From 15bb6133a3a13e973631ef0c9f9953f1a4c357c9 Mon Sep 17 00:00:00 2001 From: tzhouam Date: Wed, 9 Sep 2026 15:35:24 +0800 Subject: [PATCH] edge results: init_profile variant YAMLs; recipe GPU memory figure Co-Authored-By: Claude Fable 5.1 --- .../eager_both/eager_both.yaml | 100 ++++++++++++++++++ .../eager_stage1/eager_stage1.yaml | 99 +++++++++++++++++ recipes/Qwen/Qwen3-TTS-Edge.md | 2 +- 3 files changed, 200 insertions(+), 1 deletion(-) create mode 100644 benchmarks/tts/edge_results/init_profile/20260909-112200/eager_both/eager_both.yaml create mode 100644 benchmarks/tts/edge_results/init_profile/20260909-112200/eager_stage1/eager_stage1.yaml diff --git a/benchmarks/tts/edge_results/init_profile/20260909-112200/eager_both/eager_both.yaml b/benchmarks/tts/edge_results/init_profile/20260909-112200/eager_both/eager_both.yaml new file mode 100644 index 000000000..775a93e9e --- /dev/null +++ b/benchmarks/tts/edge_results/init_profile/20260909-112200/eager_both/eager_both.yaml @@ -0,0 +1,100 @@ +async_chunk: true +connectors: + connector_of_shared_memory: + name: SharedMemoryConnector + extra: + codec_streaming: true + connector_get_sleep_s: 0.01 + connector_get_max_wait_first_chunk: 3000 + connector_get_max_wait: 300 + codec_chunk_frames: 25 + codec_left_context_frames: 72 + initial_codec_chunk_frames: 1 + decode_batch_max_size: 4 +stages: +- stage_id: 0 + max_num_seqs: 64 + gpu_memory_utilization: 0.3 + trust_remote_code: true + enable_prefix_caching: false + async_scheduling: true + max_num_batched_tokens: 32768 + max_model_len: 4096 + devices: '0' + output_connectors: + to_stage_1: connector_of_shared_memory + default_sampling_params: + temperature: 0.9 + top_k: 50 + max_tokens: 4096 + min_tokens: 2 + repetition_penalty: 1.05 + silence_ban_frames: 0 + subtalker_sampling_params: + do_sample: true + temperature: 0.9 + top_k: 50 + top_p: 1.0 + enforce_eager: true +- stage_id: 1 + max_num_seqs: 64 + gpu_memory_utilization: 0.3 + enforce_eager: true + trust_remote_code: true + enable_prefix_caching: false + async_scheduling: true + max_num_batched_tokens: 65536 + max_model_len: 65536 + devices: '0' + input_connectors: + from_stage_0: connector_of_shared_memory + default_sampling_params: + temperature: 0.0 + top_p: 1.0 + top_k: -1 + max_tokens: 65536 + repetition_penalty: 1.0 +platforms: + rocm: + stages: + - stage_id: 1 + enforce_eager: true + npu: + stages: + - stage_id: 0 + enforce_eager: false + compilation_config: + cudagraph_mode: PIECEWISE + - stage_id: 1 + compilation_config: + cudagraph_mode: PIECEWISE + cpu: + stages: + - stage_id: 0 + engine_args: + enforce_eager: true + async_scheduling: false + max_num_seqs: 2 + max_num_batched_tokens: 2048 + max_model_len: 2048 + gpu_memory_utilization: 0.05 + dtype: bfloat16 + kv_cache_memory_bytes: 4294967296 + runtime: + devices: cpu + env: + VLLM_CPU_OMP_THREADS_BIND: 0-11 + - stage_id: 1 + engine_args: + enforce_eager: true + async_scheduling: false + max_num_seqs: 2 + max_num_batched_tokens: 8192 + max_model_len: 8192 + gpu_memory_utilization: 0.05 + dtype: bfloat16 + kv_cache_memory_bytes: 2147483648 + runtime: + devices: cpu + env: + VLLM_CPU_OMP_THREADS_BIND: 12-15 diff --git