Add Nova XL LCM OpenVINO INT4: model + 10 examples + README/REPORT/scripts

.gitattributes

@@ -33,3 +33,23 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+examples/01_hanfu_seed42_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/01_hanfu_seed42_8step_512.png filter=lfs diff=lfs merge=lfs -text
+examples/02_astronaut_seed43_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/02_astronaut_seed43_8step_512.png filter=lfs diff=lfs merge=lfs -text
+examples/03_taipei_seed44_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/03_taipei_seed44_8step_512.png filter=lfs diff=lfs merge=lfs -text
+examples/04_shiba_seed45_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/04_shiba_seed45_8step_512.png filter=lfs diff=lfs merge=lfs -text
+examples/05_ink_seed46_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/05_ink_seed46_8step_512.png filter=lfs diff=lfs merge=lfs -text
+examples/06_ghibli_style_seed47_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/06_ghibli_style_seed47_8step_512.png filter=lfs diff=lfs merge=lfs -text
+examples/07_makoto_shinkai_seed48_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/07_makoto_shinkai_seed48_8step_512.png filter=lfs diff=lfs merge=lfs -text
+examples/08_flat_vector_seed49_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/08_flat_vector_seed49_8step_512.png filter=lfs diff=lfs merge=lfs -text
+examples/09_retro_80s_seed50_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/09_retro_80s_seed50_8step_512.png filter=lfs diff=lfs merge=lfs -text
+examples/10_impasto_thick_seed51_8step_1024.png filter=lfs diff=lfs merge=lfs -text
+examples/10_impasto_thick_seed51_8step_512.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,115 @@
+---
+library_name: diffusers
+pipeline_tag: text-to-image
+base_model: sca255/nova_xl_lcm
+tags:
+- openvino
+- int4
+- optimum
+- lcm
+- anime
+- sdxl
+- text-to-image
+license: openrail++
+---
+
+# Nova XL LCM OpenVINO INT4 (optimum)
+
+Source: [sca255/nova_xl_lcm](https://huggingface.co/sca255/nova_xl_lcm) (SDXL LCM anime model, StableDiffusionXLPipeline).
+Recipe: `num_inference_steps=8, guidance_scale=1.5`, 1024x1024 (LCM distilled model).
+Converted with **optimum-intel 2.2.0 / optimum 2.3.0 / openvino 2026.4.0 / nncf 3.4.0**:
+1. `optimum-cli export openvino -m sca255/nova_xl_lcm --task text-to-image --library diffusers --weight-format fp16 ./nova_xl_lcm-ov-fp16`
+2. `OVQuantizer` NNCF weight-only INT4 (`bits=4, sym=False, group_size=128, fallback=adjust, ratio=1.0`) on unet + text_encoder + text_encoder_2, VAE INT8 default. See `quantize_int4.py`.
+- FP16 OV 6.5G → INT4 OV 2.0G (unet 1.5G, text_encoder_2 377M, text_encoder 81M, vae_dec 48M, vae_enc 33M, ~69%↓).
+
+## Usage
+```python
+from optimum.intel import OVDiffusionPipeline
+import torch
+pipe = OVDiffusionPipeline.from_pretrained("HelloSun/nova_xl_lcm-OpenVINO-INT4", compile=True)
+img = pipe("1girl, anime style, masterpiece, best quality",
+  height=1024, width=1024, num_inference_steps=8, guidance_scale=1.5,
+  generator=torch.Generator().manual_seed(42)).images[0]
+img.save("out.png")
+```
+Env: `diffusers==0.37.1 transformers==4.57.6 tokenizers==0.22.0 huggingface-hub==0.35.1 optimum==2.3.0 optimum-intel==2.2.0 openvino==2026.4.0 nncf==3.4.0 torch pillow psutil`. Full example: `inference_int4.py`.
+
+## 5 标准提示词 × 8-step (LCM recipe: 8 steps, guidance 1.5, 1024px, CPU)
+
+### 01_hanfu — seed 42 — 32.3s
+`1girl, red hanfu, intricate embroidery, golden phoenix headdress, soft night lighting, tiered pagoda background, lantern lights, anime style, masterpiece, best quality, ultra detailed`
+![01_hanfu](examples/01_hanfu_seed42_8step_1024.png)
+
+### 02_astronaut — seed 43 — 28.8s
+`1girl astronaut in lush jungle, cold color palette, detailed foliage, cinematic lighting, anime style, masterpiece, best quality, 8k`
+![02_astronaut](examples/02_astronaut_seed43_8step_1024.png)
+
+### 03_taipei — seed 44 — 34.2s
+`1girl in cyberpunk Taipei at night, heavy rain, neon signs 'TAIPEI' and '台北', wet asphalt reflections, crowded night market, anime style, masterpiece, best quality`
+![03_taipei](examples/03_taipei_seed44_8step_1024.png)
+
+### 04_shiba — seed 45 — 31.9s
+`cute shiba inu wearing tiny astronaut helmet, sunflower field under starry sky, dreamy illustration, vibrant colors, anime style, masterpiece, best quality`
+![04_shiba](examples/04_shiba_seed45_8step_1024.png)
+
+### 05_ink — seed 46 — 29.7s
+`traditional chinese ink wash landscape, misty mountains, small pagoda on cliff, cranes flying, minimalist, elegant, anime style, masterpiece, best quality`
+![05_ink](examples/05_ink_seed46_8step_1024.png)
+
+## 5 额外风格提示词 × 8-step (seeds 47-51)
+
+### 06_ghibli_style — seed 47 — 29.7s
+`studio ghibli style, 1girl in floating castle, magical clouds, whimsical atmosphere, detailed background, masterpiece, best quality`
+![06_ghibli_style](examples/06_ghibli_style_seed47_8step_1024.png)
+
+### 07_makoto_shinkai — seed 48 — 30.0s
+`makoto shinkai style, 1girl under comet sky, breathtaking lighting, detailed clouds, emotional atmosphere, anime style, masterpiece, 8k`
+![07_makoto_shinkai](examples/07_makoto_shinkai_seed48_8step_1024.png)
+
+### 08_flat_vector — seed 49 — 30.1s
+`flat vector illustration, 1girl in modern city, clean geometric shapes, bold colors, minimalist, commercial anime art style, behance trending`
+![08_flat_vector](examples/08_flat_vector_seed49_8step_1024.png)
+
+### 09_retro_80s — seed 50 — 29.9s
+`1980s retro anime, 1girl with neon city background, VHS aesthetic, synthwave colors, cel shaded, nostalgic anime style, masterpiece`
+![09_retro_80s](examples/09_retro_80s_seed50_8step_1024.png)
+
+### 10_impasto_thick — seed 51 — 29.5s
+`thick impasto anime concept art, 1girl with heavy brushstrokes, expressive texture, dramatic lighting, artstation masterpiece, detailed anime style`
+![10_impasto_thick](examples/10_impasto_thick_seed51_8step_1024.png)
+
+## 實驗數據 (CPU 真實能力)
+測試環境:
+- CPU: Intel Xeon Platinum 8559C, x86_64, 2 sockets x 48 cores x 2 threads = 192 vCPU, KVM
+- RAM 2.0Ti, OpenVINO CPU only, openvino 2026.4.0
+- 設定: num_inference_steps=8 (LCM recipe), guidance_scale=1.5, 1024x1024
+
+| 項目 | 數據 |
+|---|---|
+| load+compile | 12.5s |
+| 1024px 平均 | ~30.6s/張, 單步 ~3.67s |
+| 01 | 32.34s, mean 3.67s, steps [6.02,3.42,3.49,3.51,3.50,3.58,3.44,3.45] |
+| 02 | 28.75s, mean 3.39s |
+| 03 | 34.20s, mean 3.99s |
+| 04 | 31.90s, mean 3.72s |
+| 05 | 29.66s, mean 3.48s |
+| 06 | 29.74s, mean 3.49s |
+| 07 | 29.97s, mean 3.51s |
+| 08 | 30.13s, mean 3.54s |
+| 09 | 29.89s, mean 3.50s |
+| 10 | 29.52s, mean 3.46s |
+| 512px 對照 (同10組/seed) | 01:9.59s/1.14s, 02:9.63s/1.13s, 03:10.68s/1.26s, 04:9.84s/1.17s, 05:9.53s/1.13s, 06:9.83s/1.17s, 07:9.53s/1.13s, 08:9.77s/1.16s, 09:9.57s/1.13s, 10:9.53s/1.13s, 平均~9.75s/單步~1.15s (見examples/*_512.png) |
+| 模型大小 | FP16 6.5G → INT4 2.0G (~69%↓) |
+
+首圖首步較慢主因: text_encoder + text_encoder_2 + 首個 UNet warmup. VAE decode + text-encode 已含在 total 內. 純 CPU, OpenVINO 自動多線程.
+
+詳細見 `REPORT.md` + `examples/benchmark.json` + `examples/prompts.txt`.
+
+## 復現
+```
+pip install diffusers==0.37.1 transformers==4.57.6 tokenizers==0.22.0 huggingface-hub==0.35.1 optimum==2.3.0 optimum-intel==2.2.0 openvino==2026.4.0 nncf==3.4.0 torch pillow psutil
+optimum-cli export openvino -m sca255/nova_xl_lcm --task text-to-image --library diffusers --weight-format fp16 ./nova_xl_lcm-ov-fp16
+python quantize_int4.py
+python inference_int4.py
+python generate5.py
+```

