From 3edafe95c59d8b81c0bdc69b5c39a5c29c6739ae Mon Sep 17 00:00:00 2001
From: seongwoo <mhs4670go@naver.com>
Date: Thu, 21 May 2026 10:07:16 +0900
Subject: [PATCH] [quantization] Add tied embedding bit-width validation

This commit adds tied embedding bit-width validation.

TICO-DCO-1.0-Signed-off-by: seongwoo <mhs4670go@naver.com>
---
 .../quantize_full_qmodel_with_gptq.py         | 83 +++++++++++++++++++
 1 file changed, 83 insertions(+)

diff --git a/tico/quantization/wrapq/examples/quantize_full_qmodel_with_gptq.py b/tico/quantization/wrapq/examples/quantize_full_qmodel_with_gptq.py
index c71f7680..80968a18 100644
--- a/tico/quantization/wrapq/examples/quantize_full_qmodel_with_gptq.py
+++ b/tico/quantization/wrapq/examples/quantize_full_qmodel_with_gptq.py
@@ -437,6 +437,88 @@ def parse_cle_pairs(raw_pairs: list[str] | None) -> list[tuple[str, str]]:
     return pairs
 
 
+def _weights_share_storage(
+    left: torch.Tensor,
+    right: torch.Tensor,
+) -> bool:
+    """Return True if two weight tensors share the exact same storage slice."""
+    if left is right:
+        return True
+
+    if not isinstance(left, torch.Tensor) or not isinstance(right, torch.Tensor):
+        return False
+
+    if left.device != right.device:
+        return False
+
+    if left.device.type == "meta" or right.device.type == "meta":
+        return False
+
+    if left.numel() == 0 or right.numel() == 0:
+        return False
+
+    return (
+        left.untyped_storage().data_ptr() == right.untyped_storage().data_ptr()
+        and left.storage_offset() == right.storage_offset()
+        and tuple(left.shape) == tuple(right.shape)
+        and tuple(left.stride()) == tuple(right.stride())
+    )
+
+
+def has_tied_input_output_embeddings(model: torch.nn.Module) -> bool:
+    """Return True if the input embedding and LM head weights are tied."""
+    get_input_embeddings = getattr(model, "get_input_embeddings", None)
+    get_output_embeddings = getattr(model, "get_output_embeddings", None)
+
+    if not callable(get_input_embeddings) or not callable(get_output_embeddings):
+        return False
+
+    input_embeddings = get_input_embeddings()
+    output_embeddings = get_output_embeddings()
+
+    if input_embeddings is None or output_embeddings is None:
+        return False
+
+    input_weight = getattr(input_embeddings, "weight", None)
+    output_weight = getattr(output_embeddings, "weight", None)
+
+    if input_weight is None or output_weight is None:
+        return False
+
+    return _weights_share_storage(input_weight, output_weight)
+
+
+def validate_tied_embedding_weight_bits(
+    model: torch.nn.Module,
+    args: argparse.Namespace,
+) -> None:
+    """
+    Reject different embedding and LM head bit-widths for tied weights.
+
+    Args:
+        model: Model whose input embedding and output projection are inspected.
+        args: Parsed command-line arguments.
+
+    Raises:
+        ValueError: If the model ties input embedding and LM head weights while
+            `--embedding_weight_bits` and `--lm_head_weight_bits` differ.
+    """
+    if args.embedding_weight_bits == args.lm_head_weight_bits:
+        return
+
+    if not has_tied_input_output_embeddings(model):
+        return
+
+    raise ValueError(
+        "Cannot use different bit-widths for tied input embedding and lm_head "
+        "weights: "
+        f"--embedding_weight_bits={args.embedding_weight_bits}, "
+        f"--lm_head_weight_bits={args.lm_head_weight_bits}. "
+        "Set both options to the same value or use a model with untied "
+        "input/output embeddings."
+    )
+
+
 def build_gptq_config(
     args,
     sensitivity: dict[str, torch.Tensor] | None = None,
@@ -1322,6 +1404,7 @@ def main():
     print_config(args, device)
 
     model, tokenizer = load_model_and_tokenizer(args, dtype)
+    validate_tied_embedding_weight_bits(model, args)
     configure_max_position_embeddings(model, args)
 
     dataset_test = load_eval_dataset(args)