Training FST with constant lr, switch to cosine or linear please

FST/FontDiffuserFST_training_phase_1_config.yaml

@@ -11,8 +11,8 @@ character_input: false
 characters: null
 characters_file: null
 ckpt_dir: null
-ckpt_interval: 2
-compile: false
+ckpt_interval: 1500
+compile: true
 compute_fid: false
 consistency_loss_weight: 0.1
 content_character: null
@@ -47,7 +47,7 @@ fst_feature_channels: 64,128,256,512,1024
 fst_num_queries: 220
 fst_num_scales: 5
 fst_query_dim: 256
-gradient_accumulation_steps: 1
+gradient_accumulation_steps: 2
 ground_truth_dir: null
 guidance_scale: 7.5
 guidance_type: classifier-free
@@ -69,12 +69,12 @@ learning_rate: 0.0001
 local_rank: -1
 log_interval: 50
 logging_dir: logs
-lr_scheduler: cosine
+lr_scheduler: constant
 lr_warmup_steps: 200
 max_grad_norm: 1.0
-max_train_steps: 4
+max_train_steps: 3000
 method: multistep
-mixed_precision: 'no'
+mixed_precision: fp16
 mode: refinement
 model_type: noise
 mss_base_channels: 64
@@ -119,7 +119,7 @@ summary: false
 t_end: null
 t_start: null
 temperature: 0.07
-train_batch_size: 1
+train_batch_size: 4
 ttf_path: ttf/KaiXinSongA.ttf
 unet_channels: !!python/tuple
 - 64

FST/checkpoint_step_1500/content_encoder.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61099e3694480a1fff91aa4bb1bbc1dc966376666f5c2d17575e29ca1e6c0323
+size 4756580

FST/checkpoint_step_1500/fst_module.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36ea783bc764b673f1eaf4baa303f49056c01106bbaae63cbd639f11e189368e
+size 238575652

FST/checkpoint_step_1500/fst_projection.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d86f9cba4a75ada85db13770a254093e66008b4c5bbd02a60d3899d838aede9
+size 4198552

FST/checkpoint_step_1500/identity_loss_module.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ac907e1995c674cbe33c55038ccc3ee9ff880974a0e6b1f6115b29ed9bbce7b
+size 193688

FST/checkpoint_step_1500/mss_encoder.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25866998ebc841c69e7a321a58747c1f42e16bc3a2909aad674a56eb287eb926
+size 25261992

FST/checkpoint_step_1500/original_style_projection.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb635fc0ce7d94fa5f0f8f5535911c176911dd3072f8a4819032c3c8456bb2ee
+size 4198552

FST/checkpoint_step_1500/style_encoder.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:482252ceee1e2a054a2f4cfc19880b0326409c5b508290cc403d2ff0653a63fc
+size 82394556

FST/checkpoint_step_1500/training_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2034e2335188e51d7a331851cec591ffb2a37cae3566ec7f2a271650474ec32e
+size 1349161373

FST/checkpoint_step_1500/unet.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab3843aa9441306fffacb462bffbc01c5619970d34e2a95f315bf1d3bde4cb2f
+size 314927748

FST/logs/FontDiffuserFST_training_phase_1/events.out.tfevents.1770100017.944673a34ada.3199.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:601e4f6b4739fb8759be7667503cd503d4bc6b49e4c09bfd8d4458d6cd7e7bad
+size 88

FST/logs/FontDiffuserFST_training_phase_1/training_config_gradient_accumulation_steps/events.out.tfevents.1770100021.944673a34ada.3199.4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57128e0d238eecee51315dc276006a4db8d014a56683286014676e2dd23401b4
+size 139

FST/logs/FontDiffuserFST_training_phase_1/training_config_learning_rate/events.out.tfevents.1770100021.944673a34ada.3199.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f4fd06c25581a6cee44a65c75b0b812cd7769aa0c197d4fb7194ac9073e6770
+size 139

FST/logs/FontDiffuserFST_training_phase_1/training_config_max_train_steps/events.out.tfevents.1770100021.944673a34ada.3199.3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0276b1d81bf59f5e71043a6f6fb0afc0f1edfd62877ffd085a51808e08b1f429
+size 139

FST/logs/FontDiffuserFST_training_phase_1/training_config_mixed_precision/events.out.tfevents.1770100021.944673a34ada.3199.5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3365a50977540adc5cf17d15d4581de4c640e248758602dc1cb6f8b7c913d3a5
+size 88

FST/logs/FontDiffuserFST_training_phase_1/training_config_train_batch_size/events.out.tfevents.1770100021.944673a34ada.3199.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78393215a116d4b97a71028bc3337370d2fd6b625d78395940dc1da168156543
+size 139