REPORT.md

@@ -0,0 +1,68 @@
+# Nova XL LCM → OpenVINO INT4 (optimum) 轉換與 CPU 實測報告
+
+來源模型: sca255/nova_xl_lcm (SDXL LCM anime, StableDiffusionXLPipeline)
+官方範例: `num_inference_steps=8, guidance_scale=1.5` (LCM 8-step distilled, 1024x1024)
+轉換工具: diffusers==0.37.1 transformers==4.57.6 tokenizers==0.22.0 huggingface-hub==0.35.1 optimum-intel==2.2.0 optimum==2.3.0 openvino==2026.4.0 nncf==3.4.0 torch pillow psutil
+方法 (見 `quantize_int4.py`):
+1. `optimum-cli export openvino -m sca255/nova_xl_lcm --task text-to-image --library diffusers --weight-format fp16 ./nova_xl_lcm-ov-fp16`
+2. `OVQuantizer` + `OVPipelineQuantizationConfig(unet=INT4, text_encoder=INT4, text_encoder_2=INT4)`，NNCF weight-only:
+   `OVWeightQuantizationConfig(bits=4, sym=False, group_size=128, group_size_fallback="adjust", ratio=1.0)`，其餘預設 INT8 (`bits=8, sym=True`)
+   `ov_config=OVConfig(quantization_config=...)`，詳見 `quantize_int4.py`。範例推論見 `inference_int4.py`，批次生成見 `generate5.py`。
+
+## 模型檔案
+- FP16 OV: 6.5 GB (unet 4.8G + text_encoder_2 1.3G + text_encoder 236M + vae_decoder 95M + vae_encoder 66M)
+- INT4 OV: 2.0 GB (unet 1.5G + text_encoder_2 377M + text_encoder 81M + vae_decoder 48M + vae_encoder 33M)
+- 壓縮比: ~69% ↓ (6.5 → 2.0G)。上傳 repo: HelloSun/nova_xl_lcm-OpenVINO-INT4
+
+## 測試環境 CPU 真實能力
+- CPU: Intel(R) Xeon(R) Platinum 8559C, x86_64, 2 sockets x 48 cores, 2 threads/core = 192 vCPU
+- RAM: 2.0Ti total, OpenVINO device: CPU only, openvino 2026.4.0
+- 設定: num_inference_steps=8 (LCM recipe), guidance_scale=1.5, height=width=1024
+
+## 實測數據 (OV INT4, CPU)
+- load+compile: 12.5s
+- 1024x1024, 8 scheduler steps:
+  - 01_hanfu(42): total 32.34s, mean step 3.67s, steps [6.02,3.42,3.49,3.51,3.50,3.58,3.44,3.45]
+  - 02_astronaut(43): total 28.75s, mean step 3.39s, steps [3.54,3.30,3.33,3.39,3.30,3.39,3.49,3.39]
+  - 03_taipei(44): total 34.20s, mean step 3.99s, steps [4.21,4.20,4.07,3.93,3.91,3.86,3.90,4.01]
+  - 04_shiba(45): total 31.90s, mean step 3.72s, steps [4.08,4.32,3.77,3.52,3.61,3.58,3.53,3.50]
+  - 05_ink(46): total 29.66s, mean step 3.48s, steps [3.56,3.50,3.49,3.51,3.50,3.33,3.50,3.50]
+  - 06_ghibli_style(47): total 29.74s, mean step 3.49s, steps [3.62,3.50,3.42,3.49,3.51,3.48,3.51,3.41]
+  - 07_makoto_shinkai(48): total 29.97s, mean step 3.51s, steps [3.66,3.50,3.41,3.61,3.49,3.49,3.60,3.42]
+  - 08_flat_vector(49): total 30.13s, mean step 3.54s, steps [3.63,3.60,3.41,3.60,3.48,3.50,3.61,3.49]
+  - 09_retro_80s(50): total 29.89s, mean step 3.50s, steps [3.71,3.60,3.47,3.41,3.41,3.51,3.51,3.48]
+  - 10_impasto_thick(51): total 29.52s, mean step 3.46s, steps [3.55,3.48,3.49,3.44,3.38,3.49,3.49,3.42]
+  - 平均: ~30.61s/張, 單步 ~3.67s, 首圖首 step 較慢 (text_encoder + text_encoder_2 + warmup)
+- 512x512 對照 (同10組prompt/seed, examples/*_8step_512.png):
+  - 01_hanfu(42): total 9.59s, mean 1.14s
+  - 02_astronaut(43): total 9.63s, mean 1.13s
+  - 03_taipei(44): total 10.68s, mean 1.26s
+  - 04_shiba(45): total 9.84s, mean 1.17s
+  - 05_ink(46): total 9.53s, mean 1.13s
+  - 06_ghibli_style(47): total 9.83s, mean 1.17s
+  - 07_makoto_shinkai(48): total 9.53s, mean 1.13s
+  - 08_flat_vector(49): total 9.77s, mean 1.16s
+  - 09_retro_80s(50): total 9.57s, mean 1.13s
+  - 10_impasto_thick(51): total 9.53s, mean 1.13s
+  - 平均: ~9.75s/張, 單步 ~1.15s。1024 (~30.6s) 是 512 的 ~3.1x 時間
+- VAE decode + text-encode 已含在 total 內。純 CPU, OpenVINO 自動用滿多線程; 複現時用 `num_inference_steps=8, guidance_scale=1.5`
+
+## 10 組 prompt (seed 固定, examples/*.png)
+01 [seed42] 1girl, red hanfu, intricate embroidery, golden phoenix headdress, soft night lighting, tiered pagoda background, lantern lights, anime style, masterpiece, best quality, ultra detailed
+02 [seed43] 1girl astronaut in lush jungle, cold color palette, detailed foliage, cinematic lighting, anime style, masterpiece, best quality, 8k
+03 [seed44] 1girl in cyberpunk Taipei at night, heavy rain, neon signs 'TAIPEI' and '台北', wet asphalt reflections, crowded night market, anime style, masterpiece, best quality
+04 [seed45] cute shiba inu wearing tiny astronaut helmet, sunflower field under starry sky, dreamy illustration, vibrant colors, anime style, masterpiece, best quality
+05 [seed46] traditional chinese ink wash landscape, misty mountains, small pagoda on cliff, cranes flying, minimalist, elegant, anime style, masterpiece, best quality
+06 [seed47] studio ghibli style, 1girl in floating castle, magical clouds, whimsical atmosphere, detailed background, masterpiece, best quality
+07 [seed48] makoto shinkai style, 1girl under comet sky, breathtaking lighting, detailed clouds, emotional atmosphere, anime style, masterpiece, 8k
+08 [seed49] flat vector illustration, 1girl in modern city, clean geometric shapes, bold colors, minimalist, commercial anime art style, behance trending
+09 [seed50] 1980s retro anime, 1girl with neon city background, VHS aesthetic, synthwave colors, cel shaded, nostalgic anime style, masterpiece
+10 [seed51] thick impasto anime concept art, 1girl with heavy brushstrokes, expressive texture, dramatic lighting, artstation masterpiece, detailed anime style
+詳見 examples/prompts.txt, examples/benchmark.json, 圖片 examples/*_8step_1024.png + examples/*_8step_512.png (512對照組)
+
+## 復現
+pip install diffusers==0.37.1 transformers==4.57.6 tokenizers==0.22.0 huggingface-hub==0.35.1 optimum==2.3.0 optimum-intel==2.2.0 openvino==2026.4.0 nncf==3.4.0 torch pillow psutil
+optimum-cli export openvino -m sca255/nova_xl_lcm --task text-to-image --library diffusers --weight-format fp16 ./nova_xl_lcm-ov-fp16
+python quantize_int4.py
+python inference_int4.py
+python generate5.py

