Improve lora implementation
Browse files- flux_pipeline.py +15 -4
- lora_loading.py +222 -81
flux_pipeline.py
CHANGED
@@ -2,7 +2,7 @@ import io
|
|
2 |
import math
|
3 |
import random
|
4 |
import warnings
|
5 |
-
from typing import TYPE_CHECKING, Callable, List
|
6 |
|
7 |
import numpy as np
|
8 |
from PIL import Image
|
@@ -148,7 +148,9 @@ class FluxPipeline:
|
|
148 |
random.seed(seed)
|
149 |
return cuda_generator, seed
|
150 |
|
151 |
-
def load_lora(
|
|
|
|
|
152 |
"""
|
153 |
Loads a LoRA checkpoint into the Flux flow transformer.
|
154 |
|
@@ -156,11 +158,20 @@ class FluxPipeline:
|
|
156 |
or loras which contain keys which start with lora_unet_[...].
|
157 |
|
158 |
Args:
|
159 |
-
lora_path (str): Path to the LoRA checkpoint.
|
160 |
scale (float): Scaling factor for the LoRA weights.
|
161 |
|
162 |
"""
|
163 |
-
self.model
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
164 |
|
165 |
@torch.inference_mode()
|
166 |
def compile(self):
|
|
|
2 |
import math
|
3 |
import random
|
4 |
import warnings
|
5 |
+
from typing import TYPE_CHECKING, Callable, List, OrderedDict, Union
|
6 |
|
7 |
import numpy as np
|
8 |
from PIL import Image
|
|
|
148 |
random.seed(seed)
|
149 |
return cuda_generator, seed
|
150 |
|
151 |
+
def load_lora(
|
152 |
+
self, lora_path: Union[str, OrderedDict[str, torch.Tensor]], scale: float
|
153 |
+
):
|
154 |
"""
|
155 |
Loads a LoRA checkpoint into the Flux flow transformer.
|
156 |
|
|
|
158 |
or loras which contain keys which start with lora_unet_[...].
|
159 |
|
160 |
Args:
|
161 |
+
lora_path (str | OrderedDict[str, torch.Tensor]): Path to the LoRA checkpoint or an ordered dictionary containing the LoRA weights.
|
162 |
scale (float): Scaling factor for the LoRA weights.
|
163 |
|
164 |
"""
|
165 |
+
self.model.load_lora(lora_path, scale)
|
166 |
+
|
167 |
+
def unload_lora(self, path_or_identifier: str):
|
168 |
+
"""
|
169 |
+
Unloads the LoRA checkpoint from the Flux flow transformer.
|
170 |
+
|
171 |
+
Args:
|
172 |
+
path_or_identifier (str): Path to the LoRA checkpoint or the name given to the LoRA checkpoint when it was loaded.
|
173 |
+
"""
|
174 |
+
self.model.unload_lora(path_or_identifier)
|
175 |
|
176 |
@torch.inference_mode()
|
177 |
def compile(self):
|
lora_loading.py
CHANGED
@@ -13,7 +13,7 @@ except Exception as e:
|
|
13 |
from float8_quantize import F8Linear
|
14 |
from modules.flux_model import Flux
|
15 |
|
16 |
-
path_regex = re.compile(r"
|
17 |
|
18 |
StateDict: TypeAlias = OrderedDict[str, torch.Tensor]
|
19 |
|
@@ -138,59 +138,126 @@ def convert_diffusers_to_flux_transformer_checkpoint(
|
|
138 |
f"double_blocks.{i}.txt_mod.lin.weight",
|
139 |
)
|
140 |
|
141 |
-
|
142 |
-
|
143 |
-
)
|
144 |
-
sample_q_B = diffusers_state_dict.pop(
|
145 |
-
f"{prefix}{block_prefix}attn.to_q.lora_B.weight"
|
146 |
-
)
|
147 |
-
|
148 |
-
sample_k_A = diffusers_state_dict.pop(
|
149 |
-
f"{prefix}{block_prefix}attn.to_k.lora_A.weight"
|
150 |
-
)
|
151 |
-
sample_k_B = diffusers_state_dict.pop(
|
152 |
-
f"{prefix}{block_prefix}attn.to_k.lora_B.weight"
|
153 |
-
)
|
154 |
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
|
|
|
|
168 |
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
194 |
|
195 |
# qk_norm
|
196 |
original_state_dict, diffusers_state_dict = convert_if_lora_exists(
|
@@ -265,32 +332,73 @@ def convert_diffusers_to_flux_transformer_checkpoint(
|
|
265 |
for i in range(num_single_layers):
|
266 |
block_prefix = f"single_transformer_blocks.{i}."
