[ROCm][CI] Cleaning and restructuring amd-ci legacy pipeline (#34839)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>
This commit is contained in:
Andreas Karatzas
2026-03-19 14:30:58 -05:00
committed by GitHub
parent 9279c59a0e
commit 040a505ff5
15 changed files with 2484 additions and 3552 deletions
@@ -1,3 +1,3 @@
# GFX942 model configurations for GPQA evaluation
# Tests different environment variable combinations
gpt-oss-20b-rocm-baseline.yaml
gpt-oss-20b-rocm-baseline.yaml
@@ -0,0 +1,12 @@
model_name: "deepseek-ai/DeepSeek-R1"
accuracy_threshold: 0.95
num_questions: 1319
num_fewshot: 5
startup_max_wait_seconds: 1200
server_args: >-
--enforce-eager
--max-model-len 4096
--data-parallel-size 8
--enable-expert-parallel
--attention-backend=TRITON_ATTN
--speculative-config '{"method":"mtp","num_speculative_tokens":3}'
@@ -0,0 +1,12 @@
model_name: "deepseek-ai/DeepSeek-R1"
accuracy_threshold: 0.95
num_questions: 1319
num_fewshot: 5
startup_max_wait_seconds: 1200
server_args: >-
--enforce-eager
--max-model-len 4096
--tensor-parallel-size 8
--enable-expert-parallel
--attention-backend=TRITON_ATTN
--speculative-config '{"method":"mtp","num_speculative_tokens":3}'
@@ -0,0 +1,12 @@
model_name: "deepseek-ai/DeepSeek-V3.2"
accuracy_threshold: 0.95
num_questions: 1319
num_fewshot: 5
startup_max_wait_seconds: 1200
server_args: >-
--enforce-eager
--max-model-len 4096
--data-parallel-size 8
--enable-expert-parallel
--attention-backend=TRITON_ATTN
--speculative-config '{"method":"mtp","num_speculative_tokens":3}'
@@ -0,0 +1,12 @@
model_name: "deepseek-ai/DeepSeek-V3.2"
accuracy_threshold: 0.95
num_questions: 1319
num_fewshot: 5
startup_max_wait_seconds: 1200
server_args: >-
--enforce-eager
--max-model-len 4096
--tensor-parallel-size 8
--enable-expert-parallel
--attention-backend=TRITON_ATTN
--speculative-config '{"method":"mtp","num_speculative_tokens":3}'
@@ -0,0 +1,4 @@
DeepSeek-R1-TP_MI325.yaml
DeepSeek-R1-DP_MI325.yaml
DeepSeek-V3.2-TP_MI325.yaml
DeepSeek-V3.2-DP_MI325.yaml
@@ -64,6 +64,16 @@ def test_gsm8k_correctness(config_filename):
"Marlin kernels are not supported."
)
# 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."
)
# Parse server arguments from config (use shlex to handle quoted strings)
server_args_str = eval_config.get("server_args", "")
server_args = shlex.split(server_args_str) if server_args_str else []