Upload README.md with huggingface_hub

README.md

@@ -40,13 +40,16 @@ retrieves worse.
 
 | build | file | size (MB) | Mac ms* | backend takes | worst cosine vs eager | retrieval budget |
 |---|---|---|---|---|---|---|
-| fp32 | `embed_multilingual_e5_large_xnnpack_fp32.pte` | 2235.7 | 103.6 | 80.7% | 1.000000 | 0% |
-| fp16 | `embed_multilingual_e5_large_xnnpack_fp16.pte` | 1118.2 | 145.5 | 69.6% | 0.999999 | 6% |
-| Core ML (fp16, iOS) | `embed_multilingual_e5_large_coreml_all.pte` | 1119.5 | 28.1 | 100.0% | 0.999992 | 40% |
+| fp32 | `embed_multilingual_e5_large_xnnpack_fp32.pte` | 2235.7 | 85.9 | 80.7% | 1.000000 | 0% |
+| fp16 | `embed_multilingual_e5_large_xnnpack_fp16.pte` | 1118.2 | 142.2 | 69.6% | 0.999999 | 6% |
+| Core ML (fp16, iOS) | `embed_multilingual_e5_large_coreml_all.pte` | 1119.5 | 27.2 | 100.0% | 0.999992 | 40% |
 
-\*Mac arm64, median of 10, one 256-token sequence — a reference point for relative
-cost, not a device number. Torch eager fp32 on the same machine is
-135.6 ms.
+\*Mac arm64, one 256-token sequence, **fastest of five medians of ten** — a reference
+point for relative cost, not a device number. The host shares its cores with other work,
+and a single median does not survive that: the same eager model here measured 19.6 ms and
+182.8 ms twenty minutes apart. Contention only ever adds time, so the fastest repetition is
+the one that means something. Torch eager fp32, measured the same way, is
+108.5 ms.
 
 Cosine is measured against the model run in eager through its own pooling, over eight
 sentences. The last column is the one that decides: rank those eight against each
@@ -57,21 +60,20 @@ top-1 results.
 ## The attention is eager, and that is the faster export
 
 `F.scaled_dot_product_attention` does not survive export as one operation. The edge
-dialect lowers it through `_safe_softmax`, whose guard against a row with no unmasked
-key at all leaves **eleven operations XNNPACK cannot take, in every attention block** —
-`scalar_tensor`, `where`, `mul.Scalar`, `logical_not`, `eq`, `full_like`, `any.dim`.
-Each one cuts the subgraph in two. The count is exact and does not vary by family:
-measured across this shelf, from a 4-layer cross-encoder to a 28-layer causal reranker,
-it is 11 per block every time.
-
-The guard is emitted whether or not it can ever fire, and here it cannot. It triggers
-only on `-inf`, which reaches the graph only because the sdpa path hands `F.sdpa` a
-**boolean** mask for PyTorch to fill; `attn_implementation="eager"` masks with
-`torch.finfo(dtype).min`, a large finite number, and never produces one. So the two
-arms differ only on rows that have no unmasked key — sdpa zeroes them, eager gives them
-a uniform row — and those are padding rows, which the pooling discards and every real
-query row masks out. Measured with all but eight positions masked, as adversarial as
-this shape gets, the two graphs agree to 1.4e-07.
+dialect lowers it through `_safe_softmax`, whose guard against a row with no unmasked key
+at all leaves **11 operations XNNPACK cannot take, in every attention
+block** — `scalar_tensor`, `where`, `mul.Scalar`, `logical_not`, `eq`, `full_like`, `any.dim`. Each one cuts the subgraph in two.
+
+The switch is `attn_implementation="eager"`: transformers then builds the mask
+itself, as `torch.finfo(dtype).min`, instead of handing `F.sdpa` a **boolean** mask
+for PyTorch to fill with `-inf`.
+
+The guard is emitted whether or not it can ever fire, and here it cannot: it triggers only
+on `-inf`, and this arm never produces one. So the two differ only about rows that have no
+unmasked key at all — sdpa zeroes them, this one gives them a uniform row — and those are
+padding rows, which the pooling discards and which every real query row masks out anyway.
+Measured with all but eight positions masked, as adversarial as this shape gets, the two
+graphs agree to 1.4e-07.
 
 XNNPACK fp32 goes from **63.8% to 80.7%** delegated.