[diffusion] quant: support pruned safetensors checkpoints for minimax-h3 (#35418)

docs/cookbook/diffusion/MiniMax/MiniMax-H3.mdx

@@ -248,6 +248,59 @@ server environment.
 
 For MiniMax-H3, `--performance-mode speed` deliberately keeps the DiT eager. The current `torch.compile` path changes the model's numerical output, so it is not enabled implicitly by any recommended lossless preset. An explicit `--enable-torch-compile true` remains available for controlled experiments, but it should not be used to generate consistency ground truth.
 
+### AdaLN-pruned safetensors transformers
+
+[Comfy-Org/MiniMax-H3](https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main/diffusion_models)
+publishes smaller DiT-only checkpoints that replace the original AdaLN branches
+with an interpolated curve table. Select one file explicitly; the base model
+still supplies the text encoder and VAEs.
+
+```bash Command
+sglang serve \
+  --model-path MiniMaxAI/MiniMax-H3 \
+  --model-variant fl2va \
+  --transformer-weights-path \
+    Comfy-Org/MiniMax-H3/diffusion_models/minimax_h3_fl2va_pruned_bf16.safetensors \
+  --num-gpus 4 \
+  --tp-size 2 \
+  --ulysses-degree 2 \
+  --performance-mode speed \
+  --port 30010
+```
+
+The pruned checkpoint is approximate and is therefore rejected by
+`quality="high"`, which remains limited to the audited official BF16 DiT.
+
+The Comfy `pruned_fp8_scaled` FL2VA and Ref2VA files are also supported. Their
+per-layer markers are detected automatically; do not add `--quantization`:
+
+```bash Command
+sglang serve \
+  --model-path MiniMaxAI/MiniMax-H3 \
+  --model-variant fl2va \
+  --transformer-weights-path \
+    Comfy-Org/MiniMax-H3/diffusion_models/minimax_h3_fl2va_pruned_fp8_scaled.safetensors \
+  --num-gpus 4 \
+  --tp-size 2 \
+  --ulysses-degree 2 \
+  --performance-mode speed \
+  --port 30010
+```
+
+SGLang uses its native static-activation FP8 linear path for attention and
+`fc1`. The checkpoint marks `fc2` for full-precision matrix multiplication, so
+SGLang retains its FP8 storage but materializes and scales one compute-dtype
+`fc2` matrix for each call. This preserves the checkpoint's mixed execution
+contract and low resident weight memory, but that part is slower than a fully
+quantized FP8 GEMM. TP, Ulysses/Ring sequence parallelism, and
+component/layerwise offload are supported; FSDP inference is rejected.
+
+The `pruned_int8_convrot` files are detected but remain unsupported. They
+require online regular-Hadamard ConvRot, dynamic INT8 activation quantization,
+and a matching W8A8 GEMM. SGLang fails before loading them instead of silently
+treating their stored INT8 values as ordinary weights. A native ConvRot kernel
+path should be added and benchmarked separately before these files are accepted.
+
 ### Advanced: precomputed AdaLN cache
 
 The [model card](https://huggingface.co/MiniMaxAI/MiniMax-H3) notes that about