Add INT8 weight-only quantized transformer for FLUX.2-dev

README.md

@@ -0,0 +1,79 @@
+# FLUX.2-dev – Transformer INT8 Weight-Only (torchao)
+
+This repository provides an **INT8 weight-only quantized transformer** for  
+[`black-forest-labs/FLUX.2-dev`](https://huggingface.co/black-forest-labs/FLUX.2-dev).
+
+Only the **transformer** is quantized and redistributed.  
+All other components (VAE, text encoders, scheduler, etc.) are loaded from the original model.
+
+---
+
+## What is included
+
+- ✅ INT8 weight-only quantized **transformer**
+- ❌ No VAE
+- ❌ No text encoders
+- ❌ No scheduler
+
+Quantization is performed using **torchao** (INT8 weight-only).
+
+---
+
+## Why this exists
+
+- Reduce VRAM usage of FLUX.2-dev
+- Keep compatibility with Diffusers pipelines
+- Avoid bitsandbytes (not supported on ROCm)
+- Enable deployment on AMD GPUs (MI200 / MI210 / MI300)
+
+---
+
+## Requirements
+
+- PyTorch with CUDA or ROCm
+- `diffusers` (git main recommended)
+- `torchao`
+- `transformers`
+- `huggingface-hub`
+
+> ⚠️ The quantized transformer **cannot be loaded with safetensors**.
+
+---
+
+## How to use
+
+```python
+import torch
+from diffusers import Flux2Pipeline, AutoModel
+
+BASE_MODEL = "black-forest-labs/FLUX.2-dev"
+INT8_REPO  = "Atech/FLUX.2-dev-transformer-int8wo"
+
+dtype = torch.bfloat16
+
+# Load INT8 transformer
+transformer = AutoModel.from_pretrained(
+    INT8_REPO,
+    subfolder="transformer_int8wo",
+    torch_dtype=dtype,
+    use_safetensors=False,
+)
+
+# Build pipeline using original FLUX.2-dev
+pipe = Flux2Pipeline.from_pretrained(
+    BASE_MODEL,
+    transformer=transformer,
+    torch_dtype=dtype,
+    device_map="balanced",   # recommended
+)
+
+# Example generation
+image = pipe(
+    prompt="A futuristic data center server rack",
+    num_inference_steps=35,
+    guidance_scale=4,
+    height=1024,
+    width=1024,
+).images[0]
+
+image.save("output.png")

quantize_transformer_int8wo.py

@@ -0,0 +1,39 @@
+import os
+import torch
+from diffusers import AutoModel
+from torchao.quantization import quantize_, Int8WeightOnlyConfig
+
+BASE_MODEL = "black-forest-labs/FLUX.2-dev"
+OUT_DIR = "transformer_int8wo"
+
+def main():
+    dtype = torch.bfloat16
+
+    print(f"Loading transformer from {BASE_MODEL}")
+    transformer = AutoModel.from_pretrained(
+        BASE_MODEL,
+        subfolder="transformer",
+        torch_dtype=dtype,
+    )
+
+    print("Applying INT8 weight-only quantization (torchao)")
+    try:
+        cfg = Int8WeightOnlyConfig(version=2)
+    except TypeError:
+        cfg = Int8WeightOnlyConfig()
+
+    quantize_(transformer, cfg)
+
+    os.makedirs(OUT_DIR, exist_ok=True)
+
+    print(f"Saving quantized transformer to ./{OUT_DIR}")
+    # IMPORTANT: torchao quantized weights are NOT safetensors-compatible
+    transformer.save_pretrained(
+        OUT_DIR,
+        safe_serialization=False
+    )
+
+    print("Done.")
+
+if __name__ == "__main__":
+    main()

requirements.txt

@@ -0,0 +1,13 @@
+
+---
+
+# 4️⃣ requirements.txt
+
+📄 `requirements.txt`
+
+```txt
+torch
+diffusers
+transformers
+torchao
+huggingface-hub

transformer_int8wo/config.json

@@ -0,0 +1,23 @@
+{
+  "_class_name": "Flux2Transformer2DModel",
+  "_diffusers_version": "0.37.0.dev0",
+  "_name_or_path": "black-forest-labs/FLUX.2-dev",
+  "attention_head_dim": 128,
+  "axes_dims_rope": [
+    32,
+    32,
+    32,
+    32
+  ],
+  "eps": 1e-06,
+  "in_channels": 128,
+  "joint_attention_dim": 15360,
+  "mlp_ratio": 3.0,
+  "num_attention_heads": 48,
+  "num_layers": 8,
+  "num_single_layers": 48,
+  "out_channels": null,
+  "patch_size": 1,
+  "rope_theta": 2000,
+  "timestep_guidance_channels": 256
+}

transformer_int8wo/diffusion_pytorch_model-00001-of-00004.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d7cb821186499fd7cb38bab41007e71ff7f2bb3d5a4d6335023dba4d374c3b2
+size 9915862927

transformer_int8wo/diffusion_pytorch_model-00002-of-00004.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce0f6fdf2adcc65fa1c1960c5f40dcefb70ebb7d563a7520367bedc829069d0e
+size 9968194367

transformer_int8wo/diffusion_pytorch_model-00003-of-00004.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c53346cb517438bbedf4cc2ef8a170cd11ff687926ba159bc3c3d5dd0190f11b
+size 9817186325

