Add files using upload-large-folder tool

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+gt_top100.tsv filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,154 @@
+---
+pretty_name: msmarco_answerai_colbert_small
+license: cc-by-sa-4.0
+tags:
+- multi-vector
+- late-interaction
+- colbert
+- embeddings
+- retrieval
+- text
+---
+
+# msmarco_answerai_colbert_small
+
+Multi-vector (late-interaction) embeddings of **BEIR msmarco** (`beir/msmarco/dev`), encoded with
+**[lightonai/answerai-colbert-small-v1](https://huggingface.co/lightonai/answerai-colbert-small-v1)** at revision `e507cd12947a2b4b52201d150967df3c19a90590`.
+
+**Source data:** [ir_datasets](https://ir-datasets.com/beir.html#beir/msmarco/dev) `beir/msmarco/dev` (ir_datasets 0.6.3), which downloads [msmarco.zip](https://public.ukp.informatik.tu-darmstadt.de/thakur/BEIR/datasets/msmarco.zip) (md5 `444067daf65d982533ea17ebd59501e4`). BEIR also publishes this corpus on the Hub as [`BeIR/msmarco`](https://huggingface.co/datasets/BeIR/msmarco), whose card gives this dataset's license; the data here was loaded through ir_datasets, not from that repo. Document, query and qrel ids are the source's own ids,
+unchanged.
+
+Every document is one variable-length set of 96-d vectors; every query is one variable-length
+set of 96-d vectors. Documents and queries are stored at different precisions (fp16 and fp32
+respectively), see [Encoding](#encoding).
+
+## Files
+
+| file | dtype | shape | contents |
+|---|---|---|---|
+| `documents.npy` | float16 (`<f2`) | `[597,909,930, 96]` | every document vector, concatenated document by document (109,480.6 MiB) |
+| `doclens.npy` | int32 | `[8,841,823]` | vectors per document; `cumsum` gives offsets |
+| `token_ids.npy` | uint32 | `[597,909,930]` | tokenizer id of each `documents.npy` row, 1:1 |
+| `doc_ids.npy` | `<U7` | `[8,841,823]` | original document ids |
+| `queries.npy` | float32 (`<f4`) | `[6,980, 32, 96]` | query vectors, zero-padded at the end (81.8 MiB) |
+| `query_lens.npy` | int32 | `[6,980]` | true vectors per query, before padding |
+| `queries_ids.npy` | `<U7` | `[6,980]` | original query ids |
+| `qrels.test.tsv` | text | 7,437 rows | TREC qrels, `qid \t 0 \t docid \t relevance`, no header |
+| `gt_top100.tsv` | text | 698,000 rows | exact MaxSim top-100, see below |
+
+All positional indices (`gt_top100.tsv`, and the row order of every `.npy` file) refer to the order
+of `doc_ids.npy` and `queries_ids.npy`. Reordering either file invalidates `gt_top100.tsv`.
+
+## Statistics
+
+| | |
+|---|---|
+| documents | 8,841,823 |
+| document vectors | 597,909,930 |
+| vectors per document (min / median / mean / max) | 4 / 60 / 67.6 / 300 |
+| queries | 6,980 |
+| vectors per query (min / median / max) | 32 / 32 / 32 |
+| queries with at least one qrel | 6,980 |
+| qrels rows | 7,437 |
+| embedding dimension | 96 |
+
+## Encoding
+
+| | |
+|---|---|
+| model | [lightonai/answerai-colbert-small-v1](https://huggingface.co/lightonai/answerai-colbert-small-v1) |
+| model revision | `e507cd12947a2b4b52201d150967df3c19a90590` |
+| library | sentence-transformers 6.1.0 `MultiVectorEncoder` (transformers 5.17.0, torch 2.13.0+cu126) |
+| document compute dtype | float16 (model weights loaded at this dtype for the document pass) |
+| document storage dtype | fp16 |
+| query compute dtype | float32 (model weights loaded at this dtype for the query pass) |
+| query storage dtype | fp32 |
+| normalization | L2, by the model's own `Normalize` module, before the storage cast |
+| document truncation | 300 tokens (the checkpoint's `document_length`), before the skiplist; longest document here 300 vectors |
+| query truncation | `query_length` unset; fixed query expansion: every query is padded to 32 tokens with the tokenizer's mask token, which are not attended to, and those expansion vectors are kept |
+| document skiplist | 32 words removed: ['!', '"', '#', '$', '%', '&', "'", '(', ')', '*', '+', ',', '-', '.', '/', ':', ';', '<', '=', '>', '?', '@', '[', '\\', ']', '^', '_', '`', '{', '|', '}', '~'] |
+| document input | `title + "\n\n" + text` when the corpus has a title, else `text`; stripped |
+| query input | query text, stripped of surrounding whitespace, formatted by the model's own query prompt/template |
+| query vectors | every vector the model emits for the query is kept, including any query-expansion tokens its template adds; `query_lens` counts them all |
