[Config] Unify indexer cache dtype under attention_config.indexer_kv_dtype (#52550)

Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Yongye Zhu
2026-08-17 10:20:04 +08:00
committed by GitHub
co-authored by Claude Opus 5
parent 0ad04cff1b
commit 967e104fad
6 changed files with 65 additions and 24 deletions
@@ -15,7 +15,7 @@ server_args: >-
--block-size 256
--gpu-memory-utilization 0.5
--kv-cache-dtype fp8
--attention_config.use_fp4_indexer_cache=True
--attention_config.indexer_kv_dtype=mxfp4
--max-num-batched-tokens 16384
--max-num-seqs 128
--speculative-config '{"method":"dspark",
@@ -2,4 +2,4 @@ model_name: "deepseek-ai/DeepSeek-V4-Flash"
accuracy_threshold: 0.95
num_questions: 1319
num_fewshot: 5
server_args: "--trust-remote-code --kv-cache-dtype fp8 --block-size 256 --enable-expert-parallel --tensor-parallel-size 2 --attention_config.use_fp4_indexer_cache=True --moe-backend deep_gemm_mega_moe --tokenizer-mode deepseek_v4 --tool-call-parser deepseek_v4 --enable-auto-tool-choice --reasoning-parser deepseek_v4 --speculative_config.method=mtp --speculative_config.num_speculative_tokens=2"
server_args: "--trust-remote-code --kv-cache-dtype fp8 --block-size 256 --enable-expert-parallel --tensor-parallel-size 2 --attention_config.indexer_kv_dtype=mxfp4 --moe-backend deep_gemm_mega_moe --tokenizer-mode deepseek_v4 --tool-call-parser deepseek_v4 --enable-auto-tool-choice --reasoning-parser deepseek_v4 --speculative_config.method=mtp --speculative_config.num_speculative_tokens=2"