[AMD] Add Kimi-K2.6 in nightly tests for MI30x and MI35x (#23848)
This commit is contained in:
Michael
@@ -0,0 +1,108 @@
|
||||
"""AMD Kimi-K2.6 GSM8K Completion Evaluation Test (8-GPU)
|
||||
|
||||
Tests moonshotai/Kimi-K2.6 with GSM8K few-shot benchmark on MI325.
|
||||
|
||||
Kimi-K2.6 shares the same architecture as Kimi-K2.5 (per the model card the
|
||||
deployment method is directly reused), so the AMD server arguments match the
|
||||
existing Kimi-K2.5 MI30x test.
|
||||
|
||||
Registry: nightly-amd-accuracy-8-gpu-kimi-k26 suite
|
||||
"""
|
||||
|
||||
import os
|
||||
import unittest
|
||||
from types import SimpleNamespace
|
||||
|
||||
import requests
|
||||
|
||||
from sglang.srt.utils import kill_process_tree
|
||||
from sglang.test.ci.ci_register import register_amd_ci
|
||||
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
|
||||
from sglang.test.test_utils import (
|
||||
DEFAULT_URL_FOR_TEST,
|
||||
CustomTestCase,
|
||||
is_in_ci,
|
||||
popen_launch_server,
|
||||
write_github_step_summary,
|
||||
)
|
||||
|
||||
# Register for AMD CI - Kimi K2.6 accuracy test (~60 min)
|
||||
register_amd_ci(
|
||||
est_time=3600, suite="nightly-amd-accuracy-8-gpu-kimi-k26", nightly=True
|
||||
)
|
||||
|
||||
KIMI_K26_MODEL_PATH = "moonshotai/Kimi-K2.6"
|
||||
SERVER_LAUNCH_TIMEOUT = 3600
|
||||
ACCURACY_THRESHOLD = 0.92
|
||||
TP_SIZE = 8
|
||||
|
||||
|
||||
class TestKimiK26EvalAMD(CustomTestCase):
|
||||
"""Kimi-K2.6 GSM8K Completion Evaluation Test for AMD MI325."""
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = KIMI_K26_MODEL_PATH
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
other_args = [
|
||||
"--tp",
|
||||
str(TP_SIZE),
|
||||
"--decode-attention-backend",
|
||||
"triton",
|
||||
"--prefill-attention-backend",
|
||||
"aiter",
|
||||
"--trust-remote-code",
|
||||
"--model-loader-extra-config",
|
||||
'{"enable_multithread_load": true}',
|
||||
]
|
||||
env = os.environ.copy()
|
||||
env["SGLANG_USE_AITER"] = "1"
|
||||
env["SGLANG_ROCM_FUSED_DECODE_MLA"] = "0"
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=SERVER_LAUNCH_TIMEOUT,
|
||||
other_args=other_args,
|
||||
env=env,
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_kimi_k26_gsm8k_accuracy(self):
|
||||
"""Test Kimi-K2.6 with GSM8K few-shot completion benchmark."""
|
||||
requests.get(self.base_url + "/flush_cache")
|
||||
|
||||
args = SimpleNamespace(
|
||||
num_shots=8,
|
||||
data_path=None,
|
||||
num_questions=1319,
|
||||
parallel=1319,
|
||||
max_new_tokens=512,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
acc = metrics["accuracy"]
|
||||
|
||||
passed = acc >= ACCURACY_THRESHOLD
|
||||
status = "✅ PASS" if passed else "❌ FAIL"
|
||||
print(f" accuracy={acc:.3f} threshold={ACCURACY_THRESHOLD} {status}")
|
||||
|
||||
if is_in_ci():
|
||||
summary = "### Kimi-K2.6 Model (MI325)\n\n"
|
||||
summary += "| Model | TP | Accuracy | Threshold | Status |\n"
|
||||
summary += "| ----- | -- | -------- | --------- | ------ |\n"
|
||||
summary += f"| {KIMI_K26_MODEL_PATH} | {TP_SIZE} | {acc:.3f} | {ACCURACY_THRESHOLD} | {status} |\n"
|
||||
write_github_step_summary(summary)
|
||||
|
||||
self.assertGreaterEqual(
|
||||
acc,
|
||||
ACCURACY_THRESHOLD,
|
||||
f"Kimi-K2.6 accuracy {acc:.3f} below threshold {ACCURACY_THRESHOLD}",
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,110 @@
|
||||
"""MI35x Kimi-K2.6 GSM8K Completion Evaluation Test (8-GPU)
|
||||
|
||||
Tests moonshotai/Kimi-K2.6 with GSM8K few-shot benchmark on MI35x.