a/benchmarks/tts/edge_results/init_profile/20260909-112200/eager_stage1/eager_stage1.yaml b/benchmarks/tts/edge_results/init_profile/20260909-112200/eager_stage1/eager_stage1.yaml new file mode 100644 index 000000000..43f317c91 --- /dev/null +++ b/benchmarks/tts/edge_results/init_profile/20260909-112200/eager_stage1/eager_stage1.yaml @@ -0,0 +1,99 @@ +async_chunk: true +connectors: + connector_of_shared_memory: + name: SharedMemoryConnector + extra: + codec_streaming: true + connector_get_sleep_s: 0.01 + connector_get_max_wait_first_chunk: 3000 + connector_get_max_wait: 300 + codec_chunk_frames: 25 + codec_left_context_frames: 72 + initial_codec_chunk_frames: 1 + decode_batch_max_size: 4 +stages: +- stage_id: 0 + max_num_seqs: 64 + gpu_memory_utilization: 0.3 + trust_remote_code: true + enable_prefix_caching: false + async_scheduling: true + max_num_batched_tokens: 32768 + max_model_len: 4096 + devices: '0' + output_connectors: + to_stage_1: connector_of_shared_memory + default_sampling_params: + temperature: 0.9 + top_k: 50 + max_tokens: 4096 + min_tokens: 2 + repetition_penalty: 1.05 + silence_ban_frames: 0 + subtalker_sampling_params: + do_sample: true + temperature: 0.9 + top_k: 50 + top_p: 1.0 +- stage_id: 1 + max_num_seqs: 64 + gpu_memory_utilization: 0.3 + enforce_eager: true + trust_remote_code: true + enable_prefix_caching: false + async_scheduling: true + max_num_batched_tokens: 65536 + max_model_len: 65536 + devices: '0' + input_connectors: + from_stage_0: connector_of_shared_memory + default_sampling_params: + temperature: 0.0 + top_p: 1.0 + top_k: -1 + max_tokens: 65536 + repetition_penalty: 1.0 +platforms: + rocm: + stages: + - stage_id: 1 + enforce_eager: true + npu: + stages: + - stage_id: 0 + enforce_eager: false + compilation_config: + cudagraph_mode: PIECEWISE + - stage_id: 1 + compilation_config: + cudagraph_mode: PIECEWISE + cpu: + stages: + - stage_id: 0 + engine_args: + enforce_eager: true + async_scheduling: false + max_num_seqs: 2 + max_num_batched_tokens: 2048 + max_model_len: 2048 + gpu_memory_utilization: 0.05 + dtype: bfloat16 + kv_cache_memory_bytes: 4294967296 + runtime: + devices: cpu + env: + VLLM_CPU_OMP_THREADS_BIND: 0-11 + - stage_id: 1 + engine_args: + enforce_eager: true + async_scheduling: false + max_num_seqs: 2 + max_num_batched_tokens: 8192 + max_model_len: 8192 + gpu_memory_utilization: 0.05 + dtype: bfloat16 + kv_cache_memory_bytes: 2147483648 + runtime: + devices: cpu + env: + VLLM_CPU_OMP_THREADS_BIND: 12-15 diff --git a/recipes/Qwen/Qwen3-TTS-Edge.md b/recipes/Qwen/Qwen3-TTS-Edge.md index ef51bfafd..62b84a620 100644 --- a/recipes/Qwen/Qwen3-TTS-Edge.md +++ b/recipes/Qwen/Qwen3-TTS-Edge.md @@ -3,7 +3,7 @@ This recipe covers the edge deploy profile, the CPU platform, hardware auto-adaptation, codec-token streaming with on-device decoding, and the profiling tools added on the `edge/omni-edge` branch. Numbers quoted here were -measured on one L20X (48 GB) and a 2x56-core Xeon 8480C host on 2026-09-09; the +measured on one NVIDIA L20X (143 GB) and a 2x56-core Xeon 8480C host on 2026-09-09; the JSON files are committed under `benchmarks/tts/edge_results/` (`python benchmarks/tts/edge_results/summarize.py` renders them).