|
267 |
# norm.linear -> single_blocks.0.modulation.lin
|
|
|
268 |
original_state_dict, diffusers_state_dict = convert_if_lora_exists(
|
269 |
original_state_dict,
|
270 |
diffusers_state_dict,
|
271 |
-
|
272 |
f"single_blocks.{i}.modulation.lin.weight",
|
273 |
)
|
274 |
|
|
|
|
|
|
|
275 |
# Q, K, V, mlp
|
276 |
q_A = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_q.lora_A.weight")
|
277 |
q_B = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_q.lora_B.weight")
|
|
|
|
|
|
|
|
|
278 |
k_A = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_k.lora_A.weight")
|
279 |
k_B = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_k.lora_B.weight")
|
|
|
|
|
|
|
|
|
280 |
v_A = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_v.lora_A.weight")
|
281 |
v_B = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_v.lora_B.weight")
|
|
|
|
|
|
|
|
|
282 |
mlp_A = diffusers_state_dict.pop(
|
283 |
f"{prefix}{block_prefix}proj_mlp.lora_A.weight"
|
284 |
)
|
285 |
mlp_B = diffusers_state_dict.pop(
|
286 |
f"{prefix}{block_prefix}proj_mlp.lora_B.weight"
|
287 |
)
|
288 |
-
|
289 |
-
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
294 |
|
295 |
# output projections
|
296 |
original_state_dict, diffusers_state_dict = convert_if_lora_exists(
|
@@ -324,9 +432,16 @@ def convert_diffusers_to_flux_transformer_checkpoint(
|
|
324 |
return original_state_dict
|
325 |
|
326 |
|
327 |
-
def convert_from_original_flux_checkpoint(
|
328 |
-
|
329 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
330 |
sd = {
|
331 |
k.replace("lora_unet_", "")
|
332 |
.replace("double_blocks_", "double_blocks.")
|
@@ -358,14 +473,39 @@ def get_module_for_key(
|
|
358 |
return module
|
359 |
|
360 |
|
361 |
-
def get_lora_for_key(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
362 |
prefix = key.split(".lora")[0]
|
363 |
-
lora_A = lora_weights
|
364 |
-
lora_B = lora_weights
|
365 |
-
alpha = lora_weights.get(f"{prefix}.alpha"
|
|
|
|
|
|
|
366 |
return lora_A, lora_B, alpha
|
367 |
|
368 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
369 |
def calculate_lora_weight(
|
370 |
lora_weights: Tuple[torch.Tensor, torch.Tensor, Union[torch.Tensor, float]],
|
371 |
rank: Optional[int] = None,
|
@@ -389,12 +529,16 @@ def calculate_lora_weight(
|
|
389 |
w_down = lora_B.to(dtype=dtype, device=device)
|
390 |
|
391 |
if alpha != rank:
|
392 |
-
w_up = w_up *
|
393 |
-
|
394 |
if uneven_rank:
|
395 |
-
|
396 |
-
|
397 |
-
|
|
|
|
|
|
|
|
|
|
|
398 |
else:
|
399 |
fused_lora = lora_scale * torch.mm(w_down, w_up)
|
400 |
return fused_lora
|
@@ -445,16 +589,6 @@ def resolve_lora_state_dict(lora_weights, has_guidance: bool = True):
|
|
445 |
lora_weights = convert_from_original_flux_checkpoint(lora_weights)
|
446 |
logger.info("LoRA weights loaded")
|
447 |
logger.debug("Extracting keys")
|
448 |
-
keys_without_ab = [
|
449 |
-
key.replace(".lora_A.weight", "")
|
450 |
-
.replace(".lora_B.weight", "")
|
451 |
-
.replace(".lora_A", "")
|
452 |
-
.replace(".lora_B", "")
|
453 |
-
.replace(".alpha", "")
|
454 |
-
for key in lora_weights.keys()
|
455 |
-
]
|
456 |
-
logger.debug("Keys extracted")
|
457 |
-
keys_without_ab = list(set(keys_without_ab))
|
458 |
keys_without_ab = list(
|
459 |
set(
|
460 |
[
|
@@ -463,10 +597,11 @@ def resolve_lora_state_dict(lora_weights, has_guidance: bool = True):
|
|
463 |
.replace(".lora_A", "")
|
464 |
.replace(".lora_B", "")
|
465 |
.replace(".alpha", "")
|
466 |
-
for key in
|
467 |
]
|
468 |
)
|
469 |
)
|
|
|
470 |
return keys_without_ab, lora_weights
|
471 |
|
472 |
|
@@ -513,6 +648,9 @@ def apply_lora_to_model(
|
|
513 |
module = get_module_for_key(key, model)
|
514 |
weight, is_f8, dtype = extract_weight_from_linear(module)
|
515 |
lora_sd = get_lora_for_key(key, lora_weights)
|
|
|
|
|
|
|
516 |
weight = apply_lora_weight_to_module(weight, lora_sd, lora_scale=lora_scale)
|
517 |
if is_f8:
|
518 |
module.set_weight_tensor(weight.type(dtype))
|
@@ -540,6 +678,9 @@ def remove_lora_from_module(
|
|
540 |
module = get_module_for_key(key, model)
|
541 |
weight, is_f8, dtype = extract_weight_from_linear(module)
|
542 |
lora_sd = get_lora_for_key(key, lora_weights)
|
|
|
|
|
|
|
543 |
weight = unfuse_lora_weight_from_module(weight, lora_sd, lora_scale=lora_scale)
|
544 |
if is_f8:
|
545 |
module.set_weight_tensor(weight.type(dtype))
|
|
|
13 |
from float8_quantize import F8Linear
|
14 |
from modules.flux_model import Flux
|
15 |
|
16 |
+
path_regex = re.compile(r"/|\\")
|
17 |
|
18 |
StateDict: TypeAlias = OrderedDict[str, torch.Tensor]
|
19 |
|
|
|
138 |
f"double_blocks.{i}.txt_mod.lin.weight",
|
139 |
)
|
140 |
|
141 |
+
# Q, K, V
|
142 |
+
temp_dict = {}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
143 |
|
144 |
+
expected_shape_qkv_a = None
|
145 |
+
expected_shape_qkv_b = None
|
146 |
+
expected_shape_add_qkv_a = None
|
147 |
+
expected_shape_add_qkv_b = None
|
148 |
+
dtype = None
|
149 |
+
device = None
|
150 |
|
151 |
+
for component in [
|
152 |
+
"to_q",
|
153 |
+
"to_k",
|
154 |
+
"to_v",
|
155 |
+
"add_q_proj",
|
156 |
+
"add_k_proj",
|
157 |
+
"add_v_proj",
|
158 |
+
]:
|
159 |
|
160 |
+
sample_component_A_key = (
|
161 |
+
f"{prefix}{block_prefix}attn.{component}.lora_A.weight"
|
162 |
+
)
|
163 |
+
sample_component_B_key = (
|
164 |
+
f"{prefix}{block_prefix}attn.{component}.lora_B.weight"
|
165 |
+
)
|
166 |
+
if (
|
167 |
+
sample_component_A_key in diffusers_state_dict
|
168 |
+
and sample_component_B_key in diffusers_state_dict
|
169 |
+
):
