[pull] master from ggml-org:master

ggml/src/ggml-sycl/ggml-sycl.cpp

@@ -372,9 +372,14 @@ static void ggml_backend_sycl_buffer_set_tensor(ggml_backend_buffer_t buffer,
     auto stream = &(dpct::dev_mgr::instance().get_device(ctx->device).default_queue());
     SYCL_CHECK(
         CHECK_TRY_ERROR(dpct::dev_mgr::instance().get_device(ctx->device).queues_wait_and_throw()));
+    // Note: Use host buffer to save the data from mmap(), then copy to device. It's workaround for mmap() issue on PVC GPU.
+    // This function will be called during load model from disk. Use memory buffer replace dynamic won't save more time and brings potential memory leak risk here.
+    char* host_buf = (char*)malloc(size);
+    memcpy(host_buf, data, size);
     SYCL_CHECK(
-        CHECK_TRY_ERROR((*stream).memcpy((char *)tensor->data + offset, data, size)
+        CHECK_TRY_ERROR((*stream).memcpy((char *)tensor->data + offset, host_buf, size)
                              .wait()));
+    free(host_buf);
 }
 catch (sycl::exception const &exc) {
   std::cerr << exc.what() << "Exception caught at file:" << __FILE__

gguf-py/gguf/lazy.py

@@ -139,6 +139,16 @@ def wrapped_fn(*args, **kwargs):
 
             if isinstance(res, cls._tensor_type):
                 return cls(meta=cls.eager_to_meta(res), args=args, kwargs=kwargs, func=fn)
+            elif isinstance(res, tuple) and all(isinstance(t, cls._tensor_type) for t in res):
+                # share the evaluation between lazy tuple elements
+                shared_args: list = [args, None]
+
+                def eager_tuple_element(a: list[Any], i: int = 0, /, **kw) -> LazyBase:
+                    assert len(a) == 2
+                    if a[1] is None:
+                        a[1] = fn(*a[0], **kw)
+                    return a[1][i]
+                return tuple(cls(meta=cls.eager_to_meta(res[i]), args=(shared_args, i), kwargs=kwargs, func=eager_tuple_element) for i in range(len(res)))
             else:
                 del res  # not needed
                 # non-tensor return likely relies on the contents of the args