examples/benchmark.json

@@ -0,0 +1,436 @@
+{
+  "model": "/tmp/opencode/nova_xl_lcm-ov-int4",
+  "load_compile_s": 12.45,
+  "height": 1024,
+  "width": 1024,
+  "results": [
+    {
+      "id": "01_hanfu",
+      "prompt": "1girl, red hanfu, intricate embroidery, golden phoenix headdress, soft night lighting, tiered pagoda background, lantern lights, anime style, masterpiece, best quality, ultra detailed",
+      "seed": 42,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        6.022,
+        3.42,
+        3.492,
+        3.507,
+        3.495,
+        3.583,
+        3.441,
+        3.453
+      ],
+      "total_s": 32.34,
+      "mean_step_s": 3.802,
+      "image": "/tmp/opencode/nova_xl_examples/01_hanfu_seed42_8step_1024.png",
+      "timestamp": "2026-09-30T08:00:16.487929",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/01_hanfu_seed42_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 9.59,
+        "mean_step_s": 1.137,
+        "step_times_s": [
+          1.397,
+          1.101,
+          1.108,
+          1.1,
+          1.097,
+          1.097,
+          1.095,
+          1.099
+        ],
+        "timestamp": "2026-09-30T08:00:26.359003"
+      }
+    },
+    {
+      "id": "02_astronaut",
+      "prompt": "1girl astronaut in lush jungle, cold color palette, detailed foliage, cinematic lighting, anime style, masterpiece, best quality, 8k",
+      "seed": 43,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        3.543,
+        3.3,
+        3.326,
+        3.394,
+        3.301,
+        3.391,
+        3.492,
+        3.393
+      ],
+      "total_s": 28.75,
+      "mean_step_s": 3.393,
+      "image": "/tmp/opencode/nova_xl_examples/02_astronaut_seed43_8step_1024.png",
+      "timestamp": "2026-09-30T08:00:55.583287",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/02_astronaut_seed43_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 9.63,
+        "mean_step_s": 1.128,
+        "step_times_s": [
+          1.23,
+          1.108,
+          1.189,
+          1.114,
+          1.087,
+          1.024,
+          1.072,
+          1.204
+        ],
+        "timestamp": "2026-09-30T08:01:05.465534"
+      }
+    },
+    {
+      "id": "03_taipei",
+      "prompt": "1girl in cyberpunk Taipei at night, heavy rain, neon signs 'TAIPEI' and '台北', wet asphalt reflections, crowded night market, anime style, masterpiece, best quality",
+      "seed": 44,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        4.209,
+        4.201,
+        4.073,
+        3.93,
+        3.913,
+        3.858,
+        3.899,
+        4.007
+      ],
+      "total_s": 34.2,
+      "mean_step_s": 4.011,
+      "image": "/tmp/opencode/nova_xl_examples/03_taipei_seed44_8step_1024.png",
+      "timestamp": "2026-09-30T08:01:40.307647",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/03_taipei_seed44_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 10.68,
+        "mean_step_s": 1.259,
+        "step_times_s": [
+          1.48,
+          1.197,
+          1.125,
+          1.289,
+          1.281,
+          1.301,
+          1.131,
+          1.269
+        ],
+        "timestamp": "2026-09-30T08:01:51.278539"
+      }
+    },
+    {
+      "id": "04_shiba",
+      "prompt": "cute shiba inu wearing tiny astronaut helmet, sunflower field under starry sky, dreamy illustration, vibrant colors, anime style, masterpiece, best quality",
+      "seed": 45,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        4.077,
+        4.32,
+        3.774,
+        3.516,
+        3.614,
+        3.577,
+        3.528,
+        3.495
+      ],
+      "total_s": 31.9,
+      "mean_step_s": 3.738,
+      "image": "/tmp/opencode/nova_xl_examples/04_shiba_seed45_8step_1024.png",
+      "timestamp": "2026-09-30T08:02:23.840968",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/04_shiba_seed45_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 9.84,
+        "mean_step_s": 1.17,
+        "step_times_s": [
+          1.352,
+          1.114,
+          1.196,
+          1.178,
+          1.127,
+          1.191,
+          1.095,
+          1.105
+        ],
+        "timestamp": "2026-09-30T08:02:33.960585"
+      }
+    },
+    {
+      "id": "05_ink",
+      "prompt": "traditional chinese ink wash landscape, misty mountains, small pagoda on cliff, cranes flying, minimalist, elegant, anime style, masterpiece, best quality",
+      "seed": 46,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        3.559,
+        3.495,
+        3.486,
+        3.512,
+        3.496,
+        3.33,
+        3.498,
+        3.496
+      ],
+      "total_s": 29.66,
+      "mean_step_s": 3.484,
+      "image": "/tmp/opencode/nova_xl_examples/05_ink_seed46_8step_1024.png",
+      "timestamp": "2026-09-30T08:03:04.275206",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/05_ink_seed46_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 9.53,
+        "mean_step_s": 1.131,
+        "step_times_s": [
+          1.238,
+          1.181,
+          1.115,
+          1.108,
+          1.18,
+          1.103,
+          1.095,
+          1.024
+        ],
+        "timestamp": "2026-09-30T08:03:14.079835"
+      }
+    },
+    {
+      "id": "06_ghibli_style",
+      "prompt": "studio ghibli style, 1girl in floating castle, magical clouds, whimsical atmosphere, detailed background, masterpiece, best quality",
+      "seed": 47,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        3.617,
+        3.501,
+        3.42,
+        3.491,
+        3.505,
+        3.482,
+        3.512,
+        3.412
+      ],
+      "total_s": 29.74,
+      "mean_step_s": 3.493,
+      "image": "/tmp/opencode/nova_xl_examples/06_ghibli_style_seed47_8step_1024.png",
+      "timestamp": "2026-09-30T08:03:44.580401",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/06_ghibli_style_seed47_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 9.83,
+        "mean_step_s": 1.167,
+        "step_times_s": [
+          1.249,
+          1.189,
+          1.114,
+          1.195,
+          1.178,
+          1.1,
+          1.115,
+          1.2
+        ],
+        "timestamp": "2026-09-30T08:03:54.698464"
+      }
+    },
+    {
+      "id": "07_makoto_shinkai",
+      "prompt": "makoto shinkai style, 1girl under comet sky, breathtaking lighting, detailed clouds, emotional atmosphere, anime style, masterpiece, 8k",
+      "seed": 48,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        3.664,
+        3.499,
+        3.412,
+        3.607,
+        3.488,
+        3.486,
+        3.598,
+        3.424
+      ],
+      "total_s": 29.97,
+      "mean_step_s": 3.522,
