feat(bedrock): add MiniMax provider support for AWS Bedrock (#4609)

tests/llms/test_aws_bedrock.py

@@ -220,6 +220,25 @@ class TestBuildInferenceConfig:
         cfg = llm._build_inference_config()
         assert cfg["maxTokens"] == 2000
 
+    def test_minimax_omits_top_p_when_explicitly_set(self, mock_boto3):
+        # MiniMax M2.x (reasoning model) rejects both temperature and topP simultaneously.
+        # Even when the user explicitly configures top_p, it must be omitted.
+        llm = _make_llm(
+            "minimax.minimax-m2.5",
+            mock_boto3,
+            temperature=0.1,
+            top_p=0.9,
+        )
+        cfg = llm._build_inference_config()
+        assert "temperature" in cfg
+        assert "topP" not in cfg, "topP must be absent for MiniMax reasoning models"
+
+    def test_minimax_only_temperature_by_default(self, mock_boto3):
+        llm = _make_llm("minimax.minimax-m2.5", mock_boto3, temperature=0.1)
+        cfg = llm._build_inference_config()
+        assert cfg["temperature"] == 0.1
+        assert "topP" not in cfg
+
 
 # ---------------------------------------------------------------------------
 # generate_response — Converse API call assertions
@@ -328,3 +347,80 @@ class TestGenerateResponseConverse:
 
         _, kwargs = mock_boto3.converse.call_args
         assert "topP" not in kwargs["inferenceConfig"]
+
+
+# ---------------------------------------------------------------------------
+# MiniMax provider
+# ---------------------------------------------------------------------------
+
+class TestMiniMaxProvider:
+    """Tests for MiniMax models via Bedrock Converse API."""
+
+    def test_extract_provider(self):
+        assert extract_provider("minimax.minimax-m2.5") == "minimax"
+        assert extract_provider("minimax.minimax-m2") == "minimax"
+
+    def test_generate_response_text_only(self, mock_boto3):
+        """Standard response: single text block."""
+        mock_boto3.converse.return_value = _converse_response("Hello!")
+        llm = _make_llm("minimax.minimax-m2.5", mock_boto3)
+
+        result = llm.generate_response([{"role": "user", "content": "say hi"}])
+
+        assert result == "Hello!"
+        _, kwargs = mock_boto3.converse.call_args
+        assert kwargs["modelId"] == "minimax.minimax-m2.5"
+        assert kwargs["messages"][0]["role"] == "user"
+        assert kwargs["messages"][0]["content"][0]["text"] == "say hi"
+
+    def test_generate_response_reasoning_model(self, mock_boto3):
+        """MiniMax M2.5 is a reasoning model: reasoningContent block comes before text."""
+        reasoning_response = {
+            "output": {
+                "message": {
+                    "content": [
+                        {"reasoningContent": {"reasoningText": {"text": "Let me think..."}}},
+                        {"text": "Hello!"},
+                    ]
+                }
+            }
+        }
+        mock_boto3.converse.return_value = reasoning_response
+        llm = _make_llm("minimax.minimax-m2.5", mock_boto3)
+
+        result = llm.generate_response([{"role": "user", "content": "say hi"}])
+
+        # Must skip reasoningContent and return the actual text block
+        assert result == "Hello!"
+
+    def test_inference_config(self, mock_boto3):
+        """inferenceConfig should include maxTokens and temperature; no topP."""
+        mock_boto3.converse.return_value = _converse_response()
+        llm = _make_llm("minimax.minimax-m2.5", mock_boto3, temperature=0.2, max_tokens=512)
+
+        llm.generate_response([{"role": "user", "content": "hi"}])
+
+        _, kwargs = mock_boto3.converse.call_args
+        assert kwargs["inferenceConfig"]["maxTokens"] == 512
+        assert kwargs["inferenceConfig"]["temperature"] == 0.2
+        assert "topP" not in kwargs["inferenceConfig"]
+
+    def test_system_prompt_passed_correctly(self, mock_boto3):
+        """System messages must be sent via top-level `system` param, not as a message role."""
+        mock_boto3.converse.return_value = _converse_response('{"facts": ["test"]}')
+        llm = _make_llm("minimax.minimax-m2.5", mock_boto3)
+
+        llm.generate_response([
+            {"role": "system", "content": "Return JSON only."},
+            {"role": "user", "content": "Extract facts from: test"},
+        ])
+
+        _, kwargs = mock_boto3.converse.call_args
+        # system prompt must be in top-level "system" key
+        assert "system" in kwargs
+        assert kwargs["system"][0]["text"] == "Return JSON only."
+        # messages list must NOT contain a system role entry
+        for msg in kwargs["messages"]:
+            assert msg["role"] != "system"
+        # user message must be present
+        assert kwargs["messages"][0]["role"] == "user"