frozenleaves/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-10-20 17:13:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

[detection] fix correct k_proj weight and bias slicing in D-FINE (#40257)

Browse Source 
Fix: correct k_proj weight and bias conversion in D-FINE
This commit is contained in:
nnul
2025-08-19 10:44:37 +01:00
committed by GitHub
parent 2f1a8ad4ba
commit c93594e286
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/transformers/models/d_fine/convert_d_fine_original_pytorch_checkpoint_to_hf.py
View File

@ -364,8 +364,8 @@ def read_in_q_k_v(state_dict, config, model_name):
if model_name in ["dfine_n_coco", "dfine_n_obj2coco_e25", "dfine_n_obj365"]:
state_dict[f"model.decoder.layers.{i}.self_attn.q_proj.weight"] = in_proj_weight[:128, :]
state_dict[f"model.decoder.layers.{i}.self_attn.q_proj.bias"] = in_proj_bias[:128]
state_dict[f"model.decoder.layers.{i}.self_attn.k_proj.weight"] = in_proj_weight[256:384, :]
state_dict[f"model.decoder.layers.{i}.self_attn.k_proj.bias"] = in_proj_bias[256:384]
state_dict[f"model.decoder.layers.{i}.self_attn.k_proj.weight"] = in_proj_weight[128:256, :]
state_dict[f"model.decoder.layers.{i}.self_attn.k_proj.bias"] = in_proj_bias[128:256]
state_dict[f"model.decoder.layers.{i}.self_attn.v_proj.weight"] = in_proj_weight[-128:, :]
state_dict[f"model.decoder.layers.{i}.self_attn.v_proj.bias"] = in_proj_bias[-128:]
else: