Support multi lora loading (#137)

* adds multi lora support * update readme
AI-Hypercomputer · Dec 13, 2024 · efa11e4 · efa11e4
1 parent b1c63d7
commit efa11e4
Show file tree

Hide file tree

Showing 5 changed files with 68 additions and 40 deletions.
diff --git a/README.md b/README.md
@@ -17,6 +17,7 @@
 [![Unit Tests](https://github.com/google/maxtext/actions/workflows/UnitTests.yml/badge.svg)](https://github.com/google/maxdiffusion/actions/workflows/UnitTests.yml)
 
 # What's new?
+- **`2024/12/12`**: Load multiple LoRAs for inference.
 - **`2024/10/22`**: LoRA support for Hyper SDXL.
 - **`2024/8/1`**: Orbax is the new default checkpointer. You can still use `pipeline.save_pretrained` after training to save in diffusers format.
 - **`2024/7/20`**: Dreambooth training for Stable Diffusion 1.x,2.x is now supported.
@@ -33,6 +34,7 @@ MaxDiffusion supports
 * Stable Diffusion XL (training and inference).
 * Stable Diffusion Lightning (inference).
 * Hyper-SD XL LoRA loading (inference).
+* Load Multiple LoRA (SDXL inference).
 * ControlNet inference (Stable Diffusion 1.4 & SDXL).
 * Dreambooth training support for Stable Diffusion 1.x,2.x.
 
@@ -45,6 +47,7 @@ MaxDiffusion supports
       * [Dreambooth](#dreambooth)
     * [Inference](#inference)
       * [Hyper-SD XL LoRA](#hyper-sdxl-lora)
+      * [Load Multiple LoRA](#load-multiple-lora)
       * [SDXL Lightning](#sdxl-lightning)
       * [ControlNet](#controlnet)
 * [Comparison To Alternatives](#comparison-to-alternatives)
@@ -139,6 +142,14 @@ To generate images, run the following command:
   python src/maxdiffusion/generate_sdxl.py src/maxdiffusion/configs/base_xl.yml run_name="test-lora" output_dir=/tmp/ jax_cache_dir=/tmp/cache_dir/ num_inference_steps=2 do_classifier_free_guidance=False prompt="a photograph of a cat wearing a hat riding a skateboard in a park." per_device_batch_size=1 pretrained_model_name_or_path="Lykon/AAM_XL_AnimeMix" from_pt=True revision=main diffusion_scheduler_config='{"_class_name" : "FlaxDDIMScheduler", "timestep_spacing" : "trailing"}' lora_config='{"lora_model_name_or_path" : ["ByteDance/Hyper-SD"], "weight_name" : ["Hyper-SDXL-2steps-lora.safetensors"], "adapter_name" : ["hyper-sdxl"], "scale": [0.7], "from_pt": ["true"]}'
   ```
 
+  ## Load Multiple LoRA
+
+    Supports loading multiple LoRAs for inference. Both from local or from HuggingFace hub.
+
+    ```bash
+    python src/maxdiffusion/generate_sdxl.py src/maxdiffusion/configs/base_xl.yml run_name="test-lora" output_dir=/tmp/tmp/ jax_cache_dir=/tmp/cache_dir/ num_inference_steps=30 do_classifier_free_guidance=True prompt="ultra detailed diagram blueprint of a papercut Sitting MaineCoon cat, wide canvas, ampereart, electrical diagram, bl3uprint, papercut" per_device_batch_size=1 diffusion_scheduler_config='{"_class_name" : "FlaxDDIMScheduler", "timestep_spacing" : "trailing"}' lora_config='{"lora_model_name_or_path" : ["/home/jfacevedo/blueprintify-sd-xl-10.safetensors","TheLastBen/Papercut_SDXL"], "weight_name" : ["/home/jfacevedo/blueprintify-sd-xl-10.safetensors","papercut.safetensors"], "adapter_name" : ["blueprint","papercut"], "scale": [0.8, 0.7], "from_pt": ["true", "true"]}'
+    ```
+
   ## SDXL Lightning
 
   Single and Multi host inference is supported with sharding annotations:

diff --git a/src/maxdiffusion/generate_sdxl.py b/src/maxdiffusion/generate_sdxl.py
@@ -16,6 +16,7 @@
 
 import functools
 from absl import app
+from contextlib import ExitStack
 from typing import Sequence
 import time
 
@@ -233,14 +234,15 @@ def run(config):
     params["unet"] = unet_params
 
   # maybe load lora and create interceptor
-  params, lora_interceptor = maybe_load_lora(config, pipeline, params)
+  params, lora_interceptors = maybe_load_lora(config, pipeline, params)
 
   if config.lightning_repo:
     pipeline, params = load_sdxllightning_unet(config, pipeline, params)
 