+      "image": "/tmp/opencode/nova_xl_examples/07_makoto_shinkai_seed48_8step_1024.png",
+      "timestamp": "2026-09-30T08:04:25.367477",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/07_makoto_shinkai_seed48_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 9.53,
+        "mean_step_s": 1.13,
+        "step_times_s": [
+          1.23,
+          1.117,
+          1.112,
+          1.174,
+          1.021,
+          1.098,
+          1.181,
+          1.108
+        ],
+        "timestamp": "2026-09-30T08:04:35.186794"
+      }
+    },
+    {
+      "id": "08_flat_vector",
+      "prompt": "flat vector illustration, 1girl in modern city, clean geometric shapes, bold colors, minimalist, commercial anime art style, behance trending",
+      "seed": 49,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        3.633,
+        3.598,
+        3.412,
+        3.604,
+        3.484,
+        3.503,
+        3.608,
+        3.487
+      ],
+      "total_s": 30.13,
+      "mean_step_s": 3.541,
+      "image": "/tmp/opencode/nova_xl_examples/08_flat_vector_seed49_8step_1024.png",
+      "timestamp": "2026-09-30T08:05:05.802048",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/08_flat_vector_seed49_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 9.77,
+        "mean_step_s": 1.16,
+        "step_times_s": [
+          1.274,
+          1.182,
+          1.124,
+          1.187,
+          1.108,
+          1.182,
+          1.109,
+          1.114
+        ],
+        "timestamp": "2026-09-30T08:05:15.884667"
+      }
+    },
+    {
+      "id": "09_retro_80s",
+      "prompt": "1980s retro anime, 1girl with neon city background, VHS aesthetic, synthwave colors, cel shaded, nostalgic anime style, masterpiece",
+      "seed": 50,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        3.713,
+        3.598,
+        3.475,
+        3.406,
+        3.407,
+        3.507,
+        3.51,
+        3.476
+      ],
+      "total_s": 29.89,
+      "mean_step_s": 3.511,
+      "image": "/tmp/opencode/nova_xl_examples/09_retro_80s_seed50_8step_1024.png",
+      "timestamp": "2026-09-30T08:05:46.392780",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/09_retro_80s_seed50_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 9.57,
+        "mean_step_s": 1.133,
+        "step_times_s": [
+          1.262,
+          1.108,
+          1.175,
+          1.109,
+          1.102,
+          1.103,
+          1.105,
+          1.1
+        ],
+        "timestamp": "2026-09-30T08:05:56.288080"
+      }
+    },
+    {
+      "id": "10_impasto_thick",
+      "prompt": "thick impasto anime concept art, 1girl with heavy brushstrokes, expressive texture, dramatic lighting, artstation masterpiece, detailed anime style",
+      "seed": 51,
+      "height": 1024,
+      "width": 1024,
+      "num_inference_steps_param": 8,
+      "guidance_scale": 1.5,
+      "scheduler_steps_observed": 8,
+      "step_times_s": [
+        3.546,
+        3.482,
+        3.494,
+        3.444,
+        3.383,
+        3.493,
+        3.486,
+        3.417
+      ],
+      "total_s": 29.52,
+      "mean_step_s": 3.468,
+      "image": "/tmp/opencode/nova_xl_examples/10_impasto_thick_seed51_8step_1024.png",
+      "timestamp": "2026-09-30T08:06:26.353283",
+      "control_512": {
+        "image": "/tmp/opencode/nova_xl_examples/10_impasto_thick_seed51_8step_512.png",
+        "height": 512,
+        "width": 512,
+        "total_s": 9.53,
+        "mean_step_s": 1.13,
+        "step_times_s": [
+          1.222,
+          1.121,
+          1.105,
+          1.094,
+          1.1,
+          1.107,
+          1.173,
+          1.121
+        ],
+        "timestamp": "2026-09-30T08:06:36.175755"
+      }
+    }
+  ],
+  "sysinfo": {
+    "cpu_model": "Intel(R) Xeon(R) Platinum 8559C",
+    "os_cpu_count": 192,
+    "lscpu": "Architecture:                            x86_64\nCPU op-mode(s):                          32-bit, 64-bit\nAddress sizes:                           46 bits physical, 48 bits virtual\nByte Order:                              Little Endian\nCPU(s):                                  192\nOn-line CPU(s) list:                     0-191\nVendor ID:                               GenuineIntel\nModel name:                              Intel(R) Xeon(R) Platinum 8559C\nCPU family:                              6\nModel:                                   207\nThread(s) per core:                      2\nCore(s) per socket:                      48\nSocket(s):                               2\nStepping:                                2\nBogoMIPS:                                4800.00\nFlags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd ida arat avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq rdpid cldemote movdiri movdir64b md_clear serialize amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities\nHypervisor vendor:                       KVM\nVirtualization type:                     full\nL1d cache:                               4.5 MiB (96 instances)\nL1i cache:                               3 MiB (96 instances)\nL2 cache:                                192 MiB (96 instances)\nL3 cache:                                640 MiB (2 instances)\nNUMA node(s):                            2\nNUMA node0 CPU(s):                       0-47,96-143\nNUMA node1 CPU(s):                       48-95,144-191\nVulnerability Gather data sampling:      Not affected\nVulnerability Ghostwrite:                Not affected\nVulnerability Indirect target selection: Not affected\nVulnerability Itlb multihit:             Not affected\nVulnerability L1tf:                      Not affected\nVulnerability Mds:                       Not affected\nVulnerability Meltdown:                  Not affected\nVulnerability Mmio stale data:           Not affected\nVulnerability Old microcode:             Not affected\nVulnerability Reg file data sampling:    Not affected\nVulnerability Retbleed:                  Not affected\nVulnerability Spec rstack overflow:      Not affected\nVulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl\nVulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization\nVulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S\nVulnerability Srbds:                     Not affected\nVulnerability Tsa:                       Not affected\nVulnerability Tsx async abort:           Not affected\nVulnerability Vmscape:                   Not affected\n",
+    "openvino_version": "2026.4.0-22959-99c81491cc3-releases/2026/4",
+    "settings": "num_inference_steps=8, guidance_scale=1.5, 1024 main + 512 control",
+    "source_model": "sca255/nova_xl_lcm (SDXL LCM anime)"
+  }
+}

