drop _tied_weights_keys

Signed-off-by: Mayank Mishra <mayank31398@gmail.com>

Author: mayank31398

lm_engine/hf_models/config/__init__.py

@@ -36,30 +36,6 @@ def _run(self, *args, **kwargs):
     return _holded_function
 
 
-# for erroring out on legacy configs
-_NAKED_DISALLOWED_ARGS = [
-    "activation_function",
-    "attn_pdrop",
-    "embd_pdrop",
-    "resid_pdrop",
-    "intermediate_size",
-    "shared_intermediate_size",
-    "num_experts",
-    "num_experts_per_tok",
-    "add_bias",
-    "attention_blocks",
-    "num_key_value_heads",
-    "attention_head_type",
-    "attention_multiplier",
-    "n_embd",
-    "n_head",
-    "n_inner",
-    "n_layer",
-    "n_positions",
-    "scale_attn_weights",
-    "num_attention_heads",
-]
-
 _SEQUENCE_MIXER_CONFIG_CLASSES = {
     "causal_convolution": _CausalConvolution,
     "gru": _GRUArgs,
@@ -148,9 +124,6 @@ def __init__(
 
         self.router_aux_loss_coef = router_aux_loss_coef
 
-        for i in _NAKED_DISALLOWED_ARGS:
-            assert i not in kwargs, f"found naked argument ({i})"
-
         super().__init__(
             bos_token_id=bos_token_id,
             eos_token_id=eos_token_id,