   # Don't restore the full train state, instead, just restore params
   # and create an inference state.
-  with nn.intercept_methods(lora_interceptor):
+  with ExitStack() as stack:
+    _ = [stack.enter_context(nn.intercept_methods(interceptor)) for interceptor in lora_interceptors]
     unet_state, unet_state_shardings = max_utils.setup_initial_state(
         model=pipeline.unet,
         tx=None,
@@ -254,7 +256,8 @@ def run(config):
   vae_state, vae_state_shardings = checkpoint_loader.create_vae_state(
       pipeline, params, checkpoint_item_name="vae_state", is_training=False
   )
-  with nn.intercept_methods(lora_interceptor):
+  with ExitStack() as stack:
+    _ = [stack.enter_context(nn.intercept_methods(interceptor)) for interceptor in lora_interceptors]
     text_encoder_state, text_encoder_state_shardings = checkpoint_loader.create_text_encoder_state(
         pipeline, params, checkpoint_item_name="text_encoder_state", is_training=False
     )
@@ -293,11 +296,13 @@ def run(config):
   )
 
   s = time.time()
-  with nn.intercept_methods(lora_interceptor):
+  with ExitStack() as stack:
+    _ = [stack.enter_context(nn.intercept_methods(interceptor)) for interceptor in lora_interceptors]
     p_run_inference(states).block_until_ready()
   print("compile time: ", (time.time() - s))
   s = time.time()
-  with nn.intercept_methods(lora_interceptor):
+  with ExitStack() as stack:
+    _ = [stack.enter_context(nn.intercept_methods(interceptor)) for interceptor in lora_interceptors]
     images = p_run_inference(states).block_until_ready()
   print("inference time: ", (time.time() - s))
   images = jax.experimental.multihost_utils.process_allgather(images)

diff --git a/src/maxdiffusion/loaders/lora_pipeline.py b/src/maxdiffusion/loaders/lora_pipeline.py
@@ -88,7 +88,7 @@ def load_lora_weights(
     return params, rank, network_alphas
 
   @classmethod
-  def _get_lora_layer(cls, module_path, module, rank, network_alphas):
+  def _get_lora_layer(cls, module_path, module, rank, network_alphas, adapter_name):
     is_conv = any("conv" in str_ for str_ in module_path)
     network_alpha = network_alphas.get(module_path, None)
     if is_conv:
@@ -105,7 +105,7 @@ def _get_lora_layer(cls, module_path, module, rank, network_alphas):
           dtype=module.dtype,
           weights_dtype=module.param_dtype,
           precision=module.precision,
-          name="lora",
+          name=f"lora-{adapter_name}",
       )
     else:
       lora_module = LoRALinearLayer(
@@ -115,39 +115,41 @@ def _get_lora_layer(cls, module_path, module, rank, network_alphas):
           dtype=module.dtype,
           weights_dtype=module.param_dtype,
           precision=module.precision,
-          name="lora",
+          name=f"lora-{adapter_name}",
       )
     return lora_module
 
-  def rename_for_interceptor(params_keys, network_alphas):
+  def rename_for_interceptor(params_keys, network_alphas, adapter_name):
     new_params_keys = []
     new_network_alphas = {}
+    lora_name = f"lora-{adapter_name}"
     for layer_lora in params_keys:
-      if "lora" in layer_lora:
-        new_layer_lora = layer_lora[: layer_lora.index("lora")]
+      if lora_name in layer_lora:
+        new_layer_lora = layer_lora[: layer_lora.index(lora_name)]
         if new_layer_lora not in new_params_keys:
           new_params_keys.append(new_layer_lora)
           network_alpha = network_alphas[layer_lora]
           new_network_alphas[new_layer_lora] = network_alpha
     return new_params_keys, new_network_alphas
 
   @classmethod
-  def make_lora_interceptor(cls, params, rank, network_alphas):
+  def make_lora_interceptor(cls, params, rank, network_alphas, adapter_name):