examples/prompts.txt

@@ -0,0 +1,20 @@
+01_hanfu	seed=42	1girl, red hanfu, intricate embroidery, golden phoenix headdress, soft night lighting, tiered pagoda background, lantern lights, anime style, masterpiece, best quality, ultra detailed	2026-09-30T08:06:36.210047
+
+02_astronaut	seed=43	1girl astronaut in lush jungle, cold color palette, detailed foliage, cinematic lighting, anime style, masterpiece, best quality, 8k	2026-09-30T08:06:36.210057
+
+03_taipei	seed=44	1girl in cyberpunk Taipei at night, heavy rain, neon signs 'TAIPEI' and '台北', wet asphalt reflections, crowded night market, anime style, masterpiece, best quality	2026-09-30T08:06:36.210059
+
+04_shiba	seed=45	cute shiba inu wearing tiny astronaut helmet, sunflower field under starry sky, dreamy illustration, vibrant colors, anime style, masterpiece, best quality	2026-09-30T08:06:36.210065
+
+05_ink	seed=46	traditional chinese ink wash landscape, misty mountains, small pagoda on cliff, cranes flying, minimalist, elegant, anime style, masterpiece, best quality	2026-09-30T08:06:36.210067
+
+06_ghibli_style	seed=47	studio ghibli style, 1girl in floating castle, magical clouds, whimsical atmosphere, detailed background, masterpiece, best quality	2026-09-30T08:06:36.210068
+
+07_makoto_shinkai	seed=48	makoto shinkai style, 1girl under comet sky, breathtaking lighting, detailed clouds, emotional atmosphere, anime style, masterpiece, 8k	2026-09-30T08:06:36.210070
+
+08_flat_vector	seed=49	flat vector illustration, 1girl in modern city, clean geometric shapes, bold colors, minimalist, commercial anime art style, behance trending	2026-09-30T08:06:36.210072
+
+09_retro_80s	seed=50	1980s retro anime, 1girl with neon city background, VHS aesthetic, synthwave colors, cel shaded, nostalgic anime style, masterpiece	2026-09-30T08:06:36.210073
+
+10_impasto_thick	seed=51	thick impasto anime concept art, 1girl with heavy brushstrokes, expressive texture, dramatic lighting, artstation masterpiece, detailed anime style	2026-09-30T08:06:36.210075
+