|
170 |
+
sample_component_A = diffusers_state_dict.pop(sample_component_A_key)
|
171 |
+
sample_component_B = diffusers_state_dict.pop(sample_component_B_key)
|
172 |
+
temp_dict[f"{component}"] = [sample_component_A, sample_component_B]
|
173 |
+
if expected_shape_qkv_a is None and not component.startswith("add_"):
|
174 |
+
expected_shape_qkv_a = sample_component_A.shape
|
175 |
+
expected_shape_qkv_b = sample_component_B.shape
|
176 |
+
dtype = sample_component_A.dtype
|
177 |
+
device = sample_component_A.device
|
178 |
+
if expected_shape_add_qkv_a is None and component.startswith("add_"):
|
179 |
+
expected_shape_add_qkv_a = sample_component_A.shape
|
180 |
+
expected_shape_add_qkv_b = sample_component_B.shape
|
181 |
+
dtype = sample_component_A.dtype
|
182 |
+
device = sample_component_A.device
|
183 |
+
else:
|
184 |
+
logger.info(
|
185 |
+
f"Skipping layer {i} since no LoRA weight is available for {sample_component_A_key}"
|
186 |
+
)
|
187 |
+
temp_dict[f"{component}"] = [None, None]
|
188 |
+
|
189 |
+
if device is not None:
|
190 |
+
if expected_shape_qkv_a is not None:
|
191 |
+
|
192 |
+
if (sq := temp_dict["to_q"])[0] is not None:
|
193 |
+
sample_q_A, sample_q_B = sq
|
194 |
+
else:
|
195 |
+
sample_q_A, sample_q_B = [
|
196 |
+
torch.zeros(expected_shape_qkv_a, dtype=dtype, device=device),
|
197 |
+
torch.zeros(expected_shape_qkv_b, dtype=dtype, device=device),
|
198 |
+
]
|
199 |
+
if (sq := temp_dict["to_k"])[0] is not None:
|
200 |
+
sample_k_A, sample_k_B = sq
|
201 |
+
else:
|
202 |
+
sample_k_A, sample_k_B = [
|
203 |
+
torch.zeros(expected_shape_qkv_a, dtype=dtype, device=device),
|
204 |
+
torch.zeros(expected_shape_qkv_b, dtype=dtype, device=device),
|
205 |
+
]
|
206 |
+
if (sq := temp_dict["to_v"])[0] is not None:
|
207 |
+
sample_v_A, sample_v_B = sq
|
208 |
+
else:
|
209 |
+
sample_v_A, sample_v_B = [
|
210 |
+
torch.zeros(expected_shape_qkv_a, dtype=dtype, device=device),
|
211 |
+
torch.zeros(expected_shape_qkv_b, dtype=dtype, device=device),
|
212 |
+
]
|
213 |
+
original_state_dict[f"double_blocks.{i}.img_attn.qkv.lora_A.weight"] = (
|
214 |
+
torch.cat([sample_q_A, sample_k_A, sample_v_A], dim=0)
|
215 |
+
)
|
216 |
+
original_state_dict[f"double_blocks.{i}.img_attn.qkv.lora_B.weight"] = (
|
217 |
+
torch.cat([sample_q_B, sample_k_B, sample_v_B], dim=0)
|
218 |
+
)
|
219 |
+
if expected_shape_add_qkv_a is not None:
|
220 |
+
|
221 |
+
if (sq := temp_dict["add_q_proj"])[0] is not None:
|
222 |
+
context_q_A, context_q_B = sq
|
223 |
+
else:
|
224 |
+
context_q_A, context_q_B = [
|
225 |
+
torch.zeros(
|
226 |
+
expected_shape_add_qkv_a, dtype=dtype, device=device
|
227 |
+
),
|
228 |
+
torch.zeros(
|
229 |
+
expected_shape_add_qkv_b, dtype=dtype, device=device