+| document padding | none: `documents.npy` holds real vectors only, `sum(doclens) == n_tokens` |
+| query padding | rows at or beyond `query_lens[i]` in `queries.npy[i]` are exactly zero |
+| token_ids | tokenizer id of each kept document token (after the skiplist above), aligned 1:1 with `documents.npy` |
+
+## Ground truth: `gt_top100.tsv`
+
+Exact brute-force MaxSim top-100 per query over the full corpus, from the vectors in this repo.
+No header; tab-separated `qidx  docidx  rank  score`:
+
+- `qidx`: 0-based row into `queries_ids.npy` / `queries.npy`
+- `docidx`: 0-based position into `doc_ids.npy` / `doclens.npy`
+- `rank`: 1-based, descending score
+- `score`: `sum over the query's query_lens[qidx] vectors of max over the document's vectors of the
+  dot product`, computed in fp32 with the fp16 document vectors upcast to fp32. Expansion vectors
+  are included in the sum. Printed to 6 decimals.
+
+No query id appears as a document id, so there are no self-matches.
+
+## Retrieval quality
+
+Sanity check of the vectors, not a leaderboard number: exact MaxSim over the full corpus scored
+against `qrels.test.tsv` with ir_measures.
+
+| nDCG@10 | Recall@100 | MRR@10 | MAP |
+|---|---|---|---|
+| 0.4352 | 0.9035 | 0.3692 | 0.3749 |
+
+## Loading
+
+```python
+import numpy as np
+
+documents = np.load("documents.npy", mmap_mode="r")      # [n_tokens, 96] float16
+doclens = np.load("doclens.npy")                         # [n_docs] int32
+offsets = np.concatenate([[0], np.cumsum(doclens)])
+doc_ids = np.load("doc_ids.npy")                         # [n_docs] str
+
+def document(i):
+    return documents[offsets[i]:offsets[i + 1]]           # [doclens[i], 96]
+
+queries = np.load("queries.npy")                         # [n_queries, 32, 96] float32
+query_lens = np.load("query_lens.npy")                   # [n_queries] int32
+query_ids = np.load("queries_ids.npy")                   # [n_queries] str
+
+def query(j):
+    return queries[j, :query_lens[j]]                     # [query_lens[j], 96]
+
+def maxsim(q, d):
+    return (q @ d.astype(np.float32).T).max(axis=1).sum()
+```
+
+## Validation
+
+Checks run by the exporter on the files exactly as written here:
+
+- ✅ file set — missing=[] extra=[]
+- ✅ documents.npy dtype/shape — <f2 (597909930, 96)
+- ✅ doclens.npy dtype/shape — <i4 (8841823,)
+- ✅ doc_ids.npy is a string array — <U7 (8841823,)
+- ✅ queries.npy dtype/shape — <f4 (6980, 32, 96)
+- ✅ query_lens.npy dtype/shape — <i4 (6980,)
+- ✅ queries_ids.npy is a string array — <U7 (6980,)
+- ✅ sum(doclens) == n_tokens — 597909930 vs 597909930
+- ✅ no empty documents — min doclen 4
+- ✅ len(doc_ids) == len(doclens) == corpus size — 8841823, 8841823, 8841823
+- ✅ doc_ids unique
+- ✅ query arrays aligned — 6980, 6980, 6980
+- ✅ doc and query dim agree — 96 / 96
+- ✅ token_ids.npy dtype/shape — <u4 (597909930,)
+- ✅ document vectors unit-norm (100k sample) — norm range [0.9995, 1.0006]
+- ✅ query vectors unit-norm — norm range [1.000000, 1.000000]
+- ✅ all vectors finite
+- ✅ gt_top100.tsv has k rows per query — 698000 rows, k=100
+- ✅ gt rows grouped by qidx with ranks 1..k and descending scores
+- ✅ gt indices in range
+
+## Provenance
+
+| | |
+|---|---|
+| exported | 2026-09-27 |
+| hardware | Tesla V100S-PCIE-32GB |

doc_ids.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0ac9bd7504d8569fb4e04b1dbb73c0a450e5dcb9b15112d9cb753bec0c374e9
+size 247571172

doclens.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb00cc3f34c5a065dededb5095edc446472f085205f6e74c18b2f19010f6b895
+size 35367420

documents.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ce6ddf05903066145277eabfca4dafc391c559c6d1685e1960e051c2de86696
+size 114798706688

gt_top100.tsv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3953fc221cff7b8b39c93a71878f865b3952f917be4dfa71cdb59f363b551526
+size 17892607

queries.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7194374728d621e80628c48f7cbfade1ba3bc3de593979f00dee6c0f0e8f71d
+size 85770368

queries_ids.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4018bf254630358c78ef1786defe075b6c0485d564836827f61521e500a6fd38
+size 195568

query_lens.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:009c17396fc151fd815942fca1227bdbd70fe8220a85322b1b2ebe8764cf32e2
+size 28048

token_ids.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:333003624c73a14f9324da58139d1af18adddb3f5992b67403d99e493085f664
+size 2391639848