generate5.py

@@ -0,0 +1,150 @@
+#!/usr/bin/env python3
+"""Generate 10 prompts with Nova XL LCM OV INT4 + benchmark.
+LCM recipe: num_inference_steps=8, guidance_scale=1.5, 1024x1024 (+512 control).
+10 fixed seeds 42-51, saved to examples/.
+Per-step timing via callback. Plus 512px control group.
+"""
+import json, time, os, gc, subprocess
+from pathlib import Path
+
+import torch
+from optimum.intel import OVDiffusionPipeline
+
+MODEL_ID = "HelloSun/nova_xl_lcm-OpenVINO-INT4"  # or local ./nova_xl_lcm-ov-int4
+OUT_DIR = Path("examples")
+OUT_DIR.mkdir(exist_ok=True)
+
+PROMPTS = [
+    ("01_hanfu", 42, "1girl, red hanfu, intricate embroidery, golden phoenix headdress, soft night lighting, tiered pagoda background, lantern lights, anime style, masterpiece, best quality, ultra detailed"),
+    ("02_astronaut", 43, "1girl astronaut in lush jungle, cold color palette, detailed foliage, cinematic lighting, anime style, masterpiece, best quality, 8k"),
+    ("03_taipei", 44, "1girl in cyberpunk Taipei at night, heavy rain, neon signs 'TAIPEI' and '台北', wet asphalt reflections, crowded night market, anime style, masterpiece, best quality"),
+    ("04_shiba", 45, "cute shiba inu wearing tiny astronaut helmet, sunflower field under starry sky, dreamy illustration, vibrant colors, anime style, masterpiece, best quality"),
+    ("05_ink", 46, "traditional chinese ink wash landscape, misty mountains, small pagoda on cliff, cranes flying, minimalist, elegant, anime style, masterpiece, best quality"),
+    ("06_ghibli_style", 47, "studio ghibli style, 1girl in floating castle, magical clouds, whimsical atmosphere, detailed background, masterpiece, best quality"),
+    ("07_makoto_shinkai", 48, "makoto shinkai style, 1girl under comet sky, breathtaking lighting, detailed clouds, emotional atmosphere, anime style, masterpiece, 8k"),
+    ("08_flat_vector", 49, "flat vector illustration, 1girl in modern city, clean geometric shapes, bold colors, minimalist, commercial anime art style, behance trending"),
+    ("09_retro_80s", 50, "1980s retro anime, 1girl with neon city background, VHS aesthetic, synthwave colors, cel shaded, nostalgic anime style, masterpiece"),
+    ("10_impasto_thick", 51, "thick impasto anime concept art, 1girl with heavy brushstrokes, expressive texture, dramatic lighting, artstation masterpiece, detailed anime style"),
+]
+
+HEIGHT = WIDTH = 1024
+NUM_STEPS = 8
+GUIDANCE = 1.5
+CONTROL_RES = 512
+
+
+def main():
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", default=MODEL_ID)
+    parser.add_argument("--outdir", default="examples")
+    args = parser.parse_args()
+    outdir = Path(args.outdir)
+    outdir.mkdir(parents=True, exist_ok=True)
+
+    print("=== Loading OV INT4 pipeline ===")
+    t_load0 = time.time()
+    pipe = OVDiffusionPipeline.from_pretrained(args.model, compile=True)
+    t_load = time.time() - t_load0
+    print(f"load+compile: {t_load:.1f}s")
+
+    results = []
+    for pid, seed, prompt in PROMPTS:
+        print(f"\n=== {pid} seed={seed} ===")
+        print(prompt)
+        gc.collect()
+
+        step_times = []
+        t_last = [time.time()]
+
+        def cb(pipe, step_idx, timestep, callback_kwargs):
+            now = time.time()
+            step_times.append(now - t_last[0])
+            t_last[0] = now
+            return callback_kwargs
+
+        t0 = time.time()
+        out = pipe(
+            prompt=prompt,
+            height=HEIGHT,
+            width=WIDTH,
+            num_inference_steps=NUM_STEPS,
+            guidance_scale=GUIDANCE,
+            generator=torch.Generator().manual_seed(seed),
+            callback_on_step_end=cb,
+            callback_on_step_end_tensor_inputs=["latents"],
+        )
+        total = time.time() - t0
+        img = out.images[0]
+        save_path = outdir / f"{pid}_seed{seed}_{NUM_STEPS}step_1024.png"
+        img.save(save_path)
+        print(f"saved {save_path} size={img.size} total={total:.1f}s")
+        print(f"step times: {[f'{x:.2f}' for x in step_times]}")
+        entry = {
+            "id": pid, "prompt": prompt, "seed": seed,
+            "height": HEIGHT, "width": WIDTH,
+            "num_inference_steps_param": NUM_STEPS,
+            "guidance_scale": GUIDANCE,
+            "scheduler_steps_observed": len(step_times),
+            "step_times_s": [round(x, 3) for x in step_times],
+            "total_s": round(total, 2),
+            "mean_step_s": round(sum(step_times) / len(step_times), 3) if step_times else None,
+            "image": str(save_path),
+        }
+        # 512px control
+        gc.collect()
+        c_times = []
+        c_last = [time.time()]
+
+        def cb2(pipe, step_idx, timestep, callback_kwargs):
+            now = time.time()
+            c_times.append(now - c_last[0])
+            c_last[0] = now
+            return callback_kwargs
+
+        tc0 = time.time()
+        cout = pipe(
+            prompt=prompt, height=CONTROL_RES, width=CONTROL_RES,
+            num_inference_steps=NUM_STEPS, guidance_scale=GUIDANCE,
+            generator=torch.Generator().manual_seed(seed),
+            callback_on_step_end=cb2,
+            callback_on_step_end_tensor_inputs=["latents"],
+        )
+        c_total = time.time() - tc0
+        cimg = cout.images[0]
+        c_path = outdir / f"{pid}_seed{seed}_{NUM_STEPS}step_512.png"
+        cimg.save(c_path)
+        print(f"control 512 saved {c_path} total={c_total:.1f}s mean={sum(c_times)/len(c_times):.2f}s")
+        entry["control_512"] = {
+            "image": str(c_path), "height": CONTROL_RES, "width": CONTROL_RES,
+            "total_s": round(c_total, 2),
+            "mean_step_s": round(sum(c_times) / len(c_times), 3) if c_times else None,
+            "step_times_s": [round(x, 3) for x in c_times],
+        }
+        results.append(entry)
+
+    with open(outdir / "benchmark.json", "w") as f:
+        import openvino as ov
+        try:
+            lscpu = subprocess.check_output("lscpu", shell=True, text=True)
+        except Exception as e:
+            lscpu = str(e)
+        sysinfo = {
+            "cpu_model": "Intel(R) Xeon(R) Platinum 8559C",
+            "os_cpu_count": os.cpu_count(),
+            "lscpu": lscpu,
+            "openvino_version": ov.__version__,
+            "settings": f"num_inference_steps={NUM_STEPS}, guidance_scale={GUIDANCE}, 1024 main + 512 control",
+            "source_model": "sca255/nova_xl_lcm (SDXL LCM anime)",
+        }
+        json.dump({"model": args.model, "load_compile_s": round(t_load, 2),
+                   "height": HEIGHT, "width": WIDTH, "results": results,
+                   "sysinfo": sysinfo}, f, indent=2, ensure_ascii=False)
+    with open(outdir / "prompts.txt", "w") as f:
+        for pid, seed, prompt in PROMPTS:
+            f.write(f"{pid}\tseed={seed}\t{prompt}\n\n")
+    print(f"\nDone. See {outdir}/benchmark.json")
+
+
+if __name__ == "__main__":
+    main()

