Skip to content

Support prequant qwen3 #10839

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
metascroy merged 3 commits into from
May 14, 2025
Merged

Support prequant qwen3 #10839

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
751eb3f
init
metascroy May 13, 2025
ea7cbdc
up
metascroy May 13, 2025
880c99b
up
metascroy May 14, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 17 additions & 8 deletions examples/models/qwen3/convert_weights.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@
_QWEN_3_FROM_META = {
"tok_embeddings.weight": "model.embed_tokens.weight",
"norm.weight": "model.norm.weight",
"output.weight": "lm_head.weight",
"layers.{}.attention.wk.weight": "model.layers.{}.self_attn.k_proj.weight",
"layers.{}.attention.k_norm_fn.weight": "model.layers.{}.self_attn.k_norm.weight",
"layers.{}.attention.wq.weight": "model.layers.{}.self_attn.q_proj.weight",
Expand Down Expand Up @@ -47,20 +48,19 @@ def qwen_3_tune_to_meta(state_dict: Dict[str, torch.Tensor]) -> Dict[str, torch.
inverted_mapping_dict = {v: k for k, v in _QWEN_3_FROM_META.items()}

for key, value in state_dict.items():
# Tied embeddings for 0.6b and 4b models.
if key == "lm_head.weight":
continue
new_key = get_mapped_key(key, inverted_mapping_dict)
converted_state_dict[new_key] = value

converted_state_dict["output.weight"] = converted_state_dict[
"tok_embeddings.weight"
]
# If lm_head.weight is not present in state dict, assume tied embeddings (e.g., 0.6b and 4b models)
if "lm_head.weight" not in state_dict:
Copy link
Contributor

@jackzhxng jackzhxng May 13, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lm_head is present in the hf checkpoints even if they are tied embeddings, it will just be the same weights as the tok_embeddings

Copy link
Contributor Author

@metascroy metascroy May 13, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmmm, are you sure it's there? I thought when config. tie_word_embeddings = true, it might not be there, but gets materialized during a tie_weights() command on the HF model.

In any case, if it is there, it's covered by the regular loop through keys and this logic is not executed. If it's not there, this sets lm_head's weight to the embeddings.

Copy link
Contributor

@jackzhxng jackzhxng May 13, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah it's here, https://huggingface.co/Qwen/Qwen3-0.6B/tree/main?show_file_info=model.safetensors. Also I remember seeing it while debugging the checkpoint. But sure, in that case could you reword the comment?

Copy link
Contributor Author

@metascroy metascroy May 14, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reworded the comment a little.

I'm not convinced by https://huggingface.co/Qwen/Qwen3-0.6B/tree/main?show_file_info=model.safetensors because it's just metadata, and doesn't prove anything about what is stored in the file.

It does not look like lm_head is present in the safetensors when I unpack them locally. Perhaps you looked at the checkpoint after running your script? (Which copied the embedding tensors into lm_head).

Copy link
Contributor

@jackzhxng jackzhxng May 14, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, if you double checked then that's fine!

converted_state_dict["output.weight"] = converted_state_dict[
"tok_embeddings.weight"
]

return converted_state_dict


def load_checkpoint(input_dir: str) -> Dict:
def load_checkpoint_from_safetensors(input_dir: str) -> Dict:
index_path = os.path.join(input_dir, "model.safetensors.index.json")
if os.path.exists(index_path):
# Sharded checkpoint.
Expand All @@ -86,6 +86,15 @@ def load_checkpoint(input_dir: str) -> Dict:
return state_dict


def load_checkpoint(input_dir: str) -> Dict:
pytorch_path = os.path.join(input_dir, "pytorch_model.bin")
if os.path.exists(pytorch_path):
print("Loading checkpoint from PyTorch .bin file")
return torch.load(pytorch_path, map_location="cpu", weights_only=True)
print("Loading checkpoint from safetensors directory")
return load_checkpoint_from_safetensors(input_dir)


def convert_weights(input_dir: str, output_file: str) -> None:
print("Loading checkpoint...")
sd = load_checkpoint(input_dir)
Expand All @@ -103,7 +112,7 @@ def main():
parser.add_argument(
"input_dir",
type=str,
help="Path to directory containing checkpoint files",
help="Path to directory containing safetensor checkpoint files, or PyTorch checkpoint file.",
)
parser.add_argument("output", type=str, help="Path to the output checkpoint")

Expand Down
Loading