zero v1 — M7 lookup table, run p35w-2m-s2500

README.md

@@ -72,6 +72,39 @@ print(docs[int(np.argmax(scores))])
 That asymmetry is the point: `doc_model` runs once per document, in the cloud. `enc` runs on
 every query, on the device, and costs almost nothing.
 
+## Using it with Qdrant
+
+The output is an ordinary 1024-d dense vector, so no special handling is needed. Both sides are
+L2-normalized, which means `DOT` ranks identically to `COSINE` and is cheaper.
+
+```python
+from qdrant_client import QdrantClient, models
+
+client = QdrantClient(":memory:")           # or your cluster
+client.create_collection("docs", vectors_config=models.VectorParams(
+    size=1024, distance=models.Distance.DOT))
+client.upsert("docs", points=[
+    models.PointStruct(id=i, vector=D[i].tolist(), payload={"text": t})
+    for i, t in enumerate(docs)])           # D from the document encoder above
+
+hits = client.query_points("docs", query=enc.encode([q])[0].tolist(), limit=5).points
+```
+
+**On the edge, put the table in the store too.** A second collection holds one point per vocab
+row, created with `hnsw_config=models.HnswConfigDiff(m=0)` — it is retrieve-by-id only, and
+indexing it inflated the shard from 466 MB to 1.82 GB for no benefit. The query path becomes
+tokenize → fetch rows by id → pool → search, with no model weights in your process at all.
+
+**If you fuse with BM25, the fusion rule matters.** The system that ties OpenSearch (0.4911)
+uses **convex score fusion at w=0.8**, not RRF: on development sets RRF scored 0.5504 against
+convex's 0.5727. Qdrant's native `Fusion.RRF` is therefore a *different, weaker* operating point,
+not the published one — combine the scores yourself to reproduce it. Dense-only in Qdrant
+reproduces 0.4339 exactly.
+
+Scalar-quantizing the document index to int8 halves it to ~1.02 GB per 1M vectors, but **that
+was never measured for quality here** — treat it as untested. (The int8 quality-free result
+above is about the *query table*, not the document index.)
+
 ### The rule, if you reimplement it
 
 Tokenize with the bundled WordPiece tokenizer (`add_special_tokens=True`, truncate at 512,