[CI][ROCm] Mirror large-model GSM8K evaluations on MI355 (#58683)
Signed-off-by: Andreas Karatzas <andreas.karatzas@protonmail.com> Co-authored-by: Codex <noreply@openai.com>
This commit is contained in:
co-authored by
Codex
parent
d86cf91240
commit
0ae480ff1a
@@ -0,0 +1,17 @@
|
||||
model_name: "deepseek-ai/DeepSeek-R1"
|
||||
accuracy_threshold: 0.95
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
startup_max_wait_seconds: 1200
|
||||
env:
|
||||
VLLM_ROCM_USE_AITER: "1"
|
||||
VLLM_ROCM_USE_AITER_MLA: "1"
|
||||
VLLM_ROCM_USE_AITER_MOE: "1"
|
||||
server_args: >-
|
||||
--enforce-eager
|
||||
--max-model-len 4096
|
||||
--data-parallel-size 4
|
||||
--enable-expert-parallel
|
||||
--attention-backend ROCM_AITER_MLA
|
||||
--moe-backend aiter
|
||||
--speculative-config '{"method":"mtp","num_speculative_tokens":3}'
|
||||
@@ -0,0 +1,17 @@
|
||||
model_name: "deepseek-ai/DeepSeek-R1"
|
||||
accuracy_threshold: 0.95
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
startup_max_wait_seconds: 1200
|
||||
env:
|
||||
VLLM_ROCM_USE_AITER: "1"
|
||||
VLLM_ROCM_USE_AITER_MLA: "1"
|
||||
VLLM_ROCM_USE_AITER_MOE: "1"
|
||||
server_args: >-
|
||||
--enforce-eager
|
||||
--max-model-len 4096
|
||||
--tensor-parallel-size 4
|
||||
--enable-expert-parallel
|
||||
--attention-backend ROCM_AITER_MLA
|
||||
--moe-backend aiter
|
||||
--speculative-config '{"method":"mtp","num_speculative_tokens":3}'
|
||||
@@ -0,0 +1,17 @@
|
||||
model_name: "deepseek-ai/DeepSeek-V3.2"
|
||||
accuracy_threshold: 0.95
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
startup_max_wait_seconds: 1200
|
||||
env:
|
||||
VLLM_ROCM_USE_AITER: "1"
|
||||
VLLM_ROCM_USE_AITER_MLA: "1"
|
||||
VLLM_ROCM_USE_AITER_MOE: "1"
|
||||
server_args: >-
|
||||
--enforce-eager
|
||||
--max-model-len 4096
|
||||
--data-parallel-size 4
|
||||
--enable-expert-parallel
|
||||
--attention-backend ROCM_AITER_MLA_SPARSE
|
||||
--moe-backend aiter
|
||||
--speculative-config '{"method":"mtp","num_speculative_tokens":3}'
|
||||
@@ -0,0 +1,17 @@
|
||||
model_name: "deepseek-ai/DeepSeek-V3.2"
|
||||
accuracy_threshold: 0.95
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
startup_max_wait_seconds: 1200
|
||||
env:
|
||||
VLLM_ROCM_USE_AITER: "1"
|
||||
VLLM_ROCM_USE_AITER_MLA: "1"
|
||||
VLLM_ROCM_USE_AITER_MOE: "1"
|
||||
server_args: >-
|
||||
--enforce-eager
|
||||
--max-model-len 4096
|
||||
--tensor-parallel-size 4
|
||||
--enable-expert-parallel
|
||||
--attention-backend ROCM_AITER_MLA_SPARSE
|
||||
--moe-backend aiter
|
||||
--speculative-config '{"method":"mtp","num_speculative_tokens":3}'
|
||||
@@ -0,0 +1,15 @@
|
||||
model_name: "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"
|
||||
accuracy_threshold: 0.93
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
startup_max_wait_seconds: 1200
|
||||
env:
|
||||
VLLM_ROCM_USE_AITER: "1"
|
||||
server_args: >-
|
||||
--enforce-eager
|
||||
--max-model-len 4096
|
||||
--tensor-parallel-size 4
|
||||
--enable-expert-parallel
|
||||
--mamba-backend triton
|
||||
--moe-backend triton
|
||||
--speculative-config '{"method":"mtp","num_speculative_tokens":5}'
|
||||
@@ -0,0 +1,15 @@
|
||||
model_name: "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8"
|
||||
accuracy_threshold: 0.93
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
startup_max_wait_seconds: 1200
|
||||
env:
|
||||
VLLM_ROCM_USE_AITER: "1"
|
||||
server_args: >-
|
||||
--enforce-eager
|
||||
--max-model-len 4096
|
||||
--tensor-parallel-size 4
|
||||
--enable-expert-parallel
|
||||
--mamba-backend triton
|
||||
--moe-backend triton
|
||||
--speculative-config '{"method":"mtp","num_speculative_tokens":5}'
|
||||
@@ -0,0 +1 @@
|
||||
DeepSeek-R1-DP_MI355.yaml
|
||||
@@ -0,0 +1 @@
|
||||
DeepSeek-R1-TP_MI355.yaml
|
||||
@@ -0,0 +1 @@
|
||||
DeepSeek-V3.2-DP_MI355.yaml
|
||||
@@ -0,0 +1 @@
|
||||
DeepSeek-V3.2-TP_MI355.yaml
|
||||
@@ -0,0 +1,2 @@
|
||||
Nemotron-3-Super-120B-A12B-BF16_MI355.yaml
|
||||
Nemotron-3-Super-120B-A12B-FP8_MI355.yaml
|
||||
@@ -109,15 +109,17 @@ def test_gsm8k_correctness(config_filename):
|
||||
"Its W2A16 (uint2b2) scheme has no kernel outside CUDA."
|
||||
)
|
||||
|
||||
# TODO(akaratza): Enable DeepSeek-V3.2 and DeepSeek-R1 on ROCm platforms
|
||||
if current_platform.is_rocm() and (
|
||||
"deepseek-ai/DeepSeek-V3.2" in eval_config["model_name"]
|
||||
or "deepseek-ai/DeepSeek-R1" in eval_config["model_name"]
|
||||
):
|
||||
pytest.skip(
|
||||
"Skipping DeepSeek-V3.2 and DeepSeek-R1 on ROCm platforms "
|
||||
"due to agent pool disk space issues and pod evictions."
|
||||
)
|
||||
from vllm.platforms.rocm import get_cdna_version
|
||||
|
||||
if get_cdna_version() < 4:
|
||||
pytest.skip(
|
||||
"Skipping DeepSeek-V3.2 and DeepSeek-R1 on ROCm platforms below "
|
||||
"CDNA 4 due to agent pool disk space issues and pod evictions."
|
||||
)
|
||||
if current_platform.is_rocm() and (
|
||||
"Qwen3.5-35B-A3B-MXFP4-AITER-TP2" in config_filename.name
|
||||
):
|
||||
|
||||
Reference in New Issue
Block a user