transformer_int8wo/diffusion_pytorch_model-00004-of-00004.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c02970fe6b55ee00b99bfeeb49da66ef7b3a4c30a42841b008a7b6d8a4a77da2
+size 2530607445

transformer_int8wo/diffusion_pytorch_model.bin.index.json

@@ -0,0 +1,338 @@
+{
+  "metadata": {
+    "total_size": 32231641344
+  },
+  "weight_map": {
+    "context_embedder.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "double_stream_modulation_img.linear.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "double_stream_modulation_txt.linear.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "norm_out.linear.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "proj_out.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_stream_modulation.linear.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.0.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.0.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.0.attn.to_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.0.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.1.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.1.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.1.attn.to_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.1.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.10.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.10.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.10.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.10.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.11.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.11.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.11.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.11.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.12.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.12.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.12.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.12.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.13.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.13.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.13.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.13.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.14.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.14.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.14.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.14.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.15.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.15.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.15.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.15.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.16.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.16.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.16.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.16.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.17.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.17.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.17.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.17.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.18.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.18.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.18.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.18.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.19.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.19.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.19.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.19.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.2.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.2.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.2.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.2.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "single_transformer_blocks.20.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.20.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.20.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.20.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.21.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.21.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.21.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.21.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.22.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.22.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.22.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.22.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.23.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.23.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.23.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.23.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.24.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.24.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.24.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.24.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.25.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.25.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.25.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.25.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.26.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.26.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.26.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.26.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.27.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.27.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.27.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.27.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.28.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.28.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.28.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.28.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.29.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.29.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.29.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.29.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.3.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.3.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.3.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.3.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.30.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.30.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.30.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.30.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.31.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.31.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.31.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.31.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.32.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.32.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.32.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.32.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.33.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.33.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.33.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.33.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.34.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.34.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.34.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.34.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.35.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.35.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.35.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.35.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.36.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.36.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.36.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.36.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.37.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.37.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.37.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.37.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.38.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.38.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.38.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.38.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.39.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.39.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.39.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.39.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.4.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.4.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.4.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.4.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.40.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.40.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.40.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.40.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.41.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.41.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.41.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.41.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.42.attn.norm_k.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.42.attn.norm_q.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.42.attn.to_out.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.42.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00003-of-00004.bin",
+    "single_transformer_blocks.43.attn.norm_k.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.43.attn.norm_q.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.43.attn.to_out.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.43.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.44.attn.norm_k.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.44.attn.norm_q.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.44.attn.to_out.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.44.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.45.attn.norm_k.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.45.attn.norm_q.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.45.attn.to_out.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.45.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.46.attn.norm_k.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.46.attn.norm_q.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.46.attn.to_out.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.46.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.47.attn.norm_k.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.47.attn.norm_q.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.47.attn.to_out.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.47.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00004-of-00004.bin",
+    "single_transformer_blocks.5.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.5.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.5.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.5.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.6.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.6.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.6.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.6.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.7.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.7.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.7.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.7.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.8.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.8.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.8.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.8.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.9.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.9.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.9.attn.to_out.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "single_transformer_blocks.9.attn.to_qkv_mlp_proj.weight": "diffusion_pytorch_model-00002-of-00004.bin",
+    "time_guidance_embed.guidance_embedder.linear_1.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "time_guidance_embed.guidance_embedder.linear_2.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "time_guidance_embed.timestep_embedder.linear_1.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "time_guidance_embed.timestep_embedder.linear_2.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.add_k_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.add_q_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.add_v_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.norm_added_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.norm_added_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.to_add_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.ff.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.ff.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.ff_context.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.0.ff_context.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.add_k_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.add_q_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.add_v_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.norm_added_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.norm_added_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.to_add_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.ff.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.ff.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.ff_context.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.1.ff_context.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.add_k_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.add_q_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.add_v_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.norm_added_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.norm_added_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.to_add_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.ff.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.ff.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.ff_context.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.2.ff_context.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.add_k_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.add_q_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.add_v_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.norm_added_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.norm_added_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.to_add_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.ff.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.ff.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.ff_context.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.3.ff_context.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.add_k_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.add_q_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.add_v_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.norm_added_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.norm_added_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.to_add_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.ff.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.ff.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.ff_context.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.4.ff_context.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.add_k_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.add_q_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.add_v_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.norm_added_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.norm_added_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.to_add_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.ff.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.ff.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.ff_context.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.5.ff_context.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.add_k_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.add_q_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.add_v_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.norm_added_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.norm_added_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.to_add_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.ff.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.ff.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.ff_context.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.6.ff_context.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.add_k_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.add_q_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.add_v_proj.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.norm_added_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.norm_added_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.to_add_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.ff.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.ff.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.ff_context.linear_in.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "transformer_blocks.7.ff_context.linear_out.weight": "diffusion_pytorch_model-00001-of-00004.bin",
+    "x_embedder.weight": "diffusion_pytorch_model-00001-of-00004.bin"
+  }
+}

upload_hf.py

@@ -0,0 +1,14 @@
+from huggingface_hub import HfApi
+
+repo_id = "AmdGoose/FLUX.2-dev-transformer-int8wo"
+
+api = HfApi()
+api.create_repo(repo_id=repo_id, repo_type="model", exist_ok=True)
+
+api.upload_folder(
+    repo_id=repo_id,
+    folder_path=".",
+    commit_message="Add INT8 weight-only quantized transformer for FLUX.2-dev",
+)
+
+print("Upload complete:", repo_id)