Removing IPEX_AVAIL. (#2115)

* Removing IPEX_AVAIL.

Chose to unify CPU and XPU under `ipex`. Most code is exactly similar
except for a very few spots.

The biggest number of spots is the kv-cache layout and the flash_xxx.py
files.
Since those files should be removed soon and factored away, we should
not need them.

* Forgot a few places.

* Unrelated change.

* Fixing HF_TOKEN.

* HF_TOKEN

server/text_generation_server/utils/dist.py

@@ -3,7 +3,7 @@ import torch
 
 from datetime import timedelta
 from loguru import logger
-from text_generation_server.utils.import_utils import IPEX_AVAIL
+from text_generation_server.utils.import_utils import SYSTEM
 
 # Tensor Parallelism settings
 RANK = int(os.getenv("RANK", "0"))
@@ -69,7 +69,7 @@ def initialize_torch_distributed():
 
         if not torch.distributed.is_initialized():
             # Call the init process.
-            if IPEX_AVAIL:
+            if SYSTEM == "ipex":
                 import intel_extension_for_pytorch as ipex
 
                 ipex.distributed.init_process_group(

server/text_generation_server/utils/import_utils.py

@@ -37,7 +37,10 @@ def get_cpu_free_memory(device, memory_fraction):
     return free_memory
 
 
-IPEX_AVAIL = is_ipex_available()
+def noop(*args, **kwargs):
+    pass
+
+
 SYSTEM = None
 if torch.version.hip is not None:
     SYSTEM = "rocm"
@@ -49,17 +52,19 @@ elif torch.version.cuda is not None and torch.cuda.is_available():
     empty_cache = torch.cuda.empty_cache
     synchronize = torch.cuda.synchronize
     get_free_memory = get_cuda_free_memory
-elif IPEX_AVAIL and hasattr(torch, "xpu") and torch.xpu.is_available():
-    SYSTEM = "xpu"
-    empty_cache = torch.xpu.empty_cache
-    synchronize = torch.xpu.synchronize
-    get_free_memory = get_xpu_free_memory
+elif is_ipex_available():
+    SYSTEM = "ipex"
+    if hasattr(torch, "xpu") and torch.xpu.is_available():
+        empty_cache = torch.xpu.empty_cache
+        synchronize = torch.xpu.synchronize
+        get_free_memory = get_xpu_free_memory
+    else:
+        empty_cache = noop
+        synchronize = noop
+        get_free_memory = get_cpu_free_memory
 else:
     SYSTEM = "cpu"
 
-    def noop(*args, **kwargs):
-        pass
-
     empty_cache = noop
     synchronize = noop
     get_free_memory = get_cpu_free_memory