Release Wan2.2 T2V A14B q8/BF16 MLX-Gen package

README.md

@@ -13,14 +13,13 @@ tags:
 - wan2.2
 - video-generation
 - text-to-video
-- image-to-video
 - wan-a14b
 ---
 # wan2.2-t2v-a14b-diffusers-8bit
 
 This repository contains MLX-Gen saved weights for `Wan-AI/Wan2.2-T2V-A14B-Diffusers`. The checkpoint is designed for local Apple Silicon inference with [`mlx-gen`](https://github.com/lpalbou/mlx-gen).
 
-It uses the mflux/MLX saved-weight layout and MLX quantization tensors. It is not a Diffusers or Transformers `from_pretrained()` checkpoint.
+It uses the mflux/MLX saved-weight layout. Quantized checkpoints include MLX quantization tensors. It is not a Diffusers or Transformers `from_pretrained()` checkpoint.
 
 ## Source Model
 
@@ -36,7 +35,7 @@ This is an MLX q8 checkpoint for Wan2.2 A14B. MLX-Gen uses 8-bit quantization fo
 
 - q8 for quantizable Wan transformer attention and feed-forward modules.
 - BF16 for the Wan VAE.
-- BF16 for Wan transformer conditioning/output projection linears, the UMT5 text encoder, scheduler metadata, tokenizer files, norms, and other non-quantizable parameters.
+- BF16 for Wan transformer conditioning/output projection linears, the UMT5 text encoder, scheduler metadata, tokenizer files, norms, convolutions, and other non-quantizable parameters.
 
 Wan q4 quality and any possible mixed q4/q8 policy are still under validation. Prefer q8 for publishable Wan checkpoints until the q4 policy is documented.
 
@@ -44,31 +43,30 @@ See the [MLX-Gen quantization docs](https://github.com/lpalbou/mlx-gen/blob/main
 
 ## Local Validation
 
-Validation used 384x224, 17 frames, 12 denoising steps, guidance 4, guidance-2 3, fps 8, and seed 4242 on Apple Silicon. The memory numbers cover the full run from model init through video save. MLX peak is the MLX allocator peak; physical peak is the Darwin process physical footprint, which better reflects Apple Silicon unified-memory pressure than RSS alone.
+These measurements are validation-sized release checks for this uploaded package. They verify package loading, video integrity, and prompt influence for this profile only; they do not claim full-size `1280x720`, 81-frame, 40-step readiness.
 
-Bottom line:
-
-- The BF16 package reduces storage, not runtime memory.
-- This mixed q8/BF16 package reduces both storage and runtime memory. This is the package to use when generation memory footprint matters.
-
-| Layout | Disk | MLX Peak | Physical Peak | Time | Result |
-| --- | ---: | ---: | ---: | ---: | --- |
-| Original source snapshot | 118 GiB | 33.0 GiB | 48.9 GiB | 108.3 s | Baseline. |
-| BF16 package | 64 GiB | 33.0 GiB | 45.1 GiB | 114.4 s | Storage only; output was byte-identical. |
-| This mixed q8/BF16 package | 40 GiB | 20.8 GiB | 31.8 GiB | 110.3 s | Storage and memory; side-by-side quality validation passed. |
-
-Compared with the original source snapshot, this mixed q8/BF16 package cuts disk usage by about 66%, MLX peak memory by about 37%, and physical peak memory by about 35% in this validation run. It is not byte-identical to BF16, but the validation contact sheet stayed in the same visual family. The prepared q8/BF16 output was byte-identical to running `--quantize 8` from the upstream source snapshot.
+| Measurement | Value |
+|---|---:|
+| Package disk usage | 39.5 GiB |
+| Validation profile | 384x224, 33 frames, 12 steps, 8.0 fps, seed 4242, `--low-ram` |
+| Prompt pair | scientist scene / red car scene |
+| Video health | 33 / 33 frames decoded, 8.0 fps, nonblank |
+| Mean temporal delta | 5.6 / 3.2 luma |
+| Prompt delta | 102.0 mean abs RGB |
+| Generation time | 162.2 s / 319.6 s |
 
 ## Compatibility
 
-Requires `mlx-gen >= 0.18.8`.
+Requires `mlx-gen >= 0.18.9`.
 
-Generated with `mlx-gen 0.18.8`.
+Generated with `mlx-gen 0.18.9`.
 
 Use the `mlxgen` command and Python import path for new MLX-Gen projects.
 
 ## Usage
 
+The q8 A14B example below is intentionally validation-sized. Do not use this card to claim full-size `1280x720`, 81-frame, 40-step readiness until that exact path has passed video integrity and quality validation.
+
 ```bash
 python -m pip install -U mlx-gen
 
@@ -78,14 +76,15 @@ mlxgen generate \
   --model AbstractFramework/wan2.2-t2v-a14b-diffusers-8bit \
   --task text-to-video \
   --prompt "Your video prompt here" \
-  --width 1280 \
-  --height 720 \
-  --frames 81 \
-  --steps 40 \
+  --width 384 \
+  --height 224 \
+  --frames 33 \
+  --steps 12 \
   --guidance 4 \
   --guidance-2 3 \
-  --fps 16 \
-  --seed 42 \
+  --fps 8 \
+  --seed 4242 \
+  --metadata \
   --output video.mp4
 ```
 

transformer/0.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d324244b04b73bd63a5e41775a9a5fc0e38a0fdc6f6f641d274496fed8d8d727
+oid sha256:778d017f1aaec61aa14ac84ee8305f47cd6b8c7ee988b213adea84cdbcd84da8
 size 2126914156

