IPEX support FP8 kvcache/softcap/slidingwindow

Dockerfile_intel

@@ -87,7 +87,7 @@ RUN echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https:/
 
 RUN mv /tmp/intel-for-pytorch-gpu-dev.list /etc/apt/sources.list.d
 
-RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt install -y xpu-smi cmake ninja-build pciutils intel-ocloc
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt install -y xpu-smi cmake ninja-build pciutils intel-ocloc libnl-genl-3-200
 
 # Text Generation Inference base env
 ENV HF_HOME=/data \
@@ -100,8 +100,6 @@ ENV HF_HOME=/data \
 WORKDIR /usr/src
 RUN pip install torch==2.6.0 --index-url https://download.pytorch.org/whl/test/xpu
 
-RUN pip install triton-xpu==3.2.0b1 --no-cache-dir
-
 # Install server
 COPY proto proto
 COPY server server
@@ -119,7 +117,9 @@ ENV CCL_TOPO_FABRIC_VERTEX_CONNECTION_CHECK=0
 ENV TORCH_DEVICE_BACKEND_AUTOLOAD=0
 
 RUN pip install https://intel-extension-for-pytorch.s3.amazonaws.com/ipex_stable/xpu/oneccl_bind_pt-2.6.0%2Bxpu-cp311-cp311-linux_x86_64.whl
-RUN pip install https://intel-extension-for-pytorch.s3.amazonaws.com/ipex_stable/xpu/intel_extension_for_pytorch-2.6.10%2Bxpu-cp311-cp311-linux_x86_64.whl
+RUN git clone https://github.com/intel/intel-extension-for-pytorch && cd intel-extension-for-pytorch && git checkout d5a7036316a01ea8220eb4da78a2207c423a1166
+RUN sed -i 's/VERSION_MINOR 7/VERSION_MINOR 6/' intel-extension-for-pytorch/version.txt
+RUN cd intel-extension-for-pytorch && git submodule update --init --recursive && USE_AOT_DEVLIST='pvc,ats-m150' BUILD_SEPARATE_OPS=OFF BUILD_WITH_CPU=OFF USE_XETLA=ON python setup.py install && rm -rf /usr/src/intel-extension-for-pytorch
 # Install benchmarker
 COPY --from=builder /usr/src/target/release-opt/text-generation-benchmark /usr/local/bin/text-generation-benchmark
 # Install router

server/text_generation_server/layers/attention/ipex.py

@@ -8,7 +8,10 @@
     BLOCK_SIZE,
 )
 
-SUPPORTS_WINDOWING = False
+if ATTENTION == "flashdecoding-ipex":
+    SUPPORTS_WINDOWING = True
+else:
+    SUPPORTS_WINDOWING = False
 
 
 def attention(
@@ -25,13 +28,19 @@ def attention(
     causal: bool = True,
     softcap: Optional[float] = None,
 ):
-    if softcap is not None:
-        raise NotImplementedError("softcap is not available in IPEX")
 
     out = torch.empty_like(query)
+    kv_cache_dtype = "auto"
+    if kv_cache.key.dtype == torch.float8_e5m2:
+        kv_cache_dtype = "fp8_e5m2"
+    if kv_cache.key.dtype == torch.float8_e4m3fn:
+        kv_cache_dtype = "fp8_e4m3"
 
     # We do not need to check window_size_left (not supported) here, so it is already checked ahead of time at model load.
     if ATTENTION == "flashdecoding-ipex":
+        window_size_right = -1 if window_size_left == -1 else 0
+        if softcap is None:
+            softcap = -1.0
         ipex.llm.modules.PagedAttention.flash_attn_varlen_func(
             out,
             query.contiguous() if query.device.type == "xpu" else query,
@@ -45,8 +54,18 @@ def attention(
             causal,
             block_tables,
             None,
+            window_size_left=window_size_left,
+            window_size_right=window_size_right,
+            kv_cache_dtype=kv_cache_dtype,
+            k_scale=kv_scales.key_scale_cpu,
+            v_scale=kv_scales.value_scale_cpu,
+            softcap=softcap,
         )
     else:
+        if softcap is not None:
+            raise NotImplementedError(
+                "softcap is not available in IPEX paged attention"
+            )
         ipex.llm.functional.varlen_attention(
             query.contiguous() if query.device.type == "xpu" else query,
             key.contiguous() if key.device.type == "xpu" else key,
@@ -80,12 +99,16 @@ def paged_attention(
     softcap: Optional[float] = None,
     window_size_left: Optional[int] = -1,
 ):
-    if softcap is not None:
-        raise NotImplementedError("softcap is not available in IPEX")
-
     out = torch.empty_like(query)
-
+    kv_cache_dtype = "auto"
+    if kv_cache.key.dtype == torch.float8_e5m2:
+        kv_cache_dtype = "fp8_e5m2"
+    if kv_cache.key.dtype == torch.float8_e4m3fn:
+        kv_cache_dtype = "fp8_e4m3"
     if ATTENTION == "flashdecoding-ipex":
+        window_size_right = -1 if window_size_left == -1 else 0
+        if softcap is None:
+            softcap = -1.0
         ipex.llm.modules.PagedAttention.flash_attn_varlen_func(
             out,
             query.contiguous() if query.device.type == "xpu" else query,
@@ -99,9 +122,19 @@ def paged_attention(
             True,
             block_tables,
             None,
+            window_size_left=window_size_left,
+            window_size_right=window_size_right,
+            kv_cache_dtype=kv_cache_dtype,
+            k_scale=kv_scales.key_scale_cpu,
+            v_scale=kv_scales.value_scale_cpu,
+            softcap=softcap,
         )
     else:
         input_lengths = seqlen.input_lengths + seqlen.cache_lengths
+        if softcap is not None:
+            raise NotImplementedError(
+                "softcap is not available in IPEX paged attention"
+            )
         ipex.llm.modules.PagedAttention.single_query_cached_kv_attention(
             out,
             query,
@@ -114,6 +147,8 @@ def paged_attention(
             BLOCK_SIZE,
             max_s,
             None,
+            k_scale=kv_scales.key_scale_cpu,
+            v_scale=kv_scales.value_scale_cpu,
         )
     return out
 

