Public-facing model card

README.md

@@ -5,106 +5,58 @@ license_link: https://huggingface.co/stabilityai/sdxl-turbo/blob/main/LICENSE.md
 tags:
   - stable-diffusion-xl
   - sdxl-turbo
-  - qnn
+  - qualcomm
   - hexagon
+  - qnn
   - npu
-  - w8a16
   - text-to-image
   - tokforge
-pipeline_tag: text-to-image
 library_name: qnn
+pipeline_tag: text-to-image
 inference: false
 ---
 
-# TokForge SDXL-Turbo · Hexagon NPU (W8A16)
-
-**PRIVATE** model bundle for the **TokForge** Android app (`dev.tokforge`). This repo hosts the
-**SDXL-Turbo** image-generation pipeline compiled to **QNN context binaries** for the
-**Qualcomm Hexagon DSP (HTP)**, so TokForge can fetch the SDXL-NPU "Faithful" backend from scratch
-(previously hand-staged on devices only).
-
-Sibling of [`darkmaniac7/TokForge-SD15-QNN-NPU`](https://huggingface.co/darkmaniac7/TokForge-SD15-QNN-NPU)
-(SD1.5). Same non-root `untrusted_app` cDSP recipe; this is the higher-quality 1024px tier.
-
-## What this is
-
-- **Model:** SDXL-Turbo (few-step distilled SDXL), native **1024×1024** (128×128 latent).
-- **Quantization:** **W8A16** (8-bit weights, 16-bit activations) on the Hexagon HTP.
-- **Runtime:** **QAIRT 2.40**. Bin set is **V75-native** (`libQnnHtpV75Skel.so`).
-- **Variant:** **fp16 text encoders + TAESDXL** tiny-VAE (see below).
-- **Footprint:** **~4.3 GB** on disk (the on-device fresh download is this big).
-
-## Pipeline / variant
-
-This bundle is the **fp16-encoders + TAESDXL** variant the `libsdxl_qnn_driver` runs end-to-end
-(no host orchestration beyond the emb-MLP):
-
-| Stage | File | Where it runs |
-|-------|------|---------------|
-| CLIP-L text encoder | `text_encoder_1_fp16.bin` | DSP (fp16) → `last_hidden [1,77,768]` |
-| OpenCLIP-bigG text encoder (+proj) | `text_encoder_2_fp16.bin` | DSP (fp16) → `last_hidden [1,77,1280]` + `pooled [1,1280]` |
-| text_embedding | concat → `[1,77,2048]` | driver |
-| combined-emb MLP | `sdxl_emb_mlp.bin` | **host CPU** → `emb [1,1280]` per step |
-| UNet (EulerAncestral, 4 steps, guidance 0) | `unet.bin` (2.6 GB) | DSP |
-| VAE decode (1024px) | `taesdxl_decoder.mnn` (2.4 MB) | **CPU** (MNN tiny-VAE) |
-
-**Why the host emb-MLP:** qairt-converter 2.40 constant-folds SDXL's two sinusoidal paths
-(`time_proj` and `add_time_proj`). The driver dodges both by precomputing the **combined**
-`emb = time_embedding(time_proj(t)) + add_embedding(cat(pooled, add_time_proj(time_ids)))` on host
-(weights in `sdxl_emb_mlp.bin`) and feeding it as ONE float UNet input, so no foldable path remains.
-
-**Why TAESDXL CPU VAE:** it is the **only validated** VAE path. The DSP VAE is blocked at 1024px on
-V75 (unsigned-PD HTP cap) and the OpenCL/Vulkan VAE SIGSEGVs Adreno in `Session::resize`. The driver
-always selects `--vae_mnn` + `SDXL_VAE_BACKEND=cpu`.
-
-**SDXL-Turbo is guidance-free** (`guidance_scale 0`): NO CFG → one UNet pass per step. EulerAncestral,
-4 steps native (6 for quality), VAE scaling factor **0.13025**.
-
-## Arch coverage & verification
-
-| Set | dsp_arch | Skel | Verified |
-|-----|----------|------|----------|
-| `ours/v75` | 75 (V75) | `libQnnHtpV75Skel.so` | **YES** — device-verified coherent+crisp on Lenovo **SM8650/V75** (.69) |
+# TokForge — SDXL-Turbo (Qualcomm Hexagon NPU)
 
-**Forward-compat:** a lower-arch bin runs on a higher-arch DSP. The **V75-native** set is the
-shipping set for **V73 → V81**, and was **verified by forward-compat on V81** (RedMagic **SM8850**
-handsets .82 and .80, coherent+crisp). **Native V81/V79 SDXL bins are NOT separately built** — V75-native
-is the single own-built set today.
+**SDXL-Turbo** image generation for the **Qualcomm Hexagon NPU (HTP)**, packaged for **on-device**
+image generation in the **TokForge** Android app (`dev.tokforge`). This is the higher-quality
+**1024×1024** tier alongside the SD1.5 NPU bundle.
 
