faceshq_transformer.yaml 1.4 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061
  1. model:
  2. base_learning_rate: 4.5e-06
  3. target: taming.models.cond_transformer.Net2NetTransformer
  4. params:
  5. cond_stage_key: coord
  6. transformer_config:
  7. target: taming.modules.transformer.mingpt.GPT
  8. params:
  9. vocab_size: 1024
  10. block_size: 512
  11. n_layer: 24
  12. n_head: 16
  13. n_embd: 1024
  14. first_stage_config:
  15. target: taming.models.vqgan.VQModel
  16. params:
  17. ckpt_path: logs/2020-11-09T13-33-36_faceshq_vqgan/checkpoints/last.ckpt
  18. embed_dim: 256
  19. n_embed: 1024
  20. ddconfig:
  21. double_z: false
  22. z_channels: 256
  23. resolution: 256
  24. in_channels: 3
  25. out_ch: 3
  26. ch: 128
  27. ch_mult:
  28. - 1
  29. - 1
  30. - 2
  31. - 2
  32. - 4
  33. num_res_blocks: 2
  34. attn_resolutions:
  35. - 16
  36. dropout: 0.0
  37. lossconfig:
  38. target: taming.modules.losses.DummyLoss
  39. cond_stage_config:
  40. target: taming.modules.misc.coord.CoordStage
  41. params:
  42. n_embed: 1024
  43. down_factor: 16
  44. data:
  45. target: main.DataModuleFromConfig
  46. params:
  47. batch_size: 2
  48. num_workers: 8
  49. train:
  50. target: taming.data.faceshq.FacesHQTrain
  51. params:
  52. size: 256
  53. crop_size: 256
  54. coord: True
  55. validation:
  56. target: taming.data.faceshq.FacesHQValidation
  57. params:
  58. size: 256
  59. crop_size: 256
  60. coord: True