inference_int4.py

@@ -0,0 +1,29 @@
+"""Nova XL LCM OpenVINO INT4 inference example (optimum-intel).
+Requires: diffusers==0.37.1, transformers==4.57.6, tokenizers==0.22.0,
+  huggingface-hub==0.35.1, optimum-intel==2.2.0, optimum==2.3.0,
+  openvino==2026.4.0, nncf==3.4.0, torch, pillow, psutil
+Source model: sca255/nova_xl_lcm (SDXL LCM anime).
+Recipe: num_inference_steps=8, guidance_scale=1.5, 1024x1024.
+"""
+import torch
+from optimum.intel import OVDiffusionPipeline
+
+MODEL_ID = "HelloSun/nova_xl_lcm-OpenVINO-INT4"  # or local ./nova_xl_lcm-ov-int4
+
+
+def main():
+    pipe = OVDiffusionPipeline.from_pretrained(MODEL_ID, compile=True)
+    prompt = "1girl, anime style, masterpiece, best quality, detailed face, beautiful eyes"
+    image = pipe(
+        prompt=prompt,
+        height=1024, width=1024,
+        num_inference_steps=8,
+        guidance_scale=1.5,
+        generator=torch.Generator().manual_seed(42),
+    ).images[0]
+    image.save("nova_xl_lcm_ov_int4.png")
+    print("saved nova_xl_lcm_ov_int4.png")
+
+
+if __name__ == "__main__":
+    main()

model_index.json

@@ -0,0 +1,42 @@
+{
+  "_class_name": "StableDiffusionXLPipeline",
+  "_diffusers_version": "0.37.1",
+  "_name_or_path": "/tmp/opencode/nova_xl_lcm_pt_v2",
+  "feature_extractor": [
+    null,
+    null
+  ],
+  "force_zeros_for_empty_prompt": true,
+  "image_encoder": [
+    null,
+    null
+  ],
+  "scheduler": [
+    "diffusers",
+    "LCMScheduler"
+  ],
+  "text_encoder": [
+    "transformers",
+    "CLIPTextModel"
+  ],
+  "text_encoder_2": [
+    "transformers",
+    "CLIPTextModelWithProjection"
+  ],
+  "tokenizer": [
+    "transformers",
+    "CLIPTokenizer"
+  ],
+  "tokenizer_2": [
+    "transformers",
+    "CLIPTokenizer"
+  ],
+  "unet": [
+    "diffusers",
+    "UNet2DConditionModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKL"
+  ]
+}

openvino_config.json

@@ -0,0 +1,90 @@
+{
+  "dtype": "int4_int4_int4",
+  "input_info": null,
+  "optimum_version": "2.3.0",
+  "output_attentions": false,
+  "quantization_config": {
+    "_dataset_kwargs": {},
+    "dataset": null,
+    "default_config": {
+      "quant_method": "default"
+    },
+    "ignored_scope": null,
+    "num_samples": null,
+    "processor": null,
+    "quantization_configs": {
+      "text_encoder": {
+        "_dataset_kwargs": {},
+        "all_layers": null,
+        "backup_precision": null,
+        "bits": 4,
+        "dataset": null,
+        "dq_group_size": null,
+        "dtype": "int4",
+        "gptq": null,
+        "group_size": 128,
+        "group_size_fallback": "adjust",
+        "ignored_scope": null,
+        "lora_correction": null,
+        "num_samples": null,
+        "processor": null,
+        "quant_method": "default",
+        "ratio": 1.0,
+        "scale_estimation": null,
+        "sensitivity_metric": null,
+        "statistics_path": null,
+        "sym": false,
+        "tokenizer": null
+      },
+      "text_encoder_2": {
+        "_dataset_kwargs": {},
+        "all_layers": null,
+        "backup_precision": null,
+        "bits": 4,
+        "dataset": null,
+        "dq_group_size": null,
+        "dtype": "int4",
+        "gptq": null,
+        "group_size": 128,
+        "group_size_fallback": "adjust",
+        "ignored_scope": null,
+        "lora_correction": null,
+        "num_samples": null,
+        "processor": null,
+        "quant_method": "default",
+        "ratio": 1.0,
+        "scale_estimation": null,
+        "sensitivity_metric": null,
+        "statistics_path": null,
+        "sym": false,
+        "tokenizer": null
+      },
+      "unet": {
+        "_dataset_kwargs": {},
+        "all_layers": null,
+        "backup_precision": null,
+        "bits": 4,
+        "dataset": null,
+        "dq_group_size": null,
+        "dtype": "int4",
+        "gptq": null,
+        "group_size": 128,
+        "group_size_fallback": "adjust",
+        "ignored_scope": null,
+        "lora_correction": null,
+        "num_samples": null,
+        "processor": null,
+        "quant_method": "default",
+        "ratio": 1.0,
+        "scale_estimation": null,
+        "sensitivity_metric": null,
+        "statistics_path": null,
+        "sym": false,
+        "tokenizer": null
+      }
+    },
+    "tokenizer": null
+  },
+  "save_onnx_model": false,
+  "transformers_version": "4.57.6"
+}

quantize_int4.py

@@ -0,0 +1,35 @@
+"""Quantize Nova XL LCM OV FP16 -> INT4 using optimum (NNCF weight-only).
+Source: sca255/nova_xl_lcm (SDXL LCM anime, example: 8 steps, guidance 1.5)
+FP16 export: optimum-cli export openvino -m sca255/nova_xl_lcm --task text-to-image --library diffusers --weight-format fp16 ./nova_xl_lcm-ov-fp16
+INT4: unet + text_encoder + text_encoder_2 -> OVWeightQuantizationConfig(bits=4,sym=False,group_size=128,group_size_fallback="adjust",ratio=1.0), rest default INT8.
+"""
+from optimum.intel import OVStableDiffusionPipeline, OVQuantizer
+from optimum.intel.openvino.configuration import (
+    OVConfig, OVWeightQuantizationConfig, OVPipelineQuantizationConfig,
+)
+
+fp16_dir = "/tmp/opencode/nova_xl_lcm-ov-fp16-cli"
+int4_dir = "/tmp/opencode/nova_xl_lcm-ov-int4"
+
+int4_config = OVWeightQuantizationConfig(
+    bits=4, sym=False, group_size=128, group_size_fallback="adjust", ratio=1.0,
+)
+
+pipeline_config = OVPipelineQuantizationConfig(
+    quantization_configs={
+        "unet": int4_config,
+        "text_encoder": int4_config,
+        "text_encoder_2": int4_config,
+    },
+    default_config=OVWeightQuantizationConfig(bits=8, sym=True),
+)
+
+print(f"Loading FP16 OV pipeline from {fp16_dir} ...")
+pipe = OVStableDiffusionPipeline.from_pretrained(fp16_dir)
+print(f"Components: {list(pipe.components.keys())}")
+
+quantizer = OVQuantizer.from_pretrained(pipe)
+ov_config = OVConfig(quantization_config=pipeline_config)
+print("Quantizing unet+text_encoder+text_encoder_2 to INT4 (NNCF, optimum)...")
+quantizer.quantize(save_directory=int4_dir, ov_config=ov_config)
+print(f"Saved INT4 model to {int4_dir}")

