Skip to content

Fix bug with state loading on lora_decoder.ipynb #104

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix bug with state loading on lora_decoder.ipynb #104

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
52 changes: 32 additions & 20 deletions notebooks/lora_decoder.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3127,7 +3127,7 @@
"id": "Q-Sa97ezzBWo",
"outputId": "63a2f6ef-6399-4bc6-8be3-9aecda18a652"
},
"execution_count": 1,
"execution_count": null,
"outputs": [
{
"output_type": "stream",
Expand Down Expand Up @@ -3261,7 +3261,7 @@
"metadata": {
"id": "h196ZQ66kOah"
},
"execution_count": 2,
"execution_count": null,
"outputs": []
},
{
Expand All @@ -3276,7 +3276,7 @@
"metadata": {
"id": "qMBAU2yOkttM"
},
"execution_count": 3,
"execution_count": null,
"outputs": []
},
{
Expand All @@ -3292,7 +3292,7 @@
"id": "RJw0fSdF1U0j",
"outputId": "1874d2ff-7694-4ada-9ae9-5254150fe368"
},
"execution_count": 4,
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
Expand Down Expand Up @@ -3583,7 +3583,7 @@
"metadata": {
"id": "_o3t6oXKy8-s"
},
"execution_count": 5,
"execution_count": null,
"outputs": []
},
{
Expand All @@ -3602,7 +3602,7 @@
"metadata": {
"id": "sLjwl1RtyzHO"
},
"execution_count": 6,
"execution_count": null,
"outputs": []
},
{
Expand Down Expand Up @@ -3636,7 +3636,7 @@
"metadata": {
"id": "HMPOrc1Y47Y7"
},
"execution_count": 7,
"execution_count": null,
"outputs": []
},
{
Expand All @@ -3651,7 +3651,7 @@
"id": "jk_8ykSi56Bh",
"outputId": "62d490a7-476c-4f21-d2eb-0b94d1a683d5"
},
"execution_count": 8,
"execution_count": null,
"outputs": [
{
"output_type": "stream",
Expand Down Expand Up @@ -3694,7 +3694,7 @@
"metadata": {
"id": "Ca-K7XRMO_wC"
},
"execution_count": 9,
"execution_count": null,
"outputs": []
},
{
Expand All @@ -3709,7 +3709,7 @@
"id": "VGaiksGgPFGP",
"outputId": "f62ceac1-27af-4df1-fcf5-5d4fbe28c268"
},
"execution_count": 10,
"execution_count": null,
"outputs": [
{
"output_type": "stream",
Expand Down Expand Up @@ -3748,7 +3748,7 @@
"metadata": {
"id": "Ig17twUvQ8cr"
},
"execution_count": 11,
"execution_count": null,
"outputs": []
},
{
Expand Down Expand Up @@ -3837,7 +3837,7 @@
"id": "Qap8MnWtO9FS",
"outputId": "89a2ff09-3170-404d-c01d-a3b4b4877e37"
},
"execution_count": 12,
"execution_count": null,
"outputs": [
{
"output_type": "display_data",
Expand Down Expand Up @@ -3941,7 +3941,20 @@
},
{
"cell_type": "code",
"source": [],
"source": [
"from safetensors import safe_open\n",
"\n",
"# Loading the weights from the `safetensors` format to a plain dictionary.\n",
"decoder_lora_state_dict = {}\n",
"with safe_open(\"/content/decoder_lora_saves/checkpoint-500/model.safetensors\", framework=\"pt\", device=\"cpu\") as f:\n",
" for key in f.keys():\n",
" decoder_lora_state_dict[key] = f.get_tensor(key)\n",
"\n",
"decoder_prior_state_dict = {}\n",
"with safe_open(\"/content/decoder_prior_saves/checkpoint-500/model.safetensors\", framework=\"pt\", device=\"cpu\") as f:\n",
" for key in f.keys():\n",
" decoder_prior_state_dict[key] = f.get_tensor(key)"
],
"metadata": {
"id": "xEZNDvvintn_"
},
Expand All @@ -3953,7 +3966,6 @@
"source": [
"from diffusers.models.attention_processor import LoRAAttnProcessor, LoRAAttnAddedKVProcessor\n",
"lora_attn_procs = {}\n",
"d = torch.load('/content/decoder_lora_saves/checkpoint-500/pytorch_model.bin')\n",
"for name in decoder.unet.attn_processors.keys():\n",
" cross_attention_dim = None if name.endswith(\"attn1.processor\") else decoder.unet.config.cross_attention_dim\n",
" if name.startswith(\"mid_block\"):\n",
Expand All @@ -3971,13 +3983,13 @@
" ).to('cuda')\n",
"\n",
"decoder.unet.set_attn_processor(lora_attn_procs)\n",
"decoder.unet.load_state_dict(d, strict=False)\n",
"decoder.unet.load_state_dict(decoder_lora_state_dict, strict=False)\n",
"None"
],
"metadata": {
"id": "cQaJGtebasLy"
},
"execution_count": 16,
"execution_count": null,
"outputs": []
},
{
Expand All @@ -3988,13 +4000,13 @@
"for name in prior.prior.attn_processors.keys():\n",
" lora_attn_procs[name] = LoRAAttnProcessor(hidden_size=2048).to('cuda')\n",
"prior.prior.set_attn_processor(lora_attn_procs)\n",
"prior.prior.load_state_dict(torch.load('/content/decoder_prior_saves/checkpoint-500/pytorch_model.bin'), strict=False)\n",
"prior.prior.load_state_dict(decoder_prior_state_dict, strict=False)\n",
"None"
],
"metadata": {
"id": "5lfV9j-9atJt"
},
"execution_count": 17,
"execution_count": null,
"outputs": []
},
{
Expand Down Expand Up @@ -4049,7 +4061,7 @@
"id": "gqnaHUCZnq3Z",
"outputId": "fb69eb33-feb1-440d-91e6-a194d0f9d974"
},
"execution_count": 20,
"execution_count": null,
"outputs": [
{
"output_type": "display_data",
Expand Down Expand Up @@ -4115,7 +4127,7 @@
"id": "AMuPm5bma-Fq",
"outputId": "5a6b959a-f742-44bc-b31a-056f9b6d9c7c"
},
"execution_count": 21,
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
Expand Down