transformer/1.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df5419e4f564780aaa48b01874bae29a765c3df5466f5d8878fd2b85537f6cd9
+oid sha256:4d515a2ca0dd66597b72baf73badfcfb2bcce966fb9c725d2e3e324aa827efe4
 size 2129375449

transformer/2.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd1774082f5db81dca8f8ed514c241e0ab58f3b434b108c579ab72c9298cc4be
+oid sha256:ec9e66ad6f169635400291353eca81e8566bdc19e66c6950ad77c39ff1dd5679
 size 2109848506

transformer/3.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b85b62cfbe121c9ba895d83e71fb22df7f793f907ba6d462da7d86e000aa7294
+oid sha256:a935fc1df9a54abd8b3c95ab963797e73e813de307df9e4fc378cad6f48cb215
 size 2137714721

transformer/4.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c966f65563f26fcce793a4cf823949926f987a60ed8ddb837eddff91f133642
+oid sha256:2ba701f70de8516d08cc0290341edadc298e89878b14b47374fd0e7200075274
 size 2129375671

transformer/5.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:43cccf0b11fc18715f05c530d2cfe8e4ee6c1c1e7c8fbc57f13d02d6c535468d
+oid sha256:34de1e3c49b2d6b32768925b31a7bfd21ec11ec68f43e6ddc50d4a8378717825
 size 2081985022

transformer/6.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:401c3d202660a8ab6d75d119cf1b7a019c94e9aed131b478a7d67cc5d71e5747
+oid sha256:92d9390b17d0a97ca5029b99cec0ca1c10de21a5c4ed6cc4ae934919eae05218
 size 2137714713

transformer/7.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3684a00b35ab9db7a5c103704ef99943a12d60ad18ad9e2932d126f40afd0abc
+oid sha256:658d8b09c15faa69ab0baa75fd28c8b743cbe14317fa93e17060984cdfacb1d3
 size 552561394

transformer/model.safetensors.index.json

@@ -1,7 +1,7 @@
 {
   "metadata": {
     "quantization_level": "8",
-    "mflux_version": "0.18.8"
+    "mflux_version": "0.18.9"
   },
   "weight_map": {
     "rope.freqs_cos": "0.safetensors",

transformer_2/0.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2732c943575c11fef937956a9e8400ced38374295954b58fbc5a1af01d32f986
+oid sha256:dbeaaaf67edd24e54ea4e9f80d314df1be530be819243e227a81547704ab4c03
 size 2126914156

transformer_2/1.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3e2f216038257ac1b3a21def6062756d862ce266fd571f1c4a9cd03ca1b7e0b7
+oid sha256:ba39c2e3bfcd1c5a704e2ddec57f49ea45803c12ea50418841fd475e1049d864
 size 2129375449

transformer_2/2.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e94647aeaca6371b650257c9879bf1d16cb273f4b158176ab1169a93703adad0
+oid sha256:df2cd8228f861d6e2a21e59c12037cfb5014edc9f509483699af5dd0fdfda46a
 size 2109848506

transformer_2/3.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:544ac869de8cf8205875ca1677617ec28f300b440cb8b414a8c868bc2d45b1e4
+oid sha256:cb84b2a865ce243d3f86da5919626c10bf14c878103d7905a6988bea287b15e1
 size 2137714721

transformer_2/4.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0efb68e19c4b89ad3d822c4d1d5c8971f3be063da25785cdf98f4ef3d23973c3
+oid sha256:f9e27a54ee311ea9a0eb9c347b5da1b198a6c5797d024b4968d3b80ed12c2f7b
 size 2129375671

transformer_2/5.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f50af08070a793cca64ade0eb3209d928511fa524e229eeb090cee13f62d1af4
+oid sha256:2ee7573a19f22cdb1b35985ce0628b652584f304eaa66d41400ed8e13ed2be6b
 size 2081985022

transformer_2/6.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68d30462066067fa5e8e8c0386f56257e71a721303e6575e7395af8e7e0c68c3
+oid sha256:7d95c814e12f9f8a55a182b09f9822bbd59a1b17a48a185964d0650092cdd728
 size 2137714713

transformer_2/7.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc7a2a4ef35f0d608768caf516d2a0dd3b2d8da782b6598294adeb5272ecaed1
+oid sha256:f4c5f350f81e43166976120fad70c218d45907474ced78002636ffcee4c2e977
 size 552561394

transformer_2/model.safetensors.index.json

@@ -1,7 +1,7 @@
 {
   "metadata": {
     "quantization_level": "8",
-    "mflux_version": "0.18.8"
+    "mflux_version": "0.18.9"
   },
   "weight_map": {
     "rope.freqs_cos": "0.safetensors",

vae/0.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:039ca54426d7cbfb4ecf852f06b0d943c2450f34a6c92c357824c69d8449587f
-size 253807817
+oid sha256:e3d43fd7e6e6370daf762a357d4fab09f34118d0b9f9dc383f322fffe0db0411
+size 253807820

vae/model.safetensors.index.json

@@ -1,7 +1,7 @@
 {
   "metadata": {
-    "quantization_level": "8",
-    "mflux_version": "0.18.8"
+    "quantization_level": "None",
+    "mflux_version": "0.18.9"
   },
   "weight_map": {
     "encoder.conv_in.conv3d.weight": "0.safetensors",