|
||||
|
||||
Kimi-K2.6 shares the same architecture as Kimi-K2.5 (per the model card the
|
||||
deployment method is directly reused), so the AMD server arguments match the
|
||||
existing Kimi-K2.5 MI35x test.
|
||||
|
||||
Registry: nightly-amd-accuracy-8-gpu-mi35x-kimi-k26 suite
|
||||
"""
|
||||
|
||||
import os
|
||||
import unittest
|
||||
from types import SimpleNamespace
|
||||
|
||||
import requests
|
||||
|
||||
from sglang.srt.utils import kill_process_tree
|
||||
from sglang.test.ci.ci_register import register_amd_ci
|
||||
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
|
||||
from sglang.test.test_utils import (
|
||||
DEFAULT_URL_FOR_TEST,
|
||||
CustomTestCase,
|
||||
is_in_ci,
|
||||
popen_launch_server,
|
||||
write_github_step_summary,
|
||||
)
|
||||
|
||||
# Register for AMD CI - Kimi K2.6 accuracy test on MI35x (~90 min)
|
||||
register_amd_ci(
|
||||
est_time=5400, suite="nightly-amd-accuracy-8-gpu-mi35x-kimi-k26", nightly=True
|
||||
)
|
||||
|
||||
KIMI_K26_MODEL_PATH = "moonshotai/Kimi-K2.6"
|
||||
SERVER_LAUNCH_TIMEOUT = 5400
|
||||
ACCURACY_THRESHOLD = 0.92
|
||||
TP_SIZE = 8
|
||||
|
||||
|
||||
class TestKimiK26EvalMI35x(CustomTestCase):
|
||||
"""Kimi-K2.6 GSM8K Completion Evaluation Test for AMD MI35x."""
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
|
||||
def test_kimi_k26_gsm8k_accuracy(self):
|
||||
"""Test Kimi-K2.6 with GSM8K few-shot completion benchmark."""
|
||||
other_args = [
|
||||
"--tp",
|
||||
str(TP_SIZE),
|
||||
"--decode-attention-backend",
|
||||
"triton",
|
||||
"--prefill-attention-backend",
|
||||
"aiter",
|
||||
"--trust-remote-code",
|
||||
"--model-loader-extra-config",
|
||||
'{"enable_multithread_load": true}',
|
||||
"--watchdog-timeout",
|
||||
"1200",
|
||||
]
|
||||
env = os.environ.copy()
|
||||
env["SGLANG_USE_AITER"] = "1"
|
||||
env["SGLANG_ROCM_FUSED_DECODE_MLA"] = "0"
|
||||
|
||||
process = popen_launch_server(
|
||||
KIMI_K26_MODEL_PATH,
|
||||
self.base_url,
|
||||
timeout=SERVER_LAUNCH_TIMEOUT,
|
||||
other_args=other_args,
|
||||
env=env,
|
||||
)
|
||||
|
||||
try:
|
||||
requests.get(self.base_url + "/flush_cache")
|
||||
|
||||
args = SimpleNamespace(
|
||||
num_shots=8,
|
||||
data_path=None,
|
||||
num_questions=1319,
|
||||
parallel=1319,
|
||||
max_new_tokens=512,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
acc = metrics["accuracy"]
|
||||
|
||||
passed = acc >= ACCURACY_THRESHOLD
|
||||
status = "✅ PASS" if passed else "❌ FAIL"
|
||||
print(f" accuracy={acc:.3f} threshold={ACCURACY_THRESHOLD} {status}")
|
||||
|
||||
if is_in_ci():
|
||||
summary = "### Kimi-K2.6 Model (MI35x)\n\n"
|
||||
summary += "| Model | TP | Accuracy | Threshold | Status |\n"
|
||||
summary += "| ----- | -- | -------- | --------- | ------ |\n"
|
||||
summary += f"| {KIMI_K26_MODEL_PATH} | {TP_SIZE} | {acc:.3f} | {ACCURACY_THRESHOLD} | {status} |\n"
|
||||
write_github_step_summary(summary)
|
||||
|
||||
self.assertGreaterEqual(
|
||||
acc,
|
||||
ACCURACY_THRESHOLD,
|
||||
f"Kimi-K2.6 accuracy {acc:.3f} below threshold {ACCURACY_THRESHOLD}",
|
||||
)
|
||||
finally:
|
||||
kill_process_tree(process.pid)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,148 @@
|
||||
"""AMD Nightly performance benchmark for Kimi-K2.6 model.
|
||||
|
||||
This test benchmarks moonshotai/Kimi-K2.6 with TP=8 on MI325/MI300X.
