Skip to content

test: Add fp8kv to DS-v3-lite integration tests. #3950

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 16 commits into from
May 9, 2025
Merged

test: Add fp8kv to DS-v3-lite integration tests. #3950

Changes from 2 commits
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
0b4d2f9
Add fp8 kv cache tests to DSV3-Lite integration tests.
bobboli Apr 29, 2025
4427f1a
Refactor. Make fp8kv parallel to attention_dp, overlap_scheduler and …
bobboli Apr 29, 2025
b9a9884
Update gsm8k.
bobboli Apr 30, 2025
4da1b2f
Merge branch 'main' into fp8_mla_gsm8k
bobboli Apr 30, 2025
8f0453f
Update CI list.
bobboli Apr 30, 2025
43fcd3c
Update TestDeepSeekR1.
bobboli Apr 30, 2025
8935a4d
Fix test list.
bobboli Apr 30, 2025
1b85a30
Need quant_config besides pytorch_config.
bobboli Apr 30, 2025
020a739
Merge branch 'main' into fp8_mla_gsm8k
bobboli May 5, 2025
9d11463
Update waive list (bug 5239087).
bobboli May 5, 2025
736bbbc
Update waive list.
bobboli May 5, 2025
c884b12
Correct test name.
bobboli May 6, 2025
cdd1309
Merge branch 'main' into fp8_mla_gsm8k
bobboli May 6, 2025
ab93ae6
Merge branch 'main' into fp8_mla_gsm8k
bobboli May 8, 2025
9610de6
Update waive list.
bobboli May 8, 2025
77c3379
Merge branch 'main' into fp8_mla_gsm8k
syuoni May 9, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
160 changes: 117 additions & 43 deletions tests/integration/defs/accuracy/test_llm_api_pytorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -329,126 +329,200 @@ def test_bfloat16_4gpus(self, tp_size, pp_size, ep_size, mtp_nextn,
task.evaluate(llm)

@pytest.mark.skip_device_not_contain(["H100"])
@parametrize_with_ids("attention_dp,cuda_graph,overlap_scheduler",
[(False, False, False), (True, False, False),
(False, True, False), (False, False, True),
(True, True, True)])
@parametrize_with_ids("fp8kv,attention_dp,cuda_graph,overlap_scheduler",
[(False, False, False, False),
(True, False, False, False),
(False, True, False, False),
(False, False, True, False),
(False, False, False, True),
(True, True, True, True)])
@parametrize_with_ids("mtp_nextn", [None, 2])
def test_fp8_block_scales(self, mtp_nextn, attention_dp, cuda_graph,
def test_fp8_block_scales(self, mtp_nextn, fp8kv, attention_dp, cuda_graph,
overlap_scheduler):
# OOM on H100 with default free_gpu_memory_fraction=0.9
kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.8)
pytorch_config = PyTorchConfig(
enable_overlap_scheduler=overlap_scheduler,
use_cuda_graph=cuda_graph)

quant_config = QuantConfig()
quant_config.quant_algo = QuantAlgo.FP8_BLOCK_SCALES
if fp8kv:
quant_config.kv_cache_quant_algo = QuantAlgo.FP8
pytorch_config.kv_cache_dtype = "fp8"

if mtp_nextn is not None and mtp_nextn > 0:
mtp_config = MTPDecodingConfig(num_nextn_predict_layers=mtp_nextn)
else:
mtp_config = None

llm = LLM(f"{llm_models_root()}/DeepSeek-V3-Lite/fp8",
kv_cache_config=kv_cache_config,
pytorch_backend_config=pytorch_config,
quant_config=quant_config,
enable_attention_dp=attention_dp,
speculative_config=mtp_config)

assert llm.args.quant_config.quant_algo == QuantAlgo.FP8_BLOCK_SCALES
if fp8kv:
assert llm.args.quant_config.kv_cache_quant_algo == QuantAlgo.FP8

with llm:
task = CnnDailymail(self.MODEL_NAME)
task.evaluate(llm)
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
if attention_dp and cuda_graph and overlap_scheduler:
# No need to run these tests for fp8kv
if not fp8kv:
task = CnnDailymail(self.MODEL_NAME)
task.evaluate(llm)
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
# Run GSM8K for fp8kv, or if all the other optimizations are enabled
if fp8kv or (attention_dp and cuda_graph and overlap_scheduler):
task = GSM8K(self.MODEL_NAME)
task.evaluate(llm)

