[Hardware][Intel GPU] Upgrade to torch 2.7 (#17444)

Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Co-authored-by: Qiming Zhang <qiming1.zhang@intel.com>

Author: jikunshang

docker/Dockerfile.xpu

@@ -40,12 +40,6 @@ RUN --mount=type=cache,target=/root/.cache/pip \
     --mount=type=bind,source=.git,target=.git \
     python3 setup.py install
 
-# Please refer xpu doc, we need manually install intel-extension-for-pytorch 2.6.10+xpu due to there are some conflict dependencies with torch 2.6.0+xpu
-# FIXME: This will be fix in ipex 2.7. just leave this here for awareness.
-RUN --mount=type=cache,target=/root/.cache/pip \
-    pip install intel-extension-for-pytorch==2.6.10+xpu \
-    --extra-index-url=https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-
 CMD ["/bin/bash"]
 
 FROM vllm-base AS vllm-openai

docs/source/getting_started/installation/gpu/xpu.inc.md

@@ -35,13 +35,6 @@ pip install -v -r requirements/xpu.txt
 VLLM_TARGET_DEVICE=xpu python setup.py install
 ```
 
-- Finally, due to a known issue of conflict dependency(oneapi related) in torch-xpu 2.6 and ipex-xpu 2.6, we install ipex here. This will be fixed in the ipex-xpu 2.7.
-
-```console
-pip install intel-extension-for-pytorch==2.6.10+xpu \
-    --extra-index-url=https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-```
-
 :::{note}
 - FP16 is the default data type in the current XPU backend. The BF16 data
   type is supported on Intel Data Center GPU, not supported on Intel Arc GPU yet.
@@ -81,5 +74,3 @@ python -m vllm.entrypoints.openai.api_server \
 ```
 
 By default, a ray instance will be launched automatically if no existing one is detected in the system, with `num-gpus` equals to `parallel_config.world_size`. We recommend properly starting a ray cluster before execution, referring to the <gh-file:examples/online_serving/run_cluster.sh> helper script.
-
-There are some new features coming with ipex-xpu 2.6, e.g. **chunked prefill**, **V1 engine support**, **lora**, **MoE**, etc.

requirements/xpu.txt

@@ -10,14 +10,14 @@ wheel
 jinja2>=3.1.6
 datasets # for benchmark scripts
 
-torch==2.6.0+xpu
+torch==2.7.0+xpu
 torchaudio
 torchvision
 pytorch-triton-xpu
 --extra-index-url=https://download.pytorch.org/whl/xpu
 
 # Please refer xpu doc, we need manually install intel-extension-for-pytorch 2.6.10+xpu due to there are some conflict dependencies with torch 2.6.0+xpu
 # FIXME: This will be fix in ipex 2.7. just leave this here for awareness.
-# intel-extension-for-pytorch==2.6.10+xpu
-oneccl_bind_pt==2.6.0+xpu
+intel-extension-for-pytorch==2.7.10+xpu
+oneccl_bind_pt==2.7.0+xpu
 --extra-index-url=https://pytorch-extension.intel.com/release-whl/stable/xpu/us/

vllm/_ipex_ops.py

@@ -177,6 +177,7 @@ def varlen_attention(
         out: torch.Tensor,
         seqlen_q: torch.Tensor,
         seqlen_k: torch.Tensor,
+        alibi_slopes: torch.Tensor,
         max_seqlen_q: int,
         max_seqlen_k: int,
         pdropout: float,
@@ -185,6 +186,8 @@ def varlen_attention(
         is_causal: bool,
         return_softmax: bool,
         gen_: torch.Generator,
+        window_size_left: float,
+        window_size_right: float,
         logits_soft_cap: float,
     ) -> None:
         if ipex.__version__.endswith("cpu"):
@@ -200,15 +203,12 @@ def varlen_attention(
                                                  is_causal, return_softmax,
                                                  gen_)
         else:  # XPU build
-            ipex.llm.functional.varlen_attention(query.contiguous(),
-                                                 key.contiguous(),
-                                                 value.contiguous(), out,
-                                                 seqlen_q.int(),
-                                                 seqlen_k.int(), max_seqlen_q,
-                                                 max_seqlen_k, pdropout,
-                                                 softmax_scale, zero_tensors,
-                                                 is_causal, return_softmax,
-                                                 gen_, logits_soft_cap)
+            ipex.llm.functional.varlen_attention(
+                query.contiguous(), key.contiguous(), value.contiguous(), out,
+                seqlen_q.int(), seqlen_k.int(), alibi_slopes, max_seqlen_q,
+                max_seqlen_k, pdropout, softmax_scale, zero_tensors, is_causal,
+                return_softmax, gen_, window_size_left, window_size_right,
+                logits_soft_cap)
 
     @staticmethod
     def reshape_and_cache(

vllm/attention/backends/ipex_attn.py

@@ -143,10 +143,9 @@ def __init__(
 
         assert self.num_heads % self.num_kv_heads == 0
         self.num_queries_per_kv = self.num_heads // self.num_kv_heads
-        self.need_mask = (self.alibi_slopes is not None
-                          or self.sliding_window is not None)
+        self.need_mask = (self.sliding_window is not None)
         if logits_soft_cap is None:
-            logits_soft_cap = 0
+            logits_soft_cap = -1
         self.logits_soft_cap = logits_soft_cap
 
         supported_head_sizes = PagedAttention.get_supported_head_sizes()
@@ -234,11 +233,7 @@ def forward(
                                                     dim=1)
 
                 if attn_metadata.attn_bias is None:
-                    if self.alibi_slopes is not None:
-                        att_masks = _make_alibi_bias(
-                            self.alibi_slopes, query.dtype,
-                            attn_metadata.seq_lens)  # type: ignore
-                    elif self.sliding_window is not None:
+                    if self.sliding_window is not None:
                         att_masks = _make_sliding_window_bias(
                             attn_metadata.seq_lens, self.sliding_window,
                             query.dtype)  # type: ignore
@@ -258,6 +253,7 @@ def forward(
                     output,
                     attn_metadata.seqlen_q,
                     attn_metadata.seqlen_q,
+                    self.alibi_slopes,
                     attn_metadata.max_seqlen,
                     attn_metadata.max_seqlen,
                     pdropout=0.0,
@@ -266,6 +262,8 @@ def forward(
                     is_causal=True,
                     return_softmax=False,
                     gen_=None,
+                    window_size_left=-1,
+                    window_size_right=-1,
                     logits_soft_cap=self.logits_soft_cap,
                 )
             else: