lym00 commited on
Commit
02469b3
·
verified ·
1 Parent(s): 3ce174d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +25 -29
README.md CHANGED
@@ -38,7 +38,7 @@ Example: `python -m deepcompressor.app.diffusion.ptq examples/diffusion/configs/
38
 
39
  Folder Structure
40
 
41
- - refer [black-forest-labs/FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev)
42
 
43
  - refer [black-forest-labs/FLUX.1-Kontext-dev](https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev/tree/main)
44
 
@@ -51,11 +51,7 @@ potential fix: app.diffusion.pipeline.config.py
51
  ```python
52
  @staticmethod
53
  def _default_build(
54
- name: str,
55
- path: str,
56
- dtype: str | torch.dtype,
57
- device: str | torch.device,
58
- shift_activations: bool
59
  ) -> DiffusionPipeline:
60
  if not path:
61
  if name == "sdxl":
@@ -64,6 +60,8 @@ potential fix: app.diffusion.pipeline.config.py
64
  path = "stabilityai/sdxl-turbo"
65
  elif name == "pixart-sigma":
66
  path = "PixArt-alpha/PixArt-Sigma-XL-2-1024-MS"
 
 
67
  elif name == "flux.1-dev":
68
  path = "black-forest-labs/FLUX.1-dev"
69
  elif name == "flux.1-canny-dev":
@@ -76,24 +74,22 @@ potential fix: app.diffusion.pipeline.config.py
76
  path = "black-forest-labs/FLUX.1-schnell"
77
  else:
78
  raise ValueError(f"Path for {name} is not specified.")
79
-
80
- # Instantiate the pipeline
81
- if name in ["flux.1-canny-dev", "flux.1-depth-dev"]:
82
  pipeline = FluxControlPipeline.from_pretrained(path, torch_dtype=dtype)
83
  elif name == "flux.1-fill-dev":
84
  pipeline = FluxFillPipeline.from_pretrained(path, torch_dtype=dtype)
85
  elif name.startswith("sana-"):
86
  if dtype == torch.bfloat16:
87
- pipeline = SanaPipeline.from_pretrained(
88
- path, variant="bf16", torch_dtype=dtype, use_safetensors=True
89
- )
90
  pipeline.vae.to(dtype)
91
  pipeline.text_encoder.to(dtype)
92
  else:
93
  pipeline = SanaPipeline.from_pretrained(path, torch_dtype=dtype)
94
  else:
95
  pipeline = AutoPipelineForText2Image.from_pretrained(path, torch_dtype=dtype)
96
-
97
  # Debug output
98
  print(">>> DEVICE:", device)
99
  print(">>> PIPELINE TYPE:", type(pipeline))
@@ -128,14 +124,14 @@ potential fix: app.diffusion.pipeline.config.py
128
 
129
  Debug Log
130
  ```
131
- 25-07-21 22:47:02 | I | === Start Evaluating ===
132
- 25-07-21 22:47:02 | I | * Building diffusion model pipeline
133
- Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 15.44it/s]
134
- Loading pipeline components...: 57%|████████████████████████████████████████████████████████████████ | 4/7 [00:00<00:00, 9.47it/s]
135
  You set `add_prefix_space`. The tokenizer needs to be converted from the slow tokenizers
136
- Loading pipeline components...: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 7/7 [00:00<00:00, 7.79it/s]
 
137
  >>> DEVICE: cuda
138
- >>> PIPELINE TYPE: <class 'diffusers.pipelines.flux.pipeline_flux.FluxPipeline'>
139
  >>> Moving transformer to cuda using to_empty()
140
  >>> WARNING: transformer.to_empty(cuda) failed: Module.to_empty() takes 1 positional argument but 2 were given
141
  >>> Falling back to transformer.to(cuda)
@@ -146,16 +142,16 @@ Loading pipeline components...: 100%|██████████████
146
  >>> Moving text_encoder to cuda using to_empty()
147
  >>> WARNING: text_encoder.to_empty(cuda) failed: Module.to_empty() takes 1 positional argument but 2 were given
148
  >>> Falling back to text_encoder.to(cuda)
149
- 25-07-21 22:47:05 | I | Replacing fused Linear with ConcatLinear.
150
- 25-07-21 22:47:05 | I | + Replacing fused Linear in single_transformer_blocks.0 with ConcatLinear.
151
- 25-07-21 22:47:05 | I | - in_features = 3072/15360
152
- 25-07-21 22:47:05 | I | - out_features = 3072
153
- 25-07-21 22:47:05 | I | + Replacing fused Linear in single_transformer_blocks.1 with ConcatLinear.
154
- 25-07-21 22:47:05 | I | - in_features = 3072/15360
155
- 25-07-21 22:47:05 | I | - out_features = 3072
156
- 25-07-21 22:47:05 | I | + Replacing fused Linear in single_transformer_blocks.2 with ConcatLinear.
157
- 25-07-21 22:47:05 | I | - in_features = 3072/15360
158
- 25-07-21 22:47:05 | I | - out_features = 3072
159
  ```
160
 
161
  2) KeyError: <class 'diffusers.models.transformers.transformer_flux.FluxAttention'>
 
38
 
39
  Folder Structure
40
 
41
+ - refer [black-forest-labs/FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev/tree/main)
42
 
43
  - refer [black-forest-labs/FLUX.1-Kontext-dev](https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev/tree/main)
44
 
 
51
  ```python
52
  @staticmethod
53
  def _default_build(
54
+ name: str, path: str, dtype: str | torch.dtype, device: str | torch.device, shift_activations: bool
 