|
||||
|
||||
Kimi-K2.6 shares the same architecture as Kimi-K2.5 (per the model card the
|
||||
deployment method is directly reused), so the AMD server arguments match the
|
||||
existing Kimi-K2.5 MI30x accuracy test (mixed aiter prefill + triton decode).
|
||||
|
||||
The model path can be configured via KIMI_K26_MODEL_PATH environment variable.
|
||||
|
||||
Registry: nightly-perf-8-gpu-kimi-k26 suite
|
||||
|
||||
Example usage:
|
||||
KIMI_K26_MODEL_PATH=moonshotai/Kimi-K2.6 python -m pytest test_kimi_k26_perf_amd.py -v
|
||||
"""
|
||||
|
||||
import os
|
||||
import unittest
|
||||
from typing import List
|
||||
|
||||
from sglang.test.ci.ci_register import register_amd_ci
|
||||
from sglang.test.nightly_bench_utils import BenchmarkResult
|
||||
from sglang.test.nightly_utils import NightlyBenchmarkRunner
|
||||
from sglang.test.test_utils import DEFAULT_URL_FOR_TEST, _parse_int_list_env
|
||||
|
||||
# Register for AMD CI - Kimi K2.6 perf benchmark (~90 min)
|
||||
register_amd_ci(est_time=5400, suite="nightly-perf-8-gpu-kimi-k26", nightly=True)
|
||||
|
||||
|
||||
def generate_simple_markdown_report(results: List[BenchmarkResult]) -> str:
|
||||
"""Generate a simplified markdown report without traces and cost columns.
|
||||
|
||||
Skips the first result if it's a warmup run (duplicate batch_size).
|
||||
"""
|
||||
model_header = results[0].model_path
|
||||
if results[0].run_name and results[0].run_name != "default":
|
||||
model_header += f" ({results[0].run_name})"
|
||||
|
||||
gpu_config = os.getenv("GPU_CONFIG", "MI325")
|
||||
if gpu_config:
|
||||
model_header += f" [{gpu_config}]"
|
||||
|
||||
summary = f"### {model_header}\n"
|
||||
summary += "| batch size | input len | latency (s) | input throughput (tok/s) | output throughput (tok/s) | ITL (ms) |\n"
|
||||
summary += "| ---------- | --------- | ----------- | ------------------------ | ------------------------- | -------- |\n"
|
||||
|
||||
report_results = (
|
||||
results[1:]
|
||||
if len(results) > 1 and results[0].batch_size == results[1].batch_size
|
||||
else results
|
||||
)
|
||||
|
||||
for result in report_results:
|
||||
itl = 1 / (result.output_throughput / result.batch_size) * 1000
|
||||
summary += f"| {result.batch_size} | {result.input_len} | {result.latency:.2f} | {result.input_throughput:.2f} | {result.output_throughput:.2f} | {itl:.2f} |\n"
|
||||
|
||||
return summary
|
||||
|
||||
|
||||
KIMI_K26_MODEL_PATH = os.environ.get("KIMI_K26_MODEL_PATH", "moonshotai/Kimi-K2.6")
|
||||
PROFILE_DIR = "performance_profiles_kimi_k26"
|
||||
|
||||
|
||||
class TestNightlyKimiK26Performance(unittest.TestCase):
|
||||
"""AMD Nightly performance benchmark for Kimi-K2.6 model.
|
||||
|
||||
Tests Kimi-K2.6 with TP=8 mixed-attention configuration on MI325/MI300X.
|
||||
"""
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.batch_sizes = [1, 8, 16, 64]
|
||||
cls.input_lens = tuple(_parse_int_list_env("NIGHTLY_INPUT_LENS", "4096"))
|
||||
cls.output_lens = tuple(_parse_int_list_env("NIGHTLY_OUTPUT_LENS", "512"))
