Keep the house card complete and runnable

README.md

@@ -12,8 +12,7 @@ base_model:
 
 # MiniMax-M3.1-Flash Vision
 
-This repository packages the Tower and Projector from
-[MiniMax M3.1 Flash](https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash).
+This repository packages the Tower and Projector from [MiniMax M3.1 Flash](https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash).
 
 ## Contents
 
@@ -25,6 +24,8 @@ This repository packages the Tower and Projector from
 | `config.json`, `preprocessor_config.json` | | Vision-only model and image-processing configuration |
 | `export_manifest.json` | | Source pin, tensor counts and file hashes |
 | `LICENSE`, `LICENSE-CODE`, `NOTICE` | | Weight license, code license and attribution |
+| `examples/inference.py` | | Single-image extraction example |
+| `README.md`, `.gitattributes` | | Model card and Hub storage metadata |
 
 ## Architecture
 
@@ -34,13 +35,13 @@ This repository packages the Tower and Projector from
 | Token compression | 2x2 spatial merge, temporal patch size 2 |
 | Projector | `Linear(1280, 6144)`, GELU, `Linear(6144, 6144)`, flatten four 6144-wide patches, `Linear(24576, 6144)`, GELU, `Linear(6144, 6144)` |
 
-The class is `MiniMaxM3VLVisionModel`. Still images are repeated as two frames for the
-temporal patch embedding. At 448 pixels the example reports Tower `(1, 1024, 1280)`,
+The class is `MiniMaxM3VLVisionModel`. Still images are repeated as two frames for the temporal patch embedding. At 448 pixels the example reports Tower `(1, 1024, 1280)`,
 merged `(1, 256, 5120)` and projected `(1, 256, 6144)`.
 
 ## Usage
 
-See [`examples/inference.py`](examples/inference.py) for image feature extraction.
+With PyTorch, Pillow, torchvision and Transformers 5.17+, run `python examples/inference.py test-image.jpg`
+from the downloaded snapshot; see [the example](examples/inference.py). It uses CUDA BF16 when available.
 Forward one image per call: the native HF attention couples images in a single call.
 
 ## Validation
@@ -49,6 +50,8 @@ The release tests compare all 515 Tower tensors and eight Projector tensors with
 pinned parent checkpoint using `torch.equal`. Fixed-image Tower, merged and projected
 outputs also match the native parent implementation at 56 and 448 pixels.
 
+[Validation evidence](https://github.com/immanuel-peter/vision-tower-bench/tree/t3/complete-issue-21/results/diagnostics/minimax-m31-flash).
+
 ## Reproduction
 
 The [export script](https://github.com/immanuel-peter/vision-tower-bench/blob/main/scripts/export_minimax_m3_vision.py)
@@ -61,5 +64,4 @@ MiniMax released the MiniMax-M3.1-Flash weights and the native Transformers impl
 
 ## License
 
-[MiniMax Community License](LICENSE), the same license as the source model. Helper code
-is MIT; see [LICENSE-CODE](LICENSE-CODE) and [NOTICE](NOTICE).
+[MiniMax Community License](LICENSE), the same license as the source model. Helper code is MIT; see [LICENSE-CODE](LICENSE-CODE) and [NOTICE](NOTICE).

export_manifest.json

@@ -17,7 +17,7 @@
     "LICENSE": "8c4978e3e0fe0349ff929cc8450463cba3039c4c9fc144d8bd5382ab00df247a",
     "LICENSE-CODE": "149f05ce186198084c9abd6af8e76e108ba7e507446a19ee5a2ffd694f69f525",
     "NOTICE": "19d19379d439b6dc2614476bb50b170fbe382a58f3a84ec67ff9bcd4e212d626",
-    "README.md": "4c85532cd87301b8ae8ec2686c6b3680aedda6d705089456ad7d65c701f3c6cc",
+    "README.md": "a246d9f3164c77d99137383e9812ee53dc87e321991d4182c2c37ff0a28bfdc0",
     "config.json": "9c3a6dac0654055594fdde1f148400331f3e55cb72ffe1b7a6806db07ab93a79",
     "examples/inference.py": "fcde5470ba4e9fce0730f544f4f3b7669bf4f9bdcaf556223cebc4ed67ef1c5a",
     "model.safetensors": "95e3731e26887d066a0b2291a1574da40a36aba5ba43e476746b9c16a64bcf08",