Document Viggle v0.2.1 QIPACK and LoRA

NOTICE

@@ -19,11 +19,18 @@ The four-step distilled pack was converted from the full transformer in:
   Viggle/Qwen-Image-2.1-viggle-turbo
   Revision bafc91e4cc934f5fb1406b22496a0bed9b99c548
 
+The six-step QIPACK uses the unchanged base transformer above. Its required
+rank-256 LoRA adapter is an unmodified copy from:
+
+  Viggle/Qwen-Image-2.1-viggle-turbo
+  File revision 139e9492e6b81e85395877a549ec8f0afbb18f8f
+  Qwen-Image-2.1-viggle-turbo-v0.2.1-6step-lora-r256.safetensors
+
 The processor, text-encoder, and VAE files are unmodified copies from:
 
   Qwen/Qwen-Image-2.1
   Revision b3179ad355be050328e483a9dfdd9e60cd62adfa
 
-The distilled model was created and published by Viggle under the Qwen
-Research License. The QIPACK conversion was produced by the
+The distilled model and LoRA were created and published by Viggle under the
+Qwen Research License. The QIPACK conversion was produced by the
 qwen-image-cplus project.

README.md

@@ -20,10 +20,11 @@ tags:
 [`qwen-image-cplus`](https://github.com/netdur/qwen-image-cplus), a native C+
 and Metal Qwen-Image-2.1 inference runtime for Apple Silicon.
 
-This repository contains the converted transformer weights and the unmodified
-processor, text encoder, and VAE files needed by `qwen-image-cplus`. All files
-come from the pinned upstream Qwen snapshot except the distilled transformer,
-which comes from the pinned Viggle snapshot. The runtime provides the scheduler.
+This repository contains converted transformer weights, a Viggle v0.2.1 LoRA,
+and the unmodified processor, text encoder, and VAE files needed by
+`qwen-image-cplus`. The base weights and shared support files come from the
+pinned upstream Qwen snapshot; the four-step full fine-tune and six-step LoRA
+come from Viggle. The runtime provides the scheduler.
 
 ## Files
 
@@ -31,34 +32,40 @@ which comes from the pinned Viggle snapshot. The runtime provides the scheduler.
 | --- | --- | --- | --- |
 | `qwen-image-2.1-fp16-v4.qipack` | `Qwen/Qwen-Image-2.1` at `b3179ad355be050328e483a9dfdd9e60cd62adfa` | 40 steps, TaylorSeer | `9fe30bcae5678c6e21618d48d9f058fc150ed5b71b7534e5283bb7a05c163750` |
 | `qwen-image-2.1-viggle-v0.1-4step-fp16-v4.qipack` | `Viggle/Qwen-Image-2.1-viggle-turbo` at `bafc91e4cc934f5fb1406b22496a0bed9b99c548` | 4 steps, no cache, unstretched schedule | `d05edecbf9d3e7b03b0e708ae41da5370d5fe245f1d3b5f6775aef18f18857d1` |
+| `qwen-image-2.1-viggle-v0.2.1-lora-fp16-v4.qipack` | Unchanged Qwen base transformer; requires the Viggle LoRA below | 6 steps, no cache, Viggle v0.2.1 schedule | `be2e72ed75d30234a7f1934502d1d95b10eaedc935de2ce06b926b34e1b7b8b4` |
+| `Qwen-Image-2.1-viggle-turbo-v0.2.1-6step-lora-r256.safetensors` | `Viggle/Qwen-Image-2.1-viggle-turbo` at `139e9492e6b81e85395877a549ec8f0afbb18f8f` | Rank-256 adapter beside the six-step QIPACK | `2a0148f5c73abbed5f97da5ea356e439318aadb281d01fce4af39cdf43728803` |
 
-Both packs share `processor/vocab.json`, `processor/merges.txt`, four
+All three packs share `processor/vocab.json`, `processor/merges.txt`, four
 `text_encoder/model-*.safetensors` shards, and
 `vae/diffusion_pytorch_model.safetensors` in this repository's root directory.
 Their sizes and checksums are recorded in `manifest.json`. The files remain in
-those paths when downloaded, so selecting either root-level QIPACK in the GUI
-also locates its support files. One 14.23 GB pack plus the 18.89 GB shared
-support files requires about 33.12 GB of local storage. “7B” describes the
-transformer's parameter count, not the complete pipeline's size in bytes.
+those paths when downloaded, so selecting a root-level QIPACK in the GUI
+also locates its support files. The six-step pack additionally requires its
+LoRA file beside it. One 14.23 GB pack plus the 18.89 GB shared files requires
+about 33.12 GB of local storage, or 34.48 GB with the 1.36 GB LoRA. “7B”
+describes the transformer's parameter count, not the pipeline's size in bytes.
 