 
 
 
55
  ) -> DiffusionPipeline:
56
  if not path:
57
  if name == "sdxl":
 
60
  path = "stabilityai/sdxl-turbo"
61
  elif name == "pixart-sigma":
62
  path = "PixArt-alpha/PixArt-Sigma-XL-2-1024-MS"
63
+ elif name == "flux.1-kontext-dev":
64
+ path = "black-forest-labs/FLUX.1-Kontext-dev"
65
  elif name == "flux.1-dev":
66
  path = "black-forest-labs/FLUX.1-dev"
67
  elif name == "flux.1-canny-dev":
 
74
  path = "black-forest-labs/FLUX.1-schnell"
75
  else:
76
  raise ValueError(f"Path for {name} is not specified.")
77
+ if name in ["flux.1-kontext-dev"]:
78
+ pipeline = FluxKontextPipeline.from_pretrained(path, torch_dtype=dtype)
79
+ elif name in ["flux.1-canny-dev", "flux.1-depth-dev"]:
80
  pipeline = FluxControlPipeline.from_pretrained(path, torch_dtype=dtype)
81
  elif name == "flux.1-fill-dev":
82
  pipeline = FluxFillPipeline.from_pretrained(path, torch_dtype=dtype)
83
  elif name.startswith("sana-"):
84
  if dtype == torch.bfloat16:
85
+ pipeline = SanaPipeline.from_pretrained(path, variant="bf16", torch_dtype=dtype, use_safetensors=True)
 
 
86
  pipeline.vae.to(dtype)
87
  pipeline.text_encoder.to(dtype)
88
  else:
89
  pipeline = SanaPipeline.from_pretrained(path, torch_dtype=dtype)
90
  else:
91
  pipeline = AutoPipelineForText2Image.from_pretrained(path, torch_dtype=dtype)
92
+
93
  # Debug output
94
  print(">>> DEVICE:", device)
95
  print(">>> PIPELINE TYPE:", type(pipeline))
 
124
 
125
  Debug Log
126
  ```
127
+ 25-07-22 20:11:56 | I | === Start Evaluating ===
128
+ 25-07-22 20:11:56 | I | * Building diffusion model pipeline
129
+ Loading pipeline components...: 0%| | 0/7 [00:00<?, ?it/s]
 
130
  You set `add_prefix_space`. The tokenizer needs to be converted from the slow tokenizers
131
+ Loading checkpoint shards: 100%|██████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 18.92it/s]
132
+ Loading pipeline components...: 100%|█████████████████████████████████████████████████████| 7/7 [00:00<00:00, 9.50it/s]
133
  >>> DEVICE: cuda
134
+ >>> PIPELINE TYPE: <class 'diffusers.pipelines.flux.pipeline_flux_kontext.FluxKontextPipeline'>
135
  >>> Moving transformer to cuda using to_empty()
136
  >>> WARNING: transformer.to_empty(cuda) failed: Module.to_empty() takes 1 positional argument but 2 were given
137
  >>> Falling back to transformer.to(cuda)
 
142
  >>> Moving text_encoder to cuda using to_empty()
143
  >>> WARNING: text_encoder.to_empty(cuda) failed: Module.to_empty() takes 1 positional argument but 2 were given
144
  >>> Falling back to text_encoder.to(cuda)
145
+ 25-07-22 20:11:59 | I | Replacing fused Linear with ConcatLinear.
146
+ 25-07-22 20:11:59 | I | + Replacing fused Linear in single_transformer_blocks.0 with ConcatLinear.
147
+ 25-07-22 20:11:59 | I | - in_features = 3072/15360
148
+ 25-07-22 20:11:59 | I | - out_features = 3072
149
+ 25-07-22 20:11:59 | I | + Replacing fused Linear in single_transformer_blocks.1 with ConcatLinear.
150
+ 25-07-22 20:11:59 | I | - in_features = 3072/15360
151
+ 25-07-22 20:11:59 | I | - out_features = 3072
152
+ 25-07-22 20:11:59 | I | + Replacing fused Linear in single_transformer_blocks.2 with ConcatLinear.
153
+ 25-07-22 20:11:59 | I | - in_features = 3072/15360
154
+ 25-07-22 20:11:59 | I | - out_features = 3072
155
  ```
156
 
157
  2) KeyError: <class 'diffusers.models.transformers.transformer_flux.FluxAttention'>