     # Only unet interceptor supported for now.
     network_alphas_for_interceptor = {}
 
     unet_lora_keys = flax.traverse_util.flatten_dict(params["unet"]).keys()
-    lora_keys, unet_alphas = cls.rename_for_interceptor(unet_lora_keys, network_alphas)
+    lora_keys, unet_alphas = cls.rename_for_interceptor(unet_lora_keys, network_alphas, adapter_name)
     network_alphas_for_interceptor.update(unet_alphas)
 
     text_encoder_keys = flax.traverse_util.flatten_dict(params["text_encoder"]).keys()
-    text_encoder_keys, text_encoder_alphas = cls.rename_for_interceptor(text_encoder_keys, network_alphas)
+    text_encoder_keys, text_encoder_alphas = cls.rename_for_interceptor(text_encoder_keys, network_alphas, adapter_name)
     lora_keys.extend(text_encoder_keys)
     network_alphas_for_interceptor.update(text_encoder_alphas)
-
     if "text_encoder_2" in params.keys():
       text_encoder_2_keys = flax.traverse_util.flatten_dict(params["text_encoder_2"]).keys()
-      text_encoder_2_keys, text_encoder_2_alphas = cls.rename_for_interceptor(text_encoder_2_keys, network_alphas)
+      text_encoder_2_keys, text_encoder_2_alphas = cls.rename_for_interceptor(
+          text_encoder_2_keys, network_alphas, adapter_name
+      )
       lora_keys.extend(text_encoder_2_keys)
       network_alphas_for_interceptor.update(text_encoder_2_alphas)
 
@@ -161,7 +163,7 @@ def _intercept(next_fn, args, kwargs, context):
       if context.method_name == "__call__":
         module_path = context.module.path
         if module_path in lora_keys:
-          lora_layer = cls._get_lora_layer(module_path, context.module, rank, network_alphas_for_interceptor)
+          lora_layer = cls._get_lora_layer(module_path, context.module, rank, network_alphas_for_interceptor, adapter_name)
           return lora_layer(h, *args, **kwargs)
       return h
 
@@ -290,5 +292,5 @@ def load_lora(cls, state_dict, network_alphas, params, adapter_name=None, _pipel
             `default_{i}` where i is the total number of adapters being loaded.
     """
     # Load the layers corresponding to Unet.
-    params, rank, network_alphas = convert_lora_pytorch_state_dict_to_flax(state_dict, params, network_alphas)
+    params, rank, network_alphas = convert_lora_pytorch_state_dict_to_flax(state_dict, params, network_alphas, adapter_name)
     return params, rank, network_alphas
diff --git a/src/maxdiffusion/maxdiffusion_utils.py b/src/maxdiffusion/maxdiffusion_utils.py
@@ -45,21 +45,24 @@ def _noop_interceptor(next_fn, args, kwargs, context):
     return next_fn(*args, **kwargs)
 
   lora_config = config.lora_config
-  interceptor = _noop_interceptor
+  interceptors = [_noop_interceptor]
   if len(lora_config["lora_model_name_or_path"]) > 0:
     # For now only first lora supported. In the future, they will be merged
     # before being loaded.
     # TODO - merge LoRAs here.
-    params, rank, network_alphas = pipeline.load_lora_weights(
-        lora_config["lora_model_name_or_path"][0],
-        weight_name=lora_config["weight_name"][0],
-        params=params,
-        adapter_name=lora_config["adapter_name"][0],
-        unet_config=pipeline.unet.config,
-    )
-    interceptor = pipeline.make_lora_interceptor(params, rank, network_alphas)
+    interceptors = []
+    for i in range(len(lora_config["lora_model_name_or_path"])):
+      params, rank, network_alphas = pipeline.load_lora_weights(
+          lora_config["lora_model_name_or_path"][i],
+          weight_name=lora_config["weight_name"][i],
+          params=params,
+          adapter_name=lora_config["adapter_name"][i],
+          unet_config=pipeline.unet.config,
+      )
+      interceptor = pipeline.make_lora_interceptor(params, rank, network_alphas, lora_config["adapter_name"][i])
+      interceptors.append(interceptor)
 
-  return params, interceptor
+  return params, interceptors
 
 
 def vae_apply(images, sample_rng, vae, vae_params):

diff --git a/src/maxdiffusion/models/modeling_flax_pytorch_utils.py b/src/maxdiffusion/models/modeling_flax_pytorch_utils.py
@@ -130,7 +130,13 @@ def get_network_alpha_value(pt_key, network_alphas):
 
 
 def create_flax_params_from_pytorch_state(
-    pt_state_dict, unet_state_dict, text_encoder_state_dict, text_encoder_2_state_dict, network_alphas, is_lora=False
+    pt_state_dict,
+    unet_state_dict,
+    text_encoder_state_dict,
+    text_encoder_2_state_dict,
+    network_alphas,
+    adapter_name,
+    is_lora=False,
 ):
   rank = None
   renamed_network_alphas = {}
@@ -157,19 +163,21 @@ def create_flax_params_from_pytorch_state(
       flax_key_list = [*pt_tuple_key]
       if "text_encoder" in pt_tuple_key or "text_encoder_2" in pt_tuple_key:
         rename_from_to = (
-            ("to_k_lora", ("k_proj", "lora")),
-            ("to_q_lora", ("q_proj", "lora")),
-            ("to_v_lora", ("v_proj", "lora")),
-            ("to_out_lora", ("out_proj", "lora")),
+            ("to_k_lora", ("k_proj", f"lora-{adapter_name}")),
+            ("to_q_lora", ("q_proj", f"lora-{adapter_name}")),
+            ("to_v_lora", ("v_proj", f"lora-{adapter_name}")),
+            ("to_out_lora", ("out_proj", f"lora-{adapter_name}")),
+            ("lora", f"lora-{adapter_name}"),
             ("weight", "kernel"),
         )
       # the unet
       else:
         rename_from_to = (
-            ("to_k_lora", ("to_k", "lora")),
-            ("to_q_lora", ("to_q", "lora")),
-            ("to_v_lora", ("to_v", "lora")),
-            ("to_out_lora", ("to_out_0", "lora")),
+            ("to_k_lora", ("to_k", f"lora-{adapter_name}")),
+            ("to_q_lora", ("to_q", f"lora-{adapter_name}")),
+            ("to_v_lora", ("to_v", f"lora-{adapter_name}")),
+            ("to_out_lora", ("to_out_0", f"lora-{adapter_name}")),
+            ("lora", f"lora-{adapter_name}"),
             ("weight", "kernel"),
         )
       for rename_from, rename_to in rename_from_to:
@@ -206,11 +214,10 @@ def create_flax_params_from_pytorch_state(
 
     if network_alpha_value >= 0:
       renamed_network_alphas[tuple(flax_key_list)] = network_alpha_value
-
   return unet_state_dict, text_encoder_state_dict, text_encoder_2_state_dict, rank, renamed_network_alphas
 
 
-def convert_lora_pytorch_state_dict_to_flax(pt_state_dict, params, network_alphas):
+def convert_lora_pytorch_state_dict_to_flax(pt_state_dict, params, network_alphas, adapter_name):
   # Step 1: Convert pytorch tensor to numpy
   # sometimes we load weights in bf16 and numpy doesn't support it
   pt_state_dict = {k: v.float().numpy() for k, v in pt_state_dict.items()}
@@ -223,7 +230,7 @@ def convert_lora_pytorch_state_dict_to_flax(pt_state_dict, params, network_alpha
     text_encoder_2_params = None
   (unet_state_dict, text_encoder_state_dict, text_encoder_2_state_dict, rank, network_alphas) = (
       create_flax_params_from_pytorch_state(
-          pt_state_dict, unet_params, text_encoder_params, text_encoder_2_params, network_alphas, is_lora=True
+          pt_state_dict, unet_params, text_encoder_params, text_encoder_2_params, network_alphas, adapter_name, is_lora=True
       )
   )
   params["unet"] = unflatten_dict(unet_state_dict)