|
||||
|
||||
# Kimi-K2.6 shares Kimi-K2.5's architecture: aiter for prefill,
|
||||
# triton for decode (aiter ASM MLA decode requires heads_per_gpu % 16,
|
||||
# but TP=8 with 64 heads gives 8 heads/GPU, so we use triton decode).
|
||||
cls.model_config = {
|
||||
"name": "default",
|
||||
"model_path": KIMI_K26_MODEL_PATH,
|
||||
"other_args": [
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"8",
|
||||
"--decode-attention-backend",
|
||||
"triton",
|
||||
"--prefill-attention-backend",
|
||||
"aiter",
|
||||
"--mem-fraction-static",
|
||||
"0.85",
|
||||
"--model-loader-extra-config",
|
||||
'{"enable_multithread_load": true}',
|
||||
"--watchdog-timeout",
|
||||
"1200",
|
||||
],
|
||||
"env_vars": {
|
||||
"SGLANG_USE_AITER": "1",
|
||||
"SGLANG_ROCM_FUSED_DECODE_MLA": "0",
|
||||
},
|
||||
}
|
||||
|
||||
cls.runner = NightlyBenchmarkRunner(PROFILE_DIR, cls.__name__, cls.base_url)
|
||||
cls.runner.setup_profile_directory()
|
||||
cls.runner.full_report = f"## {cls.__name__}\n"
|
||||
|
||||
def test_bench_kimi_k26(self):
|
||||
"""Run benchmark for Kimi-K2.6."""
|
||||
old_env = {}
|
||||
for key, value in self.model_config.get("env_vars", {}).items():
|
||||
old_env[key] = os.environ.get(key)
|
||||
os.environ[key] = value
|
||||
|
||||
try:
|
||||
result_tuple = self.runner.run_benchmark_for_model(
|
||||
model_path=self.model_config["model_path"],
|
||||
batch_sizes=self.batch_sizes,
|
||||
input_lens=self.input_lens,
|
||||
output_lens=self.output_lens,
|
||||
other_args=self.model_config["other_args"],
|
||||
variant=self.model_config["name"],
|
||||
extra_bench_args=["--trust-remote-code"],
|
||||
enable_profile=False,
|
||||
timeout=5400,
|
||||
)
|
||||
results = result_tuple[0]
|
||||
success = result_tuple[1]
|
||||
|
||||
if results:
|
||||
self.runner.full_report += (
|
||||
generate_simple_markdown_report(results) + "\n"
|
||||
)
|
||||
|
||||
self.assertTrue(
|
||||
success, f"Benchmark failed for {self.model_config['model_path']}"
|
||||
)
|
||||
finally:
|
||||
for key, value in old_env.items():
|
||||
if value is None:
|
||||
os.environ.pop(key, None)
|
||||
else:
|
||||
os.environ[key] = value
|
||||
self.runner.write_final_report()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,152 @@
|
||||
"""MI35x Nightly performance benchmark for Kimi-K2.6 model.
|
||||
|
||||
This test benchmarks moonshotai/Kimi-K2.6 with TP=8 on MI35x.
|
||||
|
||||
Kimi-K2.6 shares the same architecture as Kimi-K2.5 (per the model card the
|
||||
deployment method is directly reused), so the AMD server arguments match the
|
||||
existing Kimi-K2.5 MI35x accuracy test (mixed aiter prefill + triton decode).
|
||||
|
||||
The model path can be configured via KIMI_K26_MODEL_PATH environment variable.
|
||||
|
||||
Registry: nightly-perf-8-gpu-mi35x-kimi-k26 suite
|
||||
|
||||
Example usage:
|
||||
KIMI_K26_MODEL_PATH=moonshotai/Kimi-K2.6 python -m pytest test_kimi_k26_perf_mi35x.py -v
|
||||
"""
|
||||
|
||||
import os
|
||||
|
||||
os.environ.setdefault("HF_HOME", "/data2/models/huggingface")
|
||||
os.environ.setdefault("HF_HUB_CACHE", "/data2/models/huggingface/hub")
|
||||
|
||||
import unittest
|
||||
from typing import List
|
||||
|
||||
from sglang.test.ci.ci_register import register_amd_ci
|
||||
from sglang.test.nightly_bench_utils import BenchmarkResult
|
||||
from sglang.test.nightly_utils import NightlyBenchmarkRunner
|
||||
from sglang.test.test_utils import DEFAULT_URL_FOR_TEST, _parse_int_list_env
|
||||
|
||||
# Register for AMD CI - Kimi K2.6 perf benchmark on MI35x (~90 min)
|
||||
register_amd_ci(est_time=5400, suite="nightly-perf-8-gpu-mi35x-kimi-k26", nightly=True)
|
||||
|
||||
|
||||
def generate_simple_markdown_report(results: List[BenchmarkResult]) -> str:
|
||||
"""Generate a simplified markdown report without traces and cost columns.
|
||||
|
||||
Skips the first result if it's a warmup run (duplicate batch_size).