-The distilled file is specifically the Viggle v0.1 four-step full transformer,
-not its LoRA and not the newer v0.2.1 six-step release.
+The four-step file is Viggle v0.1's full transformer, not a LoRA. The six-step
+file contains the original Qwen base transformer with metadata selecting the
+separate Viggle v0.2.1 LoRA and its six-step schedule.
 
-Both files use QIPACK1 version 1 with policy
+The three QIPACK files use QIPACK1 version 1 with policy
 `transformer:all-matrix-f16-v4`. All 224 transformer-block matrices are stored
 as FP16; vectors and the nine global tensors remain BF16. Each pack contains
 297 tensors and 7,115,124,736 parameters. The format has fixed little-endian
-metadata plus per-tensor and payload checksums, and both packs passed exact
-round-trip verification against their source tensors.
+metadata plus per-tensor and payload checksums. The base and four-step packs
+passed exact round-trip verification against their source tensors.
 
 ## Requirements
 
 - macOS 14 or newer on Apple Silicon
-- [`qwen-image-cplus`](https://github.com/netdur/qwen-image-cplus) v0.0.1 or newer
+- [`qwen-image-cplus`](https://github.com/netdur/qwen-image-cplus); the six-step
+  LoRA requires a build with Viggle v0.2.1 support (current `main`)
 
 The runtime has been tested on an M1 Max with 32 GB unified memory. Lower-memory
-machines have not yet been validated. Generation at 1024x1024 is supported;
-the model's native 2048x2048 path is not yet supported by this runtime.
+machines have not yet been validated. Generation at 1024x1024 and the model's
+native 2048x2048 resolution is supported; 2048x2048 is a high-memory capacity
+mode on a 32 GB M1 Max.
 
 ## Download
 
@@ -93,6 +100,17 @@ qwen-image-cplus generate-1024 \
   1301
 ```
 
+Six-step Viggle v0.2.1 LoRA (keep its sidecar safetensors beside the pack):
+
+```sh
+qwen-image-cplus generate-1024 \
+  models/qwen-image-2.1-viggle-v0.2.1-lora-fp16-v4.qipack \
+  models \
+  output.png \
+  "A vintage travel poster for CASABLANCA reading 'MEET ME AT SUNSET'" \
+  1301
+```
+
 The pack metadata selects the correct step count, terminal-shift behavior, and
 cache policy when those optional CLI arguments are omitted.
 
@@ -114,7 +132,7 @@ obtained from Qwen. Read the agreement before downloading or using the files.
 The QIPACK files are modified redistributions: their tensor storage layout and
 selected matrix dtypes were converted for the Apple-Silicon runtime. The model
 architecture and learned values were not retrained by this project. The
-distilled source model was created by Viggle and is separately attributed in
+distilled source model and LoRA were created by Viggle and are attributed in
 [`NOTICE`](NOTICE).
 
 The runtime source code has its own MIT license; that license does not replace

manifest.json

@@ -53,6 +53,41 @@
         "cache": "none"
       },
       "verification": "structural validation and exact source round-trip passed"
+    },
+    {
+      "path": "qwen-image-2.1-viggle-v0.2.1-lora-fp16-v4.qipack",
+      "local_source": "models/qwen-image-2.1-viggle-v0.2.1-lora-fp16-v4.qipack",
+      "bytes": 14230327296,
+      "sha256": "be2e72ed75d30234a7f1934502d1d95b10eaedc935de2ce06b926b34e1b7b8b4",
+      "source": {
+        "repository": "Qwen/Qwen-Image-2.1",
+        "revision": "b3179ad355be050328e483a9dfdd9e60cd62adfa",
+        "component": "transformer",
+        "modifications": "QIPACK conversion and metadata selecting the Viggle v0.2.1 LoRA"
+      },
+      "generation_defaults": {
+        "kind": "distilled",
+        "steps": 6,
+        "shift_terminal": null,
+        "cache": "none",
+        "schedule": "viggle-v0.2.1-6",
+        "adapter": "Qwen-Image-2.1-viggle-turbo-v0.2.1-6step-lora-r256.safetensors"
+      },
+      "verification": "structural validation and native six-step generation passed"
+    }
+  ],
+  "adapters": [
+    {
+      "path": "Qwen-Image-2.1-viggle-turbo-v0.2.1-6step-lora-r256.safetensors",
+      "local_source": "models/Qwen-Image-2.1-viggle-turbo-v0.2.1-6step-lora-r256.safetensors",
+      "bytes": 1359147904,
+      "sha256": "2a0148f5c73abbed5f97da5ea356e439318aadb281d01fce4af39cdf43728803",
+      "source": {
+        "repository": "Viggle/Qwen-Image-2.1-viggle-turbo",
+        "revision": "139e9492e6b81e85395877a549ec8f0afbb18f8f",
+        "component": "Qwen-Image-2.1-viggle-turbo-v0.2.1-6step-lora-r256.safetensors"
+      },
+      "modifications": "none"
     }
   ],
   "support_source": {