[ROCm][CI] skip the ROCm MRV1 default where MRV1 cannot serve the config (#58535)

Signed-off-by: Stefan Koncarevic <Stefan.Koncarevic@amd.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>
This commit is contained in:
stefankoncarevic
2026-09-24 15:36:36 -05:00
committed by GitHub
co-authored by Andreas Karatzas
parent 0908116dd9
commit e33de821c0
5 changed files with 91 additions and 2 deletions
@@ -0,0 +1,28 @@
model_name: "deepseek-ai/DeepSeek-V4-Flash-DSpark"
accuracy_threshold: 0.92
tolerance: 0.08
min_acceptance_length: 1.2
num_questions: 1319
num_fewshot: 5
startup_max_wait_seconds: 1800
env:
VLLM_ROCM_USE_AITER: "1"
server_args: >-
--tokenizer-mode deepseek_v4
--trust-remote-code
--max-model-len 8192
--tensor-parallel-size 4
--enable-expert-parallel
--all2all-backend allgather_reducescatter
--block-size 256
--gpu-memory-utilization 0.5
--kv-cache-dtype fp8
--attention_config.indexer_kv_dtype=fp8
--max-num-batched-tokens 16384
--max-num-seqs 128
--enforce-eager
--speculative-config '{"method":"dspark",
"model":"deepseek-ai/DeepSeek-V4-Flash-DSpark",
"num_speculative_tokens":7,
"draft_sample_method":"probabilistic",
"enable_adaptive_verification":false}'
@@ -1,2 +1,3 @@
Kimi-K3-pruned75-DSpark-AITER-TP4.yaml
DeepSeek-V4-Flash-DSpark-AITER-TEP4.yaml
DeepSeek-V4-Flash-DSpark-FP8-TP4-ROCm.yaml