mlboydaisuke/SAM2.1-hiera-tiny-ExecuTorch

SAM 2.1 hiera-tiny — ExecuTorch XNNPACK (encoder + decoder)

Two .pte files (image-segment path: run the encoder once per image, the decoder per click):

• sam21_tiny_encoder_xnnpack_fp32.pte (109.2 MB) — image (1,3,1024,1024) → image_embed (1,256,64,64), feat_s0 (1,32,256,256), feat_s1 (1,64,128,128)
• sam21_tiny_decoder_xnnpack_fp32.pte (24.7 MB) — (image_embed, feat_s0, feat_s1, points (1,1,N,2) fp32 pixel coords in 1024-space, labels (1,1,N) int64 1=fg/0=bg) → mask logits (1,1,3,256,256), iou scores (1,1,3)

Both graphs also ship in fp16, and both hold their accuracy there:

• sam21_tiny_encoder_xnnpack_fp16.pte (55.6 MB, corr 0.999993)
• sam21_tiny_decoder_xnnpack_fp16.pte (12.6 MB, corr 0.999999)

Every file takes and returns fp32 tensors, so precision is a file swap — the fp16 pair is 68.2 MB against 133.9 MB. Hiera is attention-heavy, which is why fp16 halves it cleanly; dynamic int8 was measured too and is not published, since it leaves the convolutional parts in fp32 and came out at the full 109.2 MB for the encoder.

If size is the binding constraint, look at EdgeTAM — same output contract, 19.7 MB encoder.

• Source: facebook/sam2.1-hiera-tiny (transformers Sam2Model)
• License: Apache-2.0
• Preprocess: RGB/255, imagenet norm (mean .485/.456/.406, std .229/.224/.225), resize 1024x1024
• Postprocess: pick argmax(iou) of the 3 mask logits, threshold > 0, upsample 4x (256→1024) to image space. The prompt encoder is embedded in the decoder — pass raw click coordinates, no separate point-encoding code needed.

Verification (Mac arm64, executorch 1.4.0, torch 2.13.0)

Parity vs torch fp32 eager (random input): all 5 outputs corr 1.000000 (worst max_abs_diff 2.2e-05). End-to-end two-.pte chain on a synthetic image: mask IoU vs torch = 1.0000, iou_scores match to 6 decimals.

Median latency (Mac, reference only — ViT-heavy graphs are known to time poorly on Mac XNNPACK; device numbers to follow): encoder 2506 ms, decoder 20 ms.

Conversion notes

torch.export → to_edge_transform_and_lower(XnnpackPartitioner) → .pte, with:

• .contiguous() on the encoder outputs / decoder inputs (transformers emits channels_last; leaving it poisons XNNPACK's runtime shape propagation)
• identity repeat_interleave(1, dim) calls stripped (delegate mis-sizes their decomposition)
• constant dense_prompt_embeddings (no-mask embedding) baked as a buffer