@pytest.mark.skip_less_device(4)
@pytest.mark.skip_device_not_contain(["H100"])
@parametrize_with_ids("attention_dp,cuda_graph,overlap_scheduler",
[(False, False, False), (True, False, False),
(False, True, False), (False, False, True),
(True, True, True)])
@parametrize_with_ids("fp8kv,attention_dp,cuda_graph,overlap_scheduler",
[(False, False, False, False),
(True, False, False, False),
(False, True, False, False),
(False, False, True, False),
(False, False, False, True),
(True, True, True, True)])
@parametrize_with_ids("mtp_nextn", [None, 2])
@pytest.mark.parametrize("tp_size,pp_size,ep_size", [(4, 1, 1), (4, 1, 4),
(2, 2, 1), (1, 4, 1)],
ids=["tp4", "ep4", "tp2pp2", "pp4"])
def test_fp8_block_scales_4gpus(self, tp_size, pp_size, ep_size, mtp_nextn,
attention_dp, cuda_graph,
overlap_scheduler):
def test_fp8_block_scales_4gpus(self, mtp_nextn, fp8kv, attention_dp,
cuda_graph, overlap_scheduler, tp_size,
pp_size, ep_size):
# OOM on H100 with default free_gpu_memory_fraction=0.9
kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.8)
pytorch_config = PyTorchConfig(
enable_overlap_scheduler=overlap_scheduler,
use_cuda_graph=cuda_graph)

quant_config = QuantConfig()
quant_config.quant_algo = QuantAlgo.FP8_BLOCK_SCALES
if fp8kv:
quant_config.kv_cache_quant_algo = QuantAlgo.FP8
pytorch_config.kv_cache_dtype = "fp8"

if mtp_nextn is not None and mtp_nextn > 0:
mtp_config = MTPDecodingConfig(num_nextn_predict_layers=mtp_nextn)
else:
mtp_config = None

llm = LLM(f"{llm_models_root()}/DeepSeek-V3-Lite/fp8",
tensor_parallel_size=tp_size,
pipeline_parallel_size=pp_size,
moe_expert_parallel_size=ep_size,
kv_cache_config=kv_cache_config,
pytorch_backend_config=pytorch_config,
quant_config=quant_config,
enable_attention_dp=attention_dp,
speculative_config=mtp_config)

assert llm.args.quant_config.quant_algo == QuantAlgo.FP8_BLOCK_SCALES
if fp8kv:
assert llm.args.quant_config.kv_cache_quant_algo == QuantAlgo.FP8

with llm:
task = CnnDailymail(self.MODEL_NAME)
task.evaluate(llm)
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
if attention_dp and cuda_graph and overlap_scheduler:
# No need to run these tests for fp8kv
if not fp8kv:
task = CnnDailymail(self.MODEL_NAME)
task.evaluate(llm)
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
# Run GSM8K for fp8kv, or if all the other optimizations are enabled
if fp8kv or (attention_dp and cuda_graph and overlap_scheduler):
task = GSM8K(self.MODEL_NAME)
task.evaluate(llm)

