# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

attention_dp_config:
  kv_cache_routing_conversation_affinity: true
cache_transceiver_config:
  backend: NIXL
  kv_transfer_sender_future_timeout_ms: 20
  kv_transfer_timeout_ms: 600000
  transceiver_runtime: PYTHON
  kv_cache_bounce_size_mb: 2048
disable_overlap_scheduler: false
enable_attention_dp: true
enable_chunked_prefill: true
enable_iter_perf_stats: true
enable_iter_req_stats: false
iter_perf_stats_interval: 40
kv_cache_config:
  block_reuse_config:
    max_num_turns: 3
    policy: per_conversation
  dtype: fp8
  enable_block_reuse: true
  event_buffer_max_size: 0
  free_gpu_memory_fraction: 0.8
  host_cache_size: 137438953472
  iteration_stats_interval: 40
  mamba_state_config:
    additional_snapshot_offsets_from_end:
    - 2
    periodic_snapshot_interval: 0
  pool_ratio:
  - 0.8
  - 0.2
  tokens_per_block: 64
  use_kv_cache_manager_v2: true
max_batch_size: 64
max_num_tokens: 8192
max_seq_len: 262144
moe_config:
  backend: CUTEDSL
nvfp4_gemm_config:
  allowed_backends:
  - cutlass
  - cublaslt
  - cutedsl
  - cuda_core
pipeline_parallel_size: 1
print_iter_log: true
return_perf_metrics: false
scheduler_config:
  capacity_scheduler_policy: MAX_UTILIZATION
torch_compile_config:
  capture_num_tokens:
  - 1
  - 2
  - 4
  - 8
  - 16
  - 32
  - 64
  - 128
  - 256
  - 512
  - 1024
  - 2048
  - 4096
  - 8192
  enable_fullgraph: true
  enable_piecewise_cuda_graph: true
trust_remote_code: true
moe_expert_parallel_size: 1
tensor_parallel_size: 1