[Major] Release v0.1.4

Support 4-bit text encoder and per-layer CPU offloading, reducing FLUX's minimum memory requirement to just 4 GiB while maintaining a 2–3× speedup. Fix various issues related to resolution, LoRA, pin memory, and runtime stability. Check out the release notes for full details!

tests/sana/test_t2i.py

+import torch
+from diffusers import SanaPAGPipeline, SanaPipeline
+
+from nunchaku import NunchakuSanaTransformer2DModel
+
+
+def test_sana():
+    transformer = NunchakuSanaTransformer2DModel.from_pretrained("mit-han-lab/svdq-int4-sana-1600m")
+    pipe = SanaPipeline.from_pretrained(
+        "Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers",
+        transformer=transformer,
+        variant="bf16",
+        torch_dtype=torch.bfloat16,
+    ).to("cuda")
+    pipe.vae.to(torch.bfloat16)
+    pipe.text_encoder.to(torch.bfloat16)
+
+    prompt = "A cute 🐼 eating 🎋, ink drawing style"
+    image = pipe(
+        prompt=prompt,
+        height=1024,
+        width=1024,
+        guidance_scale=4.5,
+        num_inference_steps=20,
+        generator=torch.Generator().manual_seed(42),
+    ).images[0]
+
+    image.save("sana_1600m.png")
+
+
+def test_sana_pag():
+    transformer = NunchakuSanaTransformer2DModel.from_pretrained("mit-han-lab/svdq-int4-sana-1600m", pag_layers=8)
+    pipe = SanaPAGPipeline.from_pretrained(
+        "Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers",
+        transformer=transformer,
+        variant="bf16",
+        torch_dtype=torch.bfloat16,
+        pag_applied_layers="transformer_blocks.8",
+    ).to("cuda")
+    pipe._set_pag_attn_processor = lambda *args, **kwargs: None
+
+    pipe.text_encoder.to(torch.bfloat16)
+    pipe.vae.to(torch.bfloat16)
+
+    image = pipe(
+        prompt="A cute 🐼 eating 🎋, ink drawing style",
+        height=1024,
+        width=1024,
+        guidance_scale=5.0,
+        pag_scale=2.0,
+        num_inference_steps=20,
+    ).images[0]
+    image.save("sana_1600m_pag.png")

Block-Sparse-Attention @ 0d23f715

-Subproject commit 0d23f715690c5171fd93679de8afd149376db167

cutlass @ a75b4ac4

-Subproject commit a75b4ac483166189a45290783cb0a18af5ff0ea5

json @ 63258397

-Subproject commit 63258397761b3dd96dd171e5a5ad5aa915834c35

mio @ 8b6b7d87

-Subproject commit 8b6b7d878c89e81614d05edca7936de41ccdd2da

spdlog @ 27cb4c76

-Subproject commit 27cb4c76708608465c413f6d0e6b8d99a4d84302