diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml index 70a79a273..6835d9abc 100644 --- a/.github/configs/amd-master.yaml +++ b/.github/configs/amd-master.yaml @@ -2257,15 +2257,8 @@ dsv4-fp4-mi355x-vllm-mtp: search-space: - { tp: 8, conc-start: 4, conc-end: 512, spec-decoding: mtp } -# Day-0 single-sequence marker for DeepSeek-V4 on ATOM (ROCm/ATOM#650). -# PR1 of the ATOM DSv4 series still uses torch sparse-attention fallbacks -# that OOM once warmup/prefill batches multiple requests; keep CONC=1 until -# the AITER sparse-attention kernel / multi-request path lands upstream. -# --enforce-eager and ATOM_USE_TRITON_MOE=1 are required on gfx950. Image is -# the standard atom0.1.2.post MI355X base (matching qwen3.5-fp8-mi355x-atom); -# the DSv4 PR is overlaid at runtime by dsv4_fp4_mi355x_atom.sh at a pinned SHA. dsv4-fp4-mi355x-atom: - image: rocm/atom:rocm7.2.4_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.3 + image: rocm/atom:rocm7.2.4_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.4_20260612 model: deepseek-ai/DeepSeek-V4-Pro model-prefix: dsv4 runner: mi355x @@ -2274,16 +2267,29 @@ dsv4-fp4-mi355x-atom: multinode: false scenarios: fixed-seq-len: - - isl: 1024 - osl: 1024 - search-space: - - { tp: 8, ep: 1, conc-start: 1, conc-end: 64 } - - { tp: 8, ep: 1, dp-attn: true, conc-start: 64, conc-end: 1024 } - - isl: 8192 - osl: 1024 - search-space: - - { tp: 8, ep: 1, conc-start: 1, conc-end: 64 } - - { tp: 8, ep: 1, dp-attn: true, conc-start: 64, conc-end: 512 } + #- isl: 1024 + # osl: 1024 + # search-space: + # # conc4-64, TP8 + # # conc128-512, DPA + # # conc1024-8192, DPA TBO + # - { tp: 8, ep: 1, conc-start: 1, conc-end: 64 } + # - { tp: 8, ep: 1, dp-attn: true, conc-start: 64, conc-end: 8192 } + #- isl: 8192 + # osl: 1024 + # search-space: + # # conc4-64, TP8 + # # conc128, DPA + # # conc256-4096, DPA TBO + # - { tp: 8, ep: 1, conc-start: 4, conc-end: 64 } + # - { tp: 8, ep: 1, dp-attn: true, conc-start: 128, conc-end: 4096 } + - isl: 8192 + osl: 1024 + search-space: + # conc4-64, TP8 + # conc128, DPA + # conc256-4096, DPA TBO + - { tp: 8, ep: 1, dp-attn: true, conc-start: 1024, conc-end: 1024 } dsv4-fp4-mi355x-atom-mtp: image: rocm/atom:rocm7.2.4_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.3 diff --git a/benchmarks/single_node/fixed_seq_len/dsv4_fp4_mi355x_atom.sh b/benchmarks/single_node/fixed_seq_len/dsv4_fp4_mi355x_atom.sh index e485dc9a6..898dac45e 100644 --- a/benchmarks/single_node/fixed_seq_len/dsv4_fp4_mi355x_atom.sh +++ b/benchmarks/single_node/fixed_seq_len/dsv4_fp4_mi355x_atom.sh @@ -22,14 +22,32 @@ echo "TP: $TP, CONC: $CONC, ISL: $ISL, OSL: $OSL, EP_SIZE: $EP_SIZE, DP_ATTENTIO SERVER_LOG=/workspace/server.log PARALLEL_ARGS=(-tp "$TP") #TP +CUDAGRAPH_SIZES='[1,2,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60,64,68,72,76,80,84,88,92,96,100,104,108,112,116,120,124,128,132,136,140,144,148,152,156,160,164,168,172,176,180,184,188,192,196,200,204,208,212,216,220,224,228,232,236,240,244,248,252,256]' if [ "$DP_ATTENTION" = "true" ]; then if [ "$EP_SIZE" -gt 1 ]; then #DP+EP PARALLEL_ARGS=(-tp "$TP" --enable-expert-parallel --enable-dp-attention ) - else #DP+TP - PARALLEL_ARGS=(-tp "$TP" --enable-dp-attention ) + else #DPA+TP + #DPA+TP+TBO + if [ "$ISL" -eq 1024 ] && [ "$OSL" -eq 1024 ] && [ "$CONC" -ge 1024 ]; then + PARALLEL_ARGS=(-tp "$TP" --enable-dp-attention --enable-tbo) + CUDAGRAPH_SIZES='[1,2,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60,64,68,72,76,80,84,88,92,96,100,104,108,112,116,120,124,128,132,136,140,144,148,152,156,160,164,168,172,176,180,184,188,192,196,200,204,208,212,216,220,224,228,232,236,240,244,248,252,256,512,1024]' + elif [ "$ISL" -eq 8192 ] && [ "$OSL" -eq 1024 ] && [ "$CONC" -ge 256 ]; then + PARALLEL_ARGS=(-tp "$TP" --enable-dp-attention --enable-tbo) + else + PARALLEL_ARGS=(-tp "$TP" --enable-dp-attention ) + fi fi fi +BENCHMARK_MAX_MODEL_LEN="$MAX_MODEL_LEN" + +if [ "${EVAL_ONLY}" = "true" ]; then + EVAL_MAX_MODEL_LEN=$(compute_eval_context_length "$MODEL" "$BENCHMARK_MAX_MODEL_LEN") + export EVAL_MAX_MODEL_LEN + SERVE_MAX_MODEL_LEN="$EVAL_MAX_MODEL_LEN" +else + SERVE_MAX_MODEL_LEN="$BENCHMARK_MAX_MODEL_LEN" +fi # Start GPU monitoring (power, temperature, clocks every second) start_gpu_monitor @@ -37,8 +55,6 @@ set -x export ATOM_DISABLE_MMAP=true export AITER_BF16_FP8_MOE_BOUND=0 export ATOM_MOE_GU_ITLV=1 -# TODO: add --no-enable_chunked_prefill, when dsv4 prefix caching is supported -#https://github.com/ROCm/ATOM/commit/7df93a181da4d3c3250c2441c7d5e2745a03d0cd#diff-61b1ba0b8b74523530d2d5cdc739d4f3a23a43bedf69015a5235844d46e9373bL1127 python3 -m atom.entrypoints.openai_server \ --model $MODEL \ --server-port $PORT \ @@ -46,7 +62,11 @@ python3 -m atom.entrypoints.openai_server \ --kv_cache_dtype fp8 \ --trust-remote-code \ --gpu-memory-utilization 0.85 \ + --no-enable_prefix_caching \ + --cudagraph-capture-sizes "${CUDAGRAPH_SIZES}" \ + --max-num-seqs ${CONC} \ > $SERVER_LOG 2>&1 & + #--max-model-len "$SERVE_MAX_MODEL_LEN" \ SERVER_PID=$! diff --git a/perf-changelog.yaml b/perf-changelog.yaml index c0642188b..f236a6d60 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -3600,3 +3600,11 @@ - "MI355x DSR1-FP4: Include TP4 configurations for 8k1k" - "Expand the TP sweep (included TP=4) for 8k/1k configuration for conc=4 to 64" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1692 + +- config-keys: + - dsv4-fp4-mi355x-atom + description: + - "Update image to rocm/atom:rocm7.2.4_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.4_20260612" + - "Enable --enable-tbo (Two Batch Overlap) on top of DPA+TP8 at high concurrency: ISL=1024/OSL=1024 at CONC>=1024, ISL=8192/OSL=1024 at CONC>=256" + - "Update ISL=8192 search-space: TP8-only from conc=4-64, DPA from conc=128-1024 (previously conc=1-64 and DPA conc=64-512)" + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1717