Skip to content

Commit 48b898f

Browse files
committed
Onboard FLUX.2-klein pipeline and models into MaxDiffusion (JAX+TPU)
1 parent 55adaa2 commit 48b898f

12 files changed

Lines changed: 1208 additions & 2559 deletions

src/maxdiffusion/configs/base_flux2klein.yml

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -161,9 +161,9 @@ dcn_fsdp_parallelism: -1
161161
dcn_context_parallelism: 1
162162
dcn_tensor_parallelism: 1
163163
ici_data_parallelism: 1
164-
ici_fsdp_parallelism: 1
164+
ici_fsdp_parallelism: -1
165165
ici_context_parallelism: 1
166-
ici_tensor_parallelism: 4
166+
ici_tensor_parallelism: 1
167167

168168
allow_split_physical_axes: False
169169

src/maxdiffusion/generate_flux2klein.py

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -15,6 +15,7 @@
1515
import gc
1616
import os
1717
import time
18+
import sys
1819
from typing import List
1920

2021
from absl import app
@@ -30,7 +31,7 @@
3031
from maxdiffusion.max_utils import create_device_mesh
3132
from maxdiffusion.train_utils import transformer_engine_context
3233

33-
from maxdiffusion.models.flux.transformers.transformer_flux2klein_flax import Flux2KleinTransformer2DModel
34+
from maxdiffusion.models.flux.transformers.transformer_flux_flax import Flux2KleinTransformer2DModel
3435
from maxdiffusion.models.vae_flax import FlaxAutoencoderKL
3536
from maxdiffusion.models.qwen3_flax import FlaxQwen3Config, FlaxQwen3Model, load_and_convert_qwen3_weights
3637
from maxdiffusion.schedulers.scheduling_flow_match_flax import FlaxFlowMatchScheduler

src/maxdiffusion/models/attention_flax.py

Lines changed: 241 additions & 49 deletions
Large diffs are not rendered by default.

0 commit comments

Comments
 (0)