|
230 |
+
),
|
231 |
+
]
|
232 |
+
if (sq := temp_dict["add_k_proj"])[0] is not None:
|
233 |
+
context_k_A, context_k_B = sq
|
234 |
+
else:
|
235 |
+
context_k_A, context_k_B = [
|
236 |
+
torch.zeros(
|
237 |
+
expected_shape_add_qkv_a, dtype=dtype, device=device
|
238 |
+
),
|
239 |
+
torch.zeros(
|
240 |
+
expected_shape_add_qkv_b, dtype=dtype, device=device
|
241 |
+
),
|
242 |
+
]
|
243 |
+
if (sq := temp_dict["add_v_proj"])[0] is not None:
|
244 |
+
context_v_A, context_v_B = sq
|
245 |
+
else:
|
246 |
+
context_v_A, context_v_B = [
|
247 |
+
torch.zeros(
|
248 |
+
expected_shape_add_qkv_a, dtype=dtype, device=device
|
249 |
+
),
|
250 |
+
torch.zeros(
|
251 |
+
expected_shape_add_qkv_b, dtype=dtype, device=device
|
252 |
+
),
|
253 |
+
]
|
254 |
+
|
255 |
+
original_state_dict[f"double_blocks.{i}.txt_attn.qkv.lora_A.weight"] = (
|
256 |
+
torch.cat([context_q_A, context_k_A, context_v_A], dim=0)
|
257 |
+
)
|
258 |
+
original_state_dict[f"double_blocks.{i}.txt_attn.qkv.lora_B.weight"] = (
|
259 |
+
torch.cat([context_q_B, context_k_B, context_v_B], dim=0)
|
260 |
+
)
|
261 |
|
262 |
# qk_norm
|
263 |
original_state_dict, diffusers_state_dict = convert_if_lora_exists(
|
|
|
332 |
for i in range(num_single_layers):
|
333 |
block_prefix = f"single_transformer_blocks.{i}."
|
334 |
# norm.linear -> single_blocks.0.modulation.lin
|
335 |
+
key_norm = f"{prefix}{block_prefix}norm.linear.weight"
|
336 |
original_state_dict, diffusers_state_dict = convert_if_lora_exists(
|
337 |
original_state_dict,
|
338 |
diffusers_state_dict,
|
339 |
+
key_norm,
|
340 |
f"single_blocks.{i}.modulation.lin.weight",
|
341 |
)
|
342 |
|
343 |
+
has_q, has_k, has_v, has_mlp = False, False, False, False
|
344 |
+
shape_qkv_a = None
|
345 |
+
shape_qkv_b = None
|
346 |
# Q, K, V, mlp
|
347 |
q_A = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_q.lora_A.weight")
|
348 |
q_B = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_q.lora_B.weight")
|
349 |
+
if q_A is not None and q_B is not None:
|
350 |
+
has_q = True
|
351 |
+
shape_qkv_a = q_A.shape
|
352 |
+
shape_qkv_b = q_B.shape
|
353 |
k_A = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_k.lora_A.weight")
|
354 |
k_B = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_k.lora_B.weight")
|
355 |
+
if k_A is not None and k_B is not None:
|
356 |
+
has_k = True
|
357 |
+
shape_qkv_a = k_A.shape
|
358 |
+
shape_qkv_b = k_B.shape
|
359 |
v_A = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_v.lora_A.weight")
|
360 |
v_B = diffusers_state_dict.pop(f"{prefix}{block_prefix}attn.to_v.lora_B.weight")
|
361 |
+
if v_A is not None and v_B is not None:
|
362 |
+
has_v = True
|
363 |
+
shape_qkv_a = v_A.shape
|
364 |
+
shape_qkv_b = v_B.shape
|
365 |