-| SoC class | Hexagon | Status |
-|-----------|---------|--------|
-| 8 Gen 2 / 8s Gen 3 (SM8550/SM8635) | V73 | covered by V75-native forward-compat (untested on device) |
-| 8 Gen 3 (SM8650) | V75 | **device-verified** |
-| 8 Elite (SM8750) | V79 | forward-compat (untested on V79 silicon) |
-| 8 Elite Gen 5 (SM8850) | V81 | **forward-compat verified** (.82, .80) |
+The model is quantized to **W8A16** (8-bit weights, 16-bit activations) and compiled to **QNN HTP
+context binaries** that run on the phone's Hexagon DSP. The pipeline uses **fp16 text encoders** and a
+**TAESDXL** tiny-VAE decoder, and is **guidance-free** (few-step distilled, one UNet pass per step).
 
-## Provenance
+## Based on
 
-Clean-room conversion, our own pipeline (no third-party NPU bins):
+[`stabilityai/sdxl-turbo`](https://huggingface.co/stabilityai/sdxl-turbo).
 
-1. **Base:** `stabilityai/sdxl-turbo` (fp16 safetensors) → ONNX export (UNet emb-input/fold-proofed
-   graph, dual CLIP encoders, TAESDXL decoder).
-2. **Quantize:** **W8A16** via QAIRT (calibration set), graph name `model`.
-3. **Compile:** `qnn-context-binary-generator` → V75 HTP context binaries (**QAIRT 2.40**).
-4. **VAE:** TAESDXL exported to a CPU MNN model (`taesdxl_decoder.mnn`).
+## Format
 
-## License
+**QNN HTP context binaries** (W8A16), native **1024px** (128×128 latent). These are compiled for the
+Hexagon DSP and are **not** a portable format like GGUF. The shipping set is V75-native and is
+**forward-compatible** (a lower-arch binary also runs on a higher-arch DSP), covering Hexagon
+**V73 → V81**.
 
-**SDXL-Turbo — Stability AI Community License** (`stabilityai/sdxl-turbo`).
+| Stage | File |
+|-------|------|
+| CLIP-L text encoder (fp16) | `text_encoder_1_fp16.bin` |
+| OpenCLIP-bigG text encoder (fp16) | `text_encoder_2_fp16.bin` |
+| combined embedding MLP | `sdxl_emb_mlp.bin` |
+| UNet | `unet.bin` |
+| VAE decode (TAESDXL tiny-VAE) | `taesdxl_decoder.mnn` |
+| tokenizers | `tokenizer/`, `tokenizer_2/` |
 
-- Commercial use **permitted** for organizations/individuals with **< $1M USD annual revenue**.
-- Above $1M revenue requires a **Stability AI Enterprise license**.
-- **Attribution + license text must be retained** with the model.
-- This is **NOT** OpenRAIL-M (that covers the sibling SD1.5 bundle).
+See `manifest.json` for the authoritative file set (with per-file size + md5) that the app uses to
+download the bundle.
 
-Text encoder bases: OpenCLIP bigG (MIT-class), OpenAI CLIP-L (MIT).
+## Usage
 
-## Files
+This bundle is loaded automatically by the TokForge Android app — it is not a standalone diffusers
+checkpoint. The app downloads the file set from `manifest.json` and runs the binaries on the device NPU.
 
-See `manifest.json` for the authoritative file set with per-file md5 + size. The app reads
-`manifest.json`, resolves the device Hexagon arch, downloads the matching set into
-`filesDir/image_models/SDXL-QNN-NPU/`, and the `libsdxl_qnn_driver` loads the bins flat from there.
+## License & attribution
 
-**Driver-required files** (every one must be present + non-empty for a usable bundle):
-`unet.bin`, `text_encoder_1_fp16.bin`, `text_encoder_2_fp16.bin`, `sdxl_emb_mlp.bin`,
-`taesdxl_decoder.mnn`, `tokenizer/{vocab.json,merges.txt}`, `tokenizer_2/{vocab.json,merges.txt}`.
+This model is a derivative of [`stabilityai/sdxl-turbo`](https://huggingface.co/stabilityai/sdxl-turbo)
+and is distributed under the **Stability AI Community License**
+([LICENSE](https://huggingface.co/stabilityai/sdxl-turbo/blob/main/LICENSE.md)). Attribution to Stability AI
+and retention of the license text are required.