@skip_pre_blackwell
@parametrize_with_ids("attention_dp,cuda_graph,overlap_scheduler",
[(False, False, False), (True, False, False),
(False, True, False), (False, False, True),
(True, True, True)])
def test_nvfp4(self, attention_dp, cuda_graph, overlap_scheduler):
@parametrize_with_ids("fp8kv,attention_dp,cuda_graph,overlap_scheduler",
[(False, False, False, False),
(True, False, False, False),
(False, True, False, False),
(False, False, True, False),
(False, False, False, True),
(True, True, True, True)])
def test_nvfp4(self, fp8kv, attention_dp, cuda_graph, overlap_scheduler):
pytorch_config = PyTorchConfig(
enable_overlap_scheduler=overlap_scheduler,
use_cuda_graph=cuda_graph)

quant_config = QuantConfig()
quant_config.quant_algo = QuantAlgo.NVFP4
if fp8kv:
quant_config.kv_cache_quant_algo = QuantAlgo.FP8
pytorch_config.kv_cache_dtype = "fp8"

llm = LLM(f"{llm_models_root()}/DeepSeek-V3-Lite/nvfp4_moe_only",
pytorch_backend_config=pytorch_config,
quant_config=quant_config,
enable_attention_dp=attention_dp)

assert llm.args.quant_config.quant_algo == QuantAlgo.NVFP4
if fp8kv:
assert llm.args.quant_config.kv_cache_quant_algo == QuantAlgo.FP8

with llm:
task = CnnDailymail(self.MODEL_NAME)
task.evaluate(llm)
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
if attention_dp and cuda_graph and overlap_scheduler:
# No need to run these tests for fp8kv
if not fp8kv:
task = CnnDailymail(self.MODEL_NAME)
task.evaluate(llm)
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
# Run GSM8K for fp8kv, or if all the other optimizations are enabled
if fp8kv or (attention_dp and cuda_graph and overlap_scheduler):
task = GSM8K(self.MODEL_NAME)
task.evaluate(llm)

@pytest.mark.skip_less_device(4)
@skip_pre_blackwell
@parametrize_with_ids("attention_dp,cuda_graph,overlap_scheduler",
[(False, False, False), (True, False, False),
(False, True, False), (False, False, True),
(True, True, True)])
@parametrize_with_ids("fp8kv,attention_dp,cuda_graph,overlap_scheduler",
[(False, False, False, False),
(True, False, False, False),
(False, True, False, False),
(False, False, True, False),
(False, False, False, True),
(True, True, True, True)])
@pytest.mark.parametrize("tp_size,pp_size,ep_size", [(4, 1, 1), (4, 1, 4),
(2, 2, 1), (1, 4, 1)],
ids=["tp4", "ep4", "tp2pp2", "pp4"])
def test_nvfp4_4gpus(self, tp_size, pp_size, ep_size, attention_dp,
cuda_graph, overlap_scheduler):
def test_nvfp4_4gpus(self, fp8kv, attention_dp, cuda_graph,
overlap_scheduler, tp_size, pp_size, ep_size):
pytorch_config = PyTorchConfig(
enable_overlap_scheduler=overlap_scheduler,
use_cuda_graph=cuda_graph)

quant_config = QuantConfig()
quant_config.quant_algo = QuantAlgo.NVFP4
if fp8kv:
quant_config.kv_cache_quant_algo = QuantAlgo.FP8
pytorch_config.kv_cache_dtype = "fp8"

llm = LLM(f"{llm_models_root()}/DeepSeek-V3-Lite/nvfp4_moe_only",
tensor_parallel_size=tp_size,
pipeline_parallel_size=pp_size,
moe_expert_parallel_size=ep_size,
pytorch_backend_config=pytorch_config,
quant_config=quant_config,
enable_attention_dp=attention_dp)

assert llm.args.quant_config.quant_algo == QuantAlgo.NVFP4
if fp8kv:
assert llm.args.quant_config.kv_cache_quant_algo == QuantAlgo.FP8

with llm:
task = CnnDailymail(self.MODEL_NAME)
task.evaluate(llm)
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
if attention_dp and cuda_graph and overlap_scheduler:
# No need to run these tests for fp8kv
if not fp8kv:
task = CnnDailymail(self.MODEL_NAME)
task.evaluate(llm)
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
# Run GSM8K for fp8kv, or if all the other optimizations are enabled
if fp8kv or (attention_dp and cuda_graph and overlap_scheduler):
task = GSM8K(self.MODEL_NAME)
task.evaluate(llm)

Expand Down