mlp_A = diffusers_state_dict.pop(
|
366 |
f"{prefix}{block_prefix}proj_mlp.lora_A.weight"
|
367 |
)
|
368 |
mlp_B = diffusers_state_dict.pop(
|
369 |
f"{prefix}{block_prefix}proj_mlp.lora_B.weight"
|
370 |
)
|
371 |
+
if mlp_A is not None and mlp_B is not None:
|
372 |
+
has_mlp = True
|
373 |
+
shape_qkv_a = mlp_A.shape
|
374 |
+
shape_qkv_b = mlp_B.shape
|
375 |
+
if any([has_q, has_k, has_v, has_mlp]):
|
376 |
+
if not has_q:
|
377 |
+
q_A, q_B = [
|
378 |
+
torch.zeros(shape_qkv_a, dtype=dtype, device=device),
|
379 |
+
torch.zeros(shape_qkv_b, dtype=dtype, device=device),
|
380 |
+
]
|
381 |
+
if not has_k:
|
382 |
+
k_A, k_B = [
|
383 |
+
torch.zeros(shape_qkv_a, dtype=dtype, device=device),
|
384 |
+
torch.zeros(shape_qkv_b, dtype=dtype, device=device),
|
385 |
+
]
|
386 |
+
if not has_v:
|
387 |
+
v_A, v_B = [
|
388 |
+
torch.zeros(shape_qkv_a, dtype=dtype, device=device),
|
389 |
+
torch.zeros(shape_qkv_b, dtype=dtype, device=device),
|
390 |
+
]
|
391 |
+
if not has_mlp:
|
392 |
+
mlp_A, mlp_B = [
|
393 |
+
torch.zeros(shape_qkv_a, dtype=dtype, device=device),
|
394 |
+
torch.zeros(shape_qkv_b, dtype=dtype, device=device),
|
395 |
+
]
|
396 |
+
original_state_dict[f"single_blocks.{i}.linear1.lora_A.weight"] = torch.cat(
|
397 |
+
[q_A, k_A, v_A, mlp_A], dim=0
|
398 |
+
)
|
399 |
+
original_state_dict[f"single_blocks.{i}.linear1.lora_B.weight"] = torch.cat(
|
400 |
+
[q_B, k_B, v_B, mlp_B], dim=0
|
401 |
+
)
|
402 |
|
403 |
# output projections
|
404 |
original_state_dict, diffusers_state_dict = convert_if_lora_exists(
|
|
|
432 |
return original_state_dict
|
433 |
|
434 |
|
435 |
+
def convert_from_original_flux_checkpoint(original_state_dict: StateDict) -> StateDict:
|
436 |
+
"""
|
437 |
+
Convert the state dict from the original Flux checkpoint format to the new format.
|
438 |
+
|
439 |
+
Args:
|
440 |
+
original_state_dict (Dict[str, torch.Tensor]): The original Flux checkpoint state dict.
|
441 |
+
|
442 |
+
Returns:
|
443 |
+
Dict[str, torch.Tensor]: The converted state dict in the new format.
|
444 |
+
"""
|
445 |
sd = {
|
446 |
k.replace("lora_unet_", "")
|
447 |
.replace("double_blocks_", "double_blocks.")
|
|
|
473 |
return module
|
474 |
|
475 |
|
476 |
+
def get_lora_for_key(
|
477 |
+
key: str, lora_weights: dict
|
478 |
+
) -> Optional[Tuple[torch.Tensor, torch.Tensor, Optional[float]]]:
|
479 |
+
"""
|
480 |
+
Get LoRA weights for a specific key.
|
481 |
+
|
482 |
+
Args:
|
483 |
+
key (str): The key to look up in the LoRA weights.
|
484 |
+
lora_weights (dict): Dictionary containing LoRA weights.
|
485 |
+
|
486 |
+
Returns:
|
487 |
+
Optional[Tuple[torch.Tensor, torch.Tensor, Optional[float]]]: A tuple containing lora_A, lora_B, and alpha if found, None otherwise.
