add test_hy_v3.py and test_ministral3.py

Author: ZX-ModelCloud

tests/models/test_hunyuan_v1_dense.py

@@ -6,7 +6,7 @@
 from model_test import ModelTest
 
 
-class TestNemotronUltra(ModelTest):
+class TestHunyuanV1Dense(ModelTest):
     NATIVE_MODEL_ID = "/monster/data/model/HY-MT1.5-1.8B" # tencent/HY-MT1.5-1.8B
     EVAL_TASKS_SLOW = {
         "arc_challenge": {
@@ -24,6 +24,5 @@ class TestNemotronUltra(ModelTest):
     }
     EVAL_TASKS_FAST = ModelTest.derive_fast_eval_tasks(EVAL_TASKS_SLOW)
 
-    def test_nemotron_ultra(self):
-        # self.quantize_and_evaluate()
-        print(self.evaluate_model(self.SAVE_PATH))
+    def test_hunyuan_v1_dense(self):
+        self.quantize_and_evaluate()

tests/models/test_hunyuan_v1_moe.py

@@ -6,7 +6,7 @@
 from model_test import ModelTest
 
 
-class TestNemotronUltra(ModelTest):
+class TestHunyuanV1MoE(ModelTest):
     NATIVE_MODEL_ID = "/monster/data/model/Hunyuan-A13B-Instruct" # tencent/Hunyuan-A13B-Instruct
     EVAL_TASKS_SLOW = {
         "arc_challenge": {
@@ -24,5 +24,5 @@ class TestNemotronUltra(ModelTest):
     }
     EVAL_TASKS_FAST = ModelTest.derive_fast_eval_tasks(EVAL_TASKS_SLOW)
 
-    def test_nemotron_ultra(self):
+    def test_hunyuan_v1_moe(self):
         self.quantize_and_evaluate()

tests/models/test_hy_v3.py

@@ -0,0 +1,28 @@
+# SPDX-FileCopyrightText: 2024-2025 ModelCloud.ai
+# SPDX-FileCopyrightText: 2024-2025 qubitium@modelcloud.ai
+# SPDX-License-Identifier: Apache-2.0
+# Contact: qubitium@modelcloud.ai, x.com/qubitium
+
+from model_test import ModelTest
+
+
+class TestHyV3(ModelTest):
+    NATIVE_MODEL_ID = "/monster/data/model/Hy-MT2-30B-A3B" # tencent/Hy-MT2-30B-A3B
+    EVAL_TASKS_SLOW = {
+        "arc_challenge": {
+            "chat_template": True,
+            "acc": {"value": 0.3182, "floor_pct": 0.36},
+            "acc_norm": {"value": 0.3472, "floor_pct": 0.36},
+        },
+        "mmlu_stem": {
+            "chat_template": False,
+            "acc": {
+                "value": 0.4024,
+                "floor_pct": 0.04,
+            },
+        },
+    }
+    EVAL_TASKS_FAST = ModelTest.derive_fast_eval_tasks(EVAL_TASKS_SLOW)
+
+    def test_hy_v3(self):
+        self.quantize_and_evaluate()

tests/models/test_ministral3.py

@@ -0,0 +1,29 @@
+# SPDX-FileCopyrightText: 2024-2025 ModelCloud.ai
+# SPDX-FileCopyrightText: 2024-2025 qubitium@modelcloud.ai
+# SPDX-License-Identifier: Apache-2.0
+# Contact: qubitium@modelcloud.ai, x.com/qubitium
+
+from model_test import ModelTest
+
+
+class TestMinistral3(ModelTest):
+    NATIVE_MODEL_ID = "/monster/data/model/Ministral-3-3B-Instruct-2512-TextOnly" # "Aratako/Ministral-3-3B-Instruct-2512-TextOnly"
+    NATIVE_ARC_CHALLENGE_ACC = 0.4974
+    NATIVE_ARC_CHALLENGE_ACC_NORM = 0.5256
+    NATIVE_ARC_CHALLENGE_ACC_SLOW = NATIVE_ARC_CHALLENGE_ACC
+    NATIVE_ARC_CHALLENGE_ACC_NORM_SLOW = NATIVE_ARC_CHALLENGE_ACC_NORM
+    NATIVE_ARC_CHALLENGE_ACC_FAST = NATIVE_ARC_CHALLENGE_ACC_SLOW
+    NATIVE_ARC_CHALLENGE_ACC_NORM_FAST = NATIVE_ARC_CHALLENGE_ACC_NORM_SLOW
+    TRUST_REMOTE_CODE = False
+    EVAL_BATCH_SIZE = 6
+    EVAL_TASKS_SLOW = {
+        "arc_challenge": {
+            "chat_template": False,
+            "acc": {"value": NATIVE_ARC_CHALLENGE_ACC},
+            "acc_norm": {"value": NATIVE_ARC_CHALLENGE_ACC_NORM},
+        },
+    }
+    EVAL_TASKS_FAST = ModelTest.derive_fast_eval_tasks(EVAL_TASKS_SLOW)
+
+    def test_ministral3(self):
+        self.quantize_and_evaluate()