm.py,sha256=P-PBuDPf07GvKggLwJx_wPwIn6esN3rX8ai2JxRuZmE,5246
-ipex_llm/transformers/models/deepseek.py,sha256=bMUAbTf2GaSyWuTwQxh_6LJqx7RvlqOQpCYw4DsC6BQ,12310
+ipex_llm/transformers/models/deepseek.py,sha256=w6tGeyJ9joD7lQBiZ6A01Z00g8hAXC1N2yGtJh8kyuk,13096
 ipex_llm/transformers/models/deepseek_v3.py,sha256=CTgwIKQlUPlUCbOxc9Id5GapWkXOP6pMtkguYrWpCio,10003
 ipex_llm/transformers/models/gemma.py,sha256=_E3Yw8Y45xyNVeLqyVKcpr8kjuICtETeL82cJ-bWJuU,9424
 ipex_llm/transformers/models/gemma2.py,sha256=2WZuv-FLzJyTJFaYxOuzJt47QE64M0lHnzAiO5T6ozI,8049
@@ -255,11 +255,11 @@ ipex_llm/vllm/xpu/engine/__init__.py,sha256=pY_CpyuZd72fr6s32ejeKHKFW0K4vUU2rzZj
 ipex_llm/vllm/xpu/engine/engine.py,sha256=NvCMbp0X8NVrOqbwm4FTvXOptTRLzu9jQsy37ZHnTk8,9493
 ipex_llm/vllm/xpu/entrypoints/openai/api_server.py,sha256=D577nxWlyoWaHXNXIEvS3ViKSSWL3XZq8D8t6izD7x4,33250
 ipex_llm/vllm/xpu/entrypoints/openai/cli_args.py,sha256=hB398yYtKauASRzevctScdbFIjiiSGMAe1bwEuIHrhY,10893
-ipex_llm-2.2.0b20250227.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
-ipex_llm-2.2.0b20250227.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
-ipex_llm-2.2.0b20250227.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
-ipex_llm-2.2.0b20250227.dist-info/METADATA,sha256=cyJrw4GeU1bOptch_YZF-a-a6336yn9oMn0DeyF5ktU,12369
-ipex_llm-2.2.0b20250227.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
-ipex_llm-2.2.0b20250227.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
-ipex_llm-2.2.0b20250227.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
-ipex_llm-2.2.0b20250227.dist-info/RECORD,,
+ipex_llm-2.2.0b20250228.data/scripts/ipex-llm-init.bat,sha256=HPtCYuDYwEatq7dAwOvdfVcHYCpAVdbj75K1qh0vQek,2578
+ipex_llm-2.2.0b20250228.data/scripts/llm-chat.ps1,sha256=6qrs-hGVAV8IKh7Jx8nq_XrnZcjd7qGU5wndArM7Yag,2769
+ipex_llm-2.2.0b20250228.data/scripts/llm-cli.ps1,sha256=3qBtTLs_EjYDnM8YyCpJhzLnGCKTEGssu9UNqfkjVXs,3009
+ipex_llm-2.2.0b20250228.dist-info/METADATA,sha256=jdYYlppmiOA9b1iPE21GxdjZCz4znzlmHq96YpMG9_M,12369
+ipex_llm-2.2.0b20250228.dist-info/WHEEL,sha256=6iYPr8vTHsyDK75jr9X0V3I9wPSVmtwr_8fdATBciGk,98
+ipex_llm-2.2.0b20250228.dist-info/entry_points.txt,sha256=TiUyBB2MRmfF3ko-pyAEzqeBCRnyhu27bNOAsWPp3e8,61
+ipex_llm-2.2.0b20250228.dist-info/top_level.txt,sha256=CGCMHM-SyqUabU4h8RqJ2KTYckQUO3LvIWwmUQ6Qbzw,9
+ipex_llm-2.2.0b20250228.dist-info/RECORD,,