Support FP8 E4M3 KV Cache (#2786)

Co-authored-by: root <bjmsong@126.com>

test/srt/kv_cache_scales_llama3_1_8b.json

@@ -0,0 +1,42 @@
+{
+    "model_type": "llama",
+    "kv_cache": {
+        "dtype": "float8_e4m3fn",
+        "scaling_factor": {
+            "0": {
+                "0": 1,
+                "1": 1,
+                "2": 1,
+                "3": 1,
+                "4": 1,
+                "5": 1,
+                "6": 1,
+                "7": 1,
+                "8": 1,
+                "9": 1,
+                "10": 1,
+                "11": 1,
+                "12": 1,
+                "13": 1,
+                "14": 1,
+                "15": 1,
+                "16": 1,
+                "17": 1,
+                "18": 1,
+                "19": 1,
+                "20": 1,
+                "21": 1,
+                "22": 1,
+                "23": 1,
+                "24": 1,
+                "25": 1,
+                "26": 1,
+                "27": 1,
+                "28": 1,
+                "29": 1,
+                "30": 1,
+                "31": 1
+            }
+        }
+    }
+}

test/srt/test_fp8_kvcache.py

@@ -0,0 +1,64 @@
+import os
+import unittest
+from types import SimpleNamespace
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.run_eval import run_eval
+from sglang.test.test_utils import (
+    DEFAULT_MODEL_NAME_FOR_TEST,
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    popen_launch_server,
+)
+
+
+class TestFp8Kvcache(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = DEFAULT_MODEL_NAME_FOR_TEST
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        dirpath = os.path.dirname(__file__)
+        config_file = os.path.join(dirpath, "kv_cache_scales_llama3_8b_chat.json")
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=[
+                "--kv-cache-dtype",
+                "fp8_e4m3",
+                "--quantization-param-path",
+                config_file,
+            ],
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_mgsm_en(self):
+        args = SimpleNamespace(
+            base_url=self.base_url,
+            model=self.model,
+            eval_name="mgsm_en",
+            num_examples=None,
+            num_threads=1024,
+        )
+
+        metrics = run_eval(args)
+        self.assertGreater(metrics["score"], 0.835)
+
+    def test_mmlu(self):
+        args = SimpleNamespace(
+            base_url=self.base_url,
+            model=self.model,
+            eval_name="mmlu",
+            num_examples=64,
+            num_threads=32,
+        )
+
+        metrics = run_eval(args)
+        self.assertGreaterEqual(metrics["score"], 0.65)
+
+
+if __name__ == "__main__":
+    unittest.main()