gen-ai mock server: serve Cohere v2 rerank

Add POST /v2/rerank to the Cohere mock. It returns the SDK's
V2RerankResponse shape, ranking the request's documents in order and
trimming to top_n, so the same request always gets the same bytes.

Author: Krishnachaitanyakc

tools/gen-ai/mock-server/README.md

@@ -23,7 +23,7 @@ The conformance runner starts it for you: a directory declares it under
 | `assistants` | OpenAI Assistants, threads and runs |
 | `bedrock` | `converse`, `converse-stream`, `invoke` |
 | `bedrock_agent`, `bedrock_agentcore` | agent invocation and the memory APIs |
-| `cohere` | `/v1` and `/v2` chat and embed |
+| `cohere` | `/v1` and `/v2` chat and embed, `/v2` rerank |
 | `google_genai` | Gemini `generateContent` / streaming, and the Vertex `projects/…` paths |
 | `mistral` | chat completions, FIM and embeddings, under `/mistral/v1/…` |
 | `ollama` | `/api/chat`, including tool use and NDJSON streaming, and `/api/embed` |

tools/gen-ai/mock-server/src/genai_mock_server/cohere.py

@@ -197,3 +197,21 @@ def cohere_embed():
 @bp.route("/v1/embed", methods=["POST"])
 def cohere_embed_v1():
     return _embed_response(request.get_json(silent=True) or {}, "1")
+
+
+@bp.route("/v2/rerank", methods=["POST"])
+def cohere_rerank():
+    body = request.get_json(silent=True) or {}
+    documents = body.get("documents") or []
+    top_n = body.get("top_n") or len(documents)
+    return {
+        "id": "cohere-rerank-mock-001",
+        "results": [
+            {"index": index, "relevance_score": round(0.99 / (index + 1), 4)}
+            for index in range(min(top_n, len(documents)))
+        ],
+        "meta": {
+            "api_version": {"version": "2"},
+            "billed_units": {"search_units": 1},
+        },
+    }

tools/gen-ai/mock-server/tests/test_mock_server.py

@@ -122,6 +122,17 @@
         "/v2/embed",
         {"model": "embed-v4.0", "texts": ["hi", "there"], "input_type": "search_document"},
     ),
+    (
+        "cohere-rerank",
+        "post",
+        "/v2/rerank",
+        {
+            "model": "rerank-v3.5",
+            "query": "hi",
+            "documents": ["one", "two", "three"],
+            "top_n": 2,
+        },
+    ),
     (
         "mistral-chat",
         "post",
@@ -353,6 +364,30 @@ def test_embeddings_survive_an_unusable_dimension_count(client):
     assert len(response.json["data"][0]["embedding"]) == 256
 
 
+def test_cohere_rerank_ranks_within_the_requested_candidates(client):
+    def rerank(**extra):
+        response = client.post(
+            "/v2/rerank",
+            json={
+                "model": "rerank-v3.5",
+                "query": "hi",
+                "documents": ["one", "two", "three"],
+                **extra,
+            },
+        )
+        assert response.status_code == 200
+        return response.json["results"]
+
+    top_two = rerank(top_n=2)
+    assert [result["index"] for result in top_two] == [0, 1]
+    scores = [result["relevance_score"] for result in top_two]
+    assert scores == sorted(scores, reverse=True)
+    assert all(0 < score <= 1 for score in scores)
+
+    assert len(rerank()) == 3
+    assert len(rerank(top_n=10)) == 3
+
+
 def test_chat_reports_the_service_tier_that_served_the_request(client):
     def tier(requested):
         response = client.post(