ms-marco-MiniLM-L6-v2 — ExecuTorch

A cross-encoder reranker: a query and one document in, one relevance score out. The second stage of on-device retrieval — an embedding model fetches candidates cheaply, this reads each candidate together with the query and scores it properly.

Source: cross-encoder/ms-marco-MiniLM-L6-v2 — 22.7M parameters, 6 BERT layers, hidden 384
License: Apache-2.0
Input: input_ids, attention_mask and token_type_ids, each [1, 512] int64
Output: [1, 1] fp32 — the raw logit. sigmoid(x) maps it to 0..1 and does not change the ordering.

Variants

build	file	size (MB)	worst score error vs eager	Mac median (ms)*
fp32	`rerank_ms_marco_minilm_l6_xnnpack_fp32.pte`	91.0	0.0000 logits	27.3
fp16	`rerank_ms_marco_minilm_l6_xnnpack_fp16.pte`	45.6	0.0062 logits	53.8
Core ML (fp16, iOS)	`rerank_ms_marco_minilm_l6_coreml_all.pte`	45.7	0.0194 logits	6.7

*Mac arm64, single process, median of 10, one query-document pair at 512 tokens — a reference point for relative cost, not a device number. PyTorch eager fp32 on the same machine: 14.7 ms.

Correlation is not reported because it cannot be: the output is a single number, and the correlation of a one-element vector is undefined. The column above is the error in the units the model is used in — logits — over 6 real query-document pairs, and every build listed reproduces eager's ranking order exactly.

What it does, on the shipped fp32 build

Query: "How many people live in Berlin?"

rank	score	document
1	+8.816	Berlin has a population of 3,520,031 registered inhabitants in an area of 891.82 km².
2	+1.973	In 2019 the city recorded 3.7 million residents within its metropolitan area.
3	-3.636	Berlin is well known for its museums, its nightlife and its history.
4	-4.487	The capital of France is Paris, a city of about 2.1 million people.
5	-10.962	ベルリンの人口はおよそ350万人です。
6	-11.162	Water boils at 100 degrees Celsius at sea level.

The narrowest gap between adjacent ranks here is 0.2007 logits.

`token_type_ids` is not optional

A BERT cross-encoder marks the document half of the pair with segment id 1, and the segment embedding is doing real work. Measured on this model with the query above and the passage that answers it, feeding zeros instead of the real segment ids moves the score from +8.82 to -2.54, a drop of 11.35 logits — the best document in the list crosses into negative territory, where any threshold rejects it. The graph therefore takes three inputs. XLM-R rerankers (type_vocab_size: 1) have no second segment and take two; the signature follows the model rather than being made uniform.

Not shipped

int8 (dynamic) is not shipped: at 58.8 MB it is larger than the fp16 build's 45.6 MB. Dynamic int8 quantizes the linear weights and leaves the token embedding table in fp32, while fp16 halves that table too. The table here is 47 MB of a 91 MB model, and the arithmetic says int8 only comes out smaller when the table is under a third of the weights (30 MB) — measured on ten models on this shelf, the rule called all ten correctly.

Verification

python convert/export_rerank.py ms_marco_minilm_l6
python convert/check_rerank.py ms_marco_minilm_l6 fp32

The check has two halves. One is agreement with the model run in eager, in logits and in ranking order. The other is that the ranking is useful at all: the passage that answers the question has to outscore a passage about the same subject that does not — agreement alone would pass a build that ranked by document length in both arms.

(conversion scripts: executorch-models)