DigiDecode
diff --git a/‎llama_cpp/llama.py
Copy file name to clipboardExpand all lines: llama_cpp/llama.py
+8-5Lines changed: 8 additions & 5 deletions b/‎llama_cpp/llama.py
Copy file name to clipboardExpand all lines: llama_cpp/llama.py
+8-5Lines changed: 8 additions & 5 deletions
diff --git a/‎llama_cpp/server/model.py
Copy file name to clipboardExpand all lines: llama_cpp/server/model.py
+3-1Lines changed: 3 additions & 1 deletion b/‎llama_cpp/server/model.py
Copy file name to clipboardExpand all lines: llama_cpp/server/model.py
+3-1Lines changed: 3 additions & 1 deletion
@@ -6,6 +6,7 @@
 import time
 import json
 import ctypes
+import typing
 import fnmatch
 import multiprocessing
 
@@ -249,24 +250,26 @@ def __init__(
                 self._kv_overrides_array[i].key = k.encode("utf-8")
                 if isinstance(v, bool):
                     self._kv_overrides_array[i].tag = llama_cpp.LLAMA_KV_OVERRIDE_TYPE_BOOL
-                    self._kv_overrides_array[i].value.bool_value = v
+                    self._kv_overrides_array[i].value.val_bool = v
                 elif isinstance(v, int):
                     self._kv_overrides_array[i].tag = llama_cpp.LLAMA_KV_OVERRIDE_TYPE_INT
-                    self._kv_overrides_array[i].value.int_value = v
+                    self._kv_overrides_array[i].value.val_i64 = v
                 elif isinstance(v, float):
                     self._kv_overrides_array[i].tag = llama_cpp.LLAMA_KV_OVERRIDE_TYPE_FLOAT
-                    self._kv_overrides_array[i].value.float_value = v
+                    self._kv_overrides_array[i].value.val_f64 = v
                 elif isinstance(v, str): # type: ignore
                     v_bytes = v.encode("utf-8")
                     if len(v_bytes) > 128: # TODO: Make this a constant
                         raise ValueError(f"Value for {k} is too long: {v}")
                     v_bytes = v_bytes.ljust(128, b"\0")
                     self._kv_overrides_array[i].tag = llama_cpp.LLAMA_KV_OVERRIDE_TYPE_STR
                     # copy min(v_bytes, 128) to str_value
+                    address = typing.cast(int, ctypes.addressof(self._kv_overrides_array[i].value) + llama_cpp.llama_model_kv_override_value.val_str.offset)
+                    buffer_start = ctypes.cast(address, ctypes.POINTER(ctypes.c_char))
                     ctypes.memmove(
-                        self._kv_overrides_array[i].value.str_value,
+                        buffer_start,
                         v_bytes,
-                        min(len(v_bytes), 128),
+                        128,
                     )
                 else:
                     raise ValueError(f"Unknown value type for {k}: {v}")
 
@@ -183,7 +183,7 @@ def load_llama_from_model_settings(settings: ModelSettings) -> llama_cpp.Llama:
                 num_pred_tokens=settings.draft_model_num_pred_tokens
             )
 
-        kv_overrides: Optional[Dict[str, Union[bool, int, float]]] = None
+        kv_overrides: Optional[Dict[str, Union[bool, int, float, str]]] = None
         if settings.kv_overrides is not None:
             assert isinstance(settings.kv_overrides, list)
             kv_overrides = {}
@@ -197,6 +197,8 @@ def load_llama_from_model_settings(settings: ModelSettings) -> llama_cpp.Llama:
                         kv_overrides[key] = int(value)
                     elif value_type == "float":
                         kv_overrides[key] = float(value)
+                    elif value_type == "str":
+                        kv_overrides[key] = value
                     else:
                         raise ValueError(f"Unknown value type {value_type}")