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

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
cuda_graph_config:
  enable_padding: true
  batch_sizes:
  - 1
  - 2
  - 4
  - 8
enable_attention_dp: false
enable_iter_perf_stats: true
enable_iter_req_stats: false
iter_perf_stats_interval: 40
enable_lm_head_tp_in_adp: false
kv_cache_config:
  avg_seq_len: 102150
  dtype: fp8
  enable_block_reuse: false
  event_buffer_max_size: 0
  free_gpu_memory_fraction: 0.85
  host_cache_size: 137438953472
  iteration_stats_interval: 40
  tokens_per_block: 64
  use_kv_cache_manager_v2: true
max_seq_len: 262148
moe_config:
  backend: TRTLLM
  use_low_precision_moe_combine: true
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: GUARANTEED_NO_EVICT
speculative_config:
  decoding_type: MTP
  max_draft_len: 6
stream_interval: 20
trust_remote_code: true
max_batch_size: 8
max_num_tokens: 56
moe_expert_parallel_size: 2
tensor_parallel_size: 2