server/text_generation_server/layers/attention/kv_cache.py

@@ -68,15 +68,20 @@ def __init__(
         if dtype in {torch.float8_e5m2, torch.float8_e4m3fn}:
             if not (
                 (ATTENTION == "flashinfer" and SYSTEM == "cuda")
-                or (ATTENTION == "paged" and SYSTEM in ("cuda", "rocm"))
+                or (ATTENTION == "paged" and SYSTEM in ("cuda", "rocm", "ipex"))
+                or (ATTENTION == "flashdecoding-ipex")
             ):
                 raise ValueError(
-                    "FP8 KV cache is currently only supported for flashinfer on CUDA and paged attention on CUDA and ROCm. "
+                    "FP8 KV cache is currently only supported for flashinfer on CUDA and paged attention on CUDA, ROCm and INTEL IPEX and flashdecoding in Intel IPEX "
                 )
             if SYSTEM == "rocm" and dtype == torch.float8_e5m2:
                 raise ValueError(
                     "float8_e5m2 FP8 KV cache is not supported on AMD ROCm"
                 )
+            if device.type == "cpu" and dtype == torch.float8_e4m3fn:
+                raise ValueError(
+                    "float8_e4m3fn FP8 KV cache is not supported on Intel IPEX CPU"
+                )
 
         element_size = torch.tensor([], dtype=dtype).element_size()
         if SYSTEM == "ipex" and device.type == "xpu":
@@ -133,15 +138,16 @@ def can_scale(self, kv_scales: KVScales) -> bool:
             return False
         elif self.dtype == torch.float8_e4m3fn and (
             (ATTENTION in ("paged", "flashinfer") and SYSTEM == "cuda")
-            or (ATTENTION == "paged" and SYSTEM == "rocm")
+            or (ATTENTION == "paged" and SYSTEM in ["rocm", "ipex"])
+            or (ATTENTION == "flashdecoding-ipex")
         ):
             log_once(logger.info, "Using FP8 KV cache scales")
             return True
         else:
             # We have scales, but not the correct FP8 cache type, so warn once.
             log_once(
                 logger.info,
-                "Ignoring FP8 KV cache scales, supported only for float8_e4m3fn KV cache with flashinfer on CUDA and paged attention on ROCm",
+                "Ignoring FP8 KV cache scales, supported only for float8_e4m3fn KV cache with flashinfer on CUDA and paged attention on ROCm/IPEX and flashdecoding on IPEX",
             )
             return False
 
@@ -207,8 +213,20 @@ def store(
         elif ATTENTION == "flashdecoding-ipex" and key.device.type == "xpu":
             import intel_extension_for_pytorch as ipex
 
+            kv_cache_dtype = "auto"
+            if key_cache.dtype == torch.float8_e5m2:
+                kv_cache_dtype = "fp8_e5m2"
+            if key_cache.dtype == torch.float8_e4m3fn:
+                kv_cache_dtype = "fp8_e4m3"
             ipex.llm.modules.PagedAttention.reshape_and_cache_flash(
-                key, value, key_cache, value_cache, slots
+                key,
+                value,
+                key_cache,
+                value_cache,
+                slots,
+                kv_cache_dtype=kv_cache_dtype,
+                k_scale=kv_scales.key_scale_cpu,
+                v_scale=kv_scales.value_scale_cpu,
             )
         else:
             paged_reshape_and_cache(
@@ -267,8 +285,21 @@ def paged_reshape_and_cache(
     elif SYSTEM == "ipex":
         import intel_extension_for_pytorch as ipex
 
+        kv_cache_dtype = "auto"
+        if key_cache.dtype == torch.float8_e5m2:
+            kv_cache_dtype = "fp8_e5m2"
+        if key_cache.dtype == torch.float8_e4m3fn:
+            kv_cache_dtype = "fp8_e4m3"
+
         ipex.llm.modules.PagedAttention.reshape_and_cache(
-            key, value, key_cache, value_cache, slots
+            key,
+            value,
+            key_cache,
+            value_cache,
+            slots,
+            kv_cache_dtype=kv_cache_dtype,
+            k_scale=k_scale,
+            v_scale=v_scale,
         )
     else:
         raise NotImplementedError(