scheduler/scheduler_config.json

@@ -0,0 +1,24 @@
+{
+  "_class_name": "LCMScheduler",
+  "_diffusers_version": "0.37.1",
+  "beta_end": 0.012,
+  "beta_schedule": "scaled_linear",
+  "beta_start": 0.00085,
+  "clip_sample": false,
+  "clip_sample_range": 1.0,
+  "dynamic_thresholding_ratio": 0.995,
+  "interpolation_type": "linear",
+  "num_train_timesteps": 1000,
+  "original_inference_steps": 50,
+  "prediction_type": "epsilon",
+  "rescale_betas_zero_snr": false,
+  "sample_max_value": 1.0,
+  "set_alpha_to_one": false,
+  "skip_prk_steps": true,
+  "steps_offset": 1,
+  "thresholding": false,
+  "timestep_scaling": 10.0,
+  "timestep_spacing": "leading",
+  "trained_betas": null,
+  "use_karras_sigmas": false
+}

text_encoder/config.json

@@ -0,0 +1,26 @@
+{
+  "architectures": [
+    "CLIPTextModel"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "dropout": 0.0,
+  "dtype": "float16",
+  "eos_token_id": 2,
+  "export_model_type": "clip-text",
+  "hidden_act": "quick_gelu",
+  "hidden_size": 768,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 77,
+  "model_type": "clip_text_model",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "output_hidden_states": true,
+  "pad_token_id": 1,
+  "projection_dim": 768,
+  "transformers_version": "4.57.6",
+  "vocab_size": 49408
+}

text_encoder/openvino_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8ec241cbaf82a6d20c4917ba4861584ab3bf6b6813adcc0cd5df76bcc66fdef
+size 83658537

text_encoder_2/config.json

@@ -0,0 +1,26 @@
+{
+  "architectures": [
+    "CLIPTextModelWithProjection"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "dropout": 0.0,
+  "dtype": "float16",
+  "eos_token_id": 2,
+  "export_model_type": "clip-text-with-projection",
+  "hidden_act": "gelu",
+  "hidden_size": 1280,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 5120,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 77,
+  "model_type": "clip_text_model",
+  "num_attention_heads": 20,
+  "num_hidden_layers": 32,
+  "output_hidden_states": true,
+  "pad_token_id": 1,
+  "projection_dim": 1280,
+  "transformers_version": "4.57.6",
+  "vocab_size": 49408
+}

text_encoder_2/openvino_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8fba9fc1a9cab644d659e84c1a9253c6d1e202b37f89f027ec3d25ce430bd2d
+size 393063207

tokenizer/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer/tokenizer_config.json

@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "49406": {
+      "content": "<|startoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "49407": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<|startoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "do_lower_case": true,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 77,
+  "pad_token": "<|endoftext|>",
+  "tokenizer_class": "CLIPTokenizer",
+  "unk_token": "<|endoftext|>"
+}

tokenizer_2/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "!",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_2/tokenizer_config.json

@@ -0,0 +1,39 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "!",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "49406": {
+      "content": "<|startoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "49407": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<|startoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "do_lower_case": true,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 77,
+  "pad_token": "!",
+  "tokenizer_class": "CLIPTokenizer",
+  "unk_token": "<|endoftext|>"
+}

unet/config.json

@@ -0,0 +1,73 @@
+{
+  "_class_name": "UNet2DConditionModel",
+  "_diffusers_version": "0.37.1",
+  "_name_or_path": "/tmp/opencode/nova_xl_lcm_pt_v2/unet",
+  "act_fn": "silu",
+  "addition_embed_type": "text_time",
+  "addition_embed_type_num_heads": 64,
+  "addition_time_embed_dim": 256,
+  "attention_head_dim": [
+    5,
+    10,
+    20
+  ],
+  "attention_type": "default",
+  "block_out_channels": [
+    320,
+    640,
+    1280
+  ],
+  "center_input_sample": false,
+  "class_embed_type": null,
+  "class_embeddings_concat": false,
+  "conv_in_kernel": 3,
+  "conv_out_kernel": 3,
+  "cross_attention_dim": 2048,
+  "cross_attention_norm": null,
+  "down_block_types": [
+    "DownBlock2D",
+    "CrossAttnDownBlock2D",
+    "CrossAttnDownBlock2D"
+  ],
+  "downsample_padding": 1,
+  "dropout": 0.0,
+  "dual_cross_attention": false,
+  "encoder_hid_dim": null,
+  "encoder_hid_dim_type": null,
+  "flip_sin_to_cos": true,
+  "freq_shift": 0,
+  "in_channels": 4,
+  "layers_per_block": 2,
+  "mid_block_only_cross_attention": null,
+  "mid_block_scale_factor": 1,
+  "mid_block_type": "UNetMidBlock2DCrossAttn",
+  "norm_eps": 1e-05,
+  "norm_num_groups": 32,
+  "num_attention_heads": null,
+  "num_class_embeds": null,
+  "only_cross_attention": false,
+  "out_channels": 4,
+  "projection_class_embeddings_input_dim": 2816,
+  "resnet_out_scale_factor": 1.0,
+  "resnet_skip_time_act": false,
+  "resnet_time_scale_shift": "default",
+  "reverse_transformer_layers_per_block": null,
+  "sample_size": 128,
+  "time_cond_proj_dim": null,
+  "time_embedding_act_fn": null,
+  "time_embedding_dim": null,
+  "time_embedding_type": "positional",
+  "timestep_post_act": null,
+  "transformer_layers_per_block": [
+    1,
+    2,
+    10
+  ],
+  "up_block_types": [
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D",
+    "UpBlock2D"
+  ],
+  "upcast_attention": null,
+  "use_linear_projection": true
+}

unet/openvino_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d07de1da677a4170b71d0151e541127b759557c34a8bbe3ef297be90615b593
+size 1496655568

vae_decoder/config.json

@@ -0,0 +1,38 @@
+{
+  "_class_name": "AutoencoderKL",
+  "_diffusers_version": "0.37.1",
+  "_name_or_path": "/tmp/opencode/nova_xl_lcm_pt_v2/vae",
+  "act_fn": "silu",
+  "block_out_channels": [
+    128,
+    256,
+    512,
+    512
+  ],
+  "down_block_types": [
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D"
+  ],
+  "force_upcast": true,
+  "in_channels": 3,
+  "latent_channels": 4,
+  "latents_mean": null,
+  "latents_std": null,
+  "layers_per_block": 2,
+  "mid_block_add_attention": true,
+  "norm_num_groups": 32,
+  "out_channels": 3,
+  "sample_size": 1024,
+  "scaling_factor": 0.13025,
+  "shift_factor": null,
+  "up_block_types": [
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D"
+  ],
+  "use_post_quant_conv": true,
+  "use_quant_conv": true
+}

vae_decoder/openvino_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6f4aba6e05627958aaa16f584953395d3fd3c327abc7ce4fca4c62e1b12e7cb
+size 49557592