seed: 0
output_dir: './output' # path to save checkpoint/strategy
run_mode: 'predict'
use_parallel: True

load_checkpoint: "/home/workspace/mindspore_dataset/weight/DeepSeek-R1-bf16"
load_ckpt_format: "safetensors"
auto_trans_ckpt: True # If true, auto transform load_checkpoint to load in distributed model

# trainer config
trainer:
  type: CausalLanguageModelingTrainer
  model_name: 'DeepSeekR1'

# default parallel of device num = 32 for Atlas 800T A2
parallel_config:
  model_parallel: 2
  pipeline_stage: 1
  expert_parallel: 1
  vocab_emb_dp: False

# mindspore context init config
context:
  mode: 1 # 0--Graph Mode; 1--Pynative Mode
  max_device_memory: "13GB"
  device_id: 0

kernel_launch_group:
  thread_num: 2
  kernel_group_num: 8

# parallel context config
parallel:
  parallel_mode: "STAND_ALONE" # use 'STAND_ALONE' mode for inference with parallelism in frontend
  full_batch: False
  strategy_ckpt_save_file: "./ckpt_strategy.ckpt"

# model config
model:
  model_config:
    type: DeepseekV3Config
    auto_register: deepseek3_config.DeepseekV3Config
    batch_size: 1 # add for incre predict
    seq_length: 4096
    hidden_size: 7168
    num_layers: 1
    num_heads: 128
    max_position_embeddings: 163840
    intermediate_size: 18432
    kv_lora_rank:  512
    q_lora_rank: 1536
    qk_rope_head_dim: 64
    v_head_dim: 128
    qk_nope_head_dim: 128
    vocab_size: 129280
    multiple_of: 256
    rms_norm_eps: 1.0e-6
    bos_token_id: 0
    eos_token_id: 1
    pad_token_id: 1
    ignore_token_id: -100
    compute_dtype: "bfloat16"
    layernorm_compute_type: "bfloat16"
    softmax_compute_type: "bfloat16"
    rotary_dtype: "bfloat16"
    router_dense_type: "bfloat16"
    param_init_type: "bfloat16"
    scaling_factor:
      beta_fast: 32.0
      beta_slow: 1.0
      factor: 40.0
      mscale: 1.0
      mscale_all_dim: 1.0
      original_max_position_embeddings: 4096
    use_past: True
    extend_method: "YARN"
    use_flash_attention: True
    block_size: 16
    num_blocks: 512
    offset: 0
    checkpoint_name_or_path: ""
    repetition_penalty: 1
    max_decode_length: 1024
    top_k: 1
    top_p: 1
    theta: 10000.0
    do_sample: False
    is_dynamic: True
    qkv_concat: False
    ffn_concat: True
    auto_map:
      AutoConfig: deepseek3_config.DeepseekV3Config
      AutoModel: deepseek3.DeepseekV3ForCausalLM
  arch:
    type: DeepseekV3ForCausalLM
    auto_register: deepseek3.DeepseekV3ForCausalLM

moe_config:
  expert_num: 256
  num_experts_chosen: 8
  routing_policy: "TopkRouterV2"
  shared_expert_num: 1
  routed_scaling_factor: 2.5
  first_k_dense_replace: 3
  moe_intermediate_size: 2048
  topk_group: 4
  n_group: 8

processor:
  return_tensors: ms
  tokenizer:
    unk_token: '<unk>'
    bos_token: '<｜begin▁of▁sentence｜>'
    eos_token: '<｜end▁of▁sentence｜>'
    pad_token: '<｜end▁of▁sentence｜>'
    type: LlamaTokenizerFast
    vocab_file: '/home/workspace/mindspore_dataset/weight/DeepSeek-R1-bf16/tokenizer.json'
    tokenizer_file: '/home/workspace/mindspore_dataset/weight/DeepSeek-R1-bf16/tokenizer.json'
    chat_template: "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='', is_first_sp=true) %}{%- for message in messages %}{%- if message['role'] == 'system' %}{%- if ns.is_first_sp %}{% set ns.system_prompt = ns.system_prompt + message['content'] %}{% set ns.is_first_sp = false %}{%- else %}{% set ns.system_prompt = ns.system_prompt + '\n\n' + message['content'] %}{%- endif %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{{'<｜Assistant｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜>'}}{% endif %}"
  type: LlamaProcessor