|
488 |
+
"""
|
489 |
prefix = key.split(".lora")[0]
|
490 |
+
lora_A = lora_weights.get(f"{prefix}.lora_A.weight")
|
491 |
+
lora_B = lora_weights.get(f"{prefix}.lora_B.weight")
|
492 |
+
alpha = lora_weights.get(f"{prefix}.alpha")
|
493 |
+
|
494 |
+
if lora_A is None or lora_B is None:
|
495 |
+
return None
|
496 |
return lora_A, lora_B, alpha
|
497 |
|
498 |
|
499 |
+
def get_module_for_key(
|
500 |
+
key: str, model: Flux
|
501 |
+
) -> F8Linear | torch.nn.Linear | CublasLinear:
|
502 |
+
parts = key.split(".")
|
503 |
+
module = model
|
504 |
+
for part in parts:
|
505 |
+
module = getattr(module, part)
|
506 |
+
return module
|
507 |
+
|
508 |
+
|
509 |
def calculate_lora_weight(
|
510 |
lora_weights: Tuple[torch.Tensor, torch.Tensor, Union[torch.Tensor, float]],
|
511 |
rank: Optional[int] = None,
|
|
|
529 |
w_down = lora_B.to(dtype=dtype, device=device)
|
530 |
|
531 |
if alpha != rank:
|
532 |
+
w_up = w_up * alpha / rank
|
|
|
533 |
if uneven_rank:
|
534 |
+
# Fuse each lora instead of repeat interleave for each individual lora,
|
535 |
+
# seems to fuse more correctly.
|
536 |
+
fused_lora = torch.zeros(
|
537 |
+
(lora_B.shape[0], lora_A.shape[1]), device=device, dtype=dtype
|
538 |
+
)
|
539 |
+
w_up = w_up.chunk(int(rank_diff), dim=0)
|
540 |
+
for w_up_chunk in w_up:
|
541 |
+
fused_lora = fused_lora + (lora_scale * torch.mm(w_down, w_up_chunk))
|
542 |
else:
|
543 |
fused_lora = lora_scale * torch.mm(w_down, w_up)
|
544 |
return fused_lora
|
|
|
589 |
lora_weights = convert_from_original_flux_checkpoint(lora_weights)
|
590 |
logger.info("LoRA weights loaded")
|
591 |
logger.debug("Extracting keys")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
592 |
keys_without_ab = list(
|
593 |
set(
|
594 |
[
|
|
|
597 |
.replace(".lora_A", "")
|
598 |
.replace(".lora_B", "")
|
599 |
.replace(".alpha", "")
|
600 |
+
for key in lora_weights.keys()
|
601 |
]
|
602 |
)
|
603 |
)
|
604 |
+
logger.debug("Keys extracted")
|
605 |
return keys_without_ab, lora_weights
|
606 |
|
607 |
|
|
|
648 |
module = get_module_for_key(key, model)
|
649 |
weight, is_f8, dtype = extract_weight_from_linear(module)
|
650 |
lora_sd = get_lora_for_key(key, lora_weights)
|
651 |
+
if lora_sd is None:
|
652 |
+
# Skipping LoRA application for this module
|
653 |
+
continue
|
654 |
weight = apply_lora_weight_to_module(weight, lora_sd, lora_scale=lora_scale)
|
655 |
if is_f8:
|
656 |
module.set_weight_tensor(weight.type(dtype))
|
|
|
678 |
module = get_module_for_key(key, model)
|
679 |
weight, is_f8, dtype = extract_weight_from_linear(module)
|
680 |
lora_sd = get_lora_for_key(key, lora_weights)
|
681 |
+
if lora_sd is None:
|
682 |
+
# Skipping LoRA application for this module
|
683 |
+
continue
|
684 |
weight = unfuse_lora_weight_from_module(weight, lora_sd, lora_scale=lora_scale)
|
685 |
if is_f8:
|
686 |
module.set_weight_tensor(weight.type(dtype))
|