|
||||
"""
|
||||
model_header = results[0].model_path
|
||||
if results[0].run_name and results[0].run_name != "default":
|
||||
model_header += f" ({results[0].run_name})"
|
||||
|
||||
gpu_config = os.getenv("GPU_CONFIG", "MI35x")
|
||||
if gpu_config:
|
||||
model_header += f" [{gpu_config}]"
|
||||
|
||||
summary = f"### {model_header}\n"
|
||||
summary += "| batch size | input len | latency (s) | input throughput (tok/s) | output throughput (tok/s) | ITL (ms) |\n"
|
||||
summary += "| ---------- | --------- | ----------- | ------------------------ | ------------------------- | -------- |\n"
|
||||
|
||||
report_results = (
|
||||
results[1:]
|
||||
if len(results) > 1 and results[0].batch_size == results[1].batch_size
|
||||
else results
|
||||
)
|
||||
|
||||
for result in report_results:
|
||||
itl = 1 / (result.output_throughput / result.batch_size) * 1000
|
||||
summary += f"| {result.batch_size} | {result.input_len} | {result.latency:.2f} | {result.input_throughput:.2f} | {result.output_throughput:.2f} | {itl:.2f} |\n"
|
||||
|
||||
return summary
|
||||
|
||||
|
||||
KIMI_K26_MODEL_PATH = os.environ.get("KIMI_K26_MODEL_PATH", "moonshotai/Kimi-K2.6")
|
||||
PROFILE_DIR = "performance_profiles_kimi_k26_mi35x"
|
||||
|
||||
|
||||
class TestNightlyKimiK26PerformanceMI35x(unittest.TestCase):
|
||||
"""MI35x Nightly performance benchmark for Kimi-K2.6 model.
|
||||
|
||||
Tests Kimi-K2.6 with TP=8 mixed-attention configuration on MI35x.
|
||||
"""
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.batch_sizes = [1, 8, 16, 64]
|
||||
cls.input_lens = tuple(_parse_int_list_env("NIGHTLY_INPUT_LENS", "4096"))
|
||||
cls.output_lens = tuple(_parse_int_list_env("NIGHTLY_OUTPUT_LENS", "512"))
|
||||
|
||||
# Kimi-K2.6 shares Kimi-K2.5's architecture: aiter for prefill,
|
||||
# triton for decode (aiter ASM MLA decode requires heads_per_gpu % 16,
|
||||
# but TP=8 with 64 heads gives 8 heads/GPU, so we use triton decode).
|
||||
cls.model_config = {
|
||||
"name": "default",
|
||||
"model_path": KIMI_K26_MODEL_PATH,
|
||||
"other_args": [
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"8",
|
||||
"--decode-attention-backend",
|
||||
"triton",
|
||||
"--prefill-attention-backend",
|
||||
"aiter",
|
||||
"--mem-fraction-static",
|
||||
"0.85",
|
||||
"--model-loader-extra-config",
|
||||
'{"enable_multithread_load": true}',
|
||||
"--watchdog-timeout",
|
||||
"1200",
|
||||
],
|
||||
"env_vars": {
|
||||
"SGLANG_USE_AITER": "1",
|
||||
"SGLANG_ROCM_FUSED_DECODE_MLA": "0",
|
||||
},
|
||||
}
|
||||
|
||||
cls.runner = NightlyBenchmarkRunner(PROFILE_DIR, cls.__name__, cls.base_url)
|
||||
cls.runner.setup_profile_directory()
|
||||
cls.runner.full_report = f"## {cls.__name__}\n"
|
||||
|
||||
def test_bench_kimi_k26(self):
|
||||
"""Run benchmark for Kimi-K2.6."""
|
||||
old_env = {}
|
||||
for key, value in self.model_config.get("env_vars", {}).items():
|
||||
old_env[key] = os.environ.get(key)
|
||||
os.environ[key] = value
|
||||
|
||||
try:
|
||||
result_tuple = self.runner.run_benchmark_for_model(
|
||||
model_path=self.model_config["model_path"],
|
||||
batch_sizes=self.batch_sizes,
|
||||
input_lens=self.input_lens,
|
||||
output_lens=self.output_lens,
|
||||
other_args=self.model_config["other_args"],
|
||||
variant=self.model_config["name"],
|
||||
extra_bench_args=["--trust-remote-code"],
|
||||
enable_profile=False,
|
||||
timeout=5400,
|
||||
)
|
||||
results = result_tuple[0]
|
||||
success = result_tuple[1]
|
||||
|
||||
if results:
|
||||
self.runner.full_report += (
|
||||
generate_simple_markdown_report(results) + "\n"
|
||||
)
|
||||
|
||||
self.assertTrue(
|
||||
success, f"Benchmark failed for {self.model_config['model_path']}"
|
||||
)
|
||||
finally:
|
||||
for key, value in old_env.items():
|
||||
if value is None:
|
||||
os.environ.pop(key, None)
|
||||
else:
|
||||
os.environ[key] = value
|
||||
self.runner.write_final_report()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user