Hardcode VRAM size for AWS L4 (#92)

The way AWS reports the VRAM size for L4 is
inconsistent with other GPUs. The reported size is
24 base-10 gigabytes, while L4 is actually 24
base-2 gibibytes, same as many other NVIDIA GPUs.
Additionally, for multi-unit L4 instances AWS
reports the total VRAM size instead of the size of
a single unit.

This commit hardcodes the correct L4 VRAM size
until there are changes from AWS' end.

src/gpuhunt/providers/aws.py

@@ -138,7 +138,7 @@ def fill_gpu_details(self, offers: List[RawCatalogItem]):
                         gpu = i["GpuInfo"]["Gpus"][0]
                         gpus[i["InstanceType"]] = (
                             gpu["Name"],
-                            gpu["MemoryInfo"]["SizeInMiB"] / 1024,
+                            _get_gpu_memory_gib(gpu["Name"], gpu["MemoryInfo"]["SizeInMiB"]),
                         )
 
             regions = {
@@ -230,6 +230,22 @@ def filter(cls, offers: List[RawCatalogItem]) -> List[RawCatalogItem]:
         ]
 
 
+def _get_gpu_memory_gib(gpu_name: str, reported_memory_mib: int) -> float:
+    """
+    Fixes L4 memory size misreported by AWS API
+    """
+
+    if gpu_name != "L4":
+        return reported_memory_mib / 1024
+
+    if reported_memory_mib not in (22888, 91553, 183105):
+        logger.warning(
+            "The L4 memory size reported by AWS changed. "
+            "Please check that it is now correct and remove the hardcoded size if it is."
+        )
+    return 24
+
+
 def parse_memory(s: str) -> float:
     r = re.match(r"^([0-9.]+) GiB$", s)
     return float(r.group(1))