seed: 0
output_dir: './output' # path to save checkpoint/strategy
load_checkpoint: ''
src_strategy_path_or_dir: ''
auto_trans_ckpt: False  # If true, auto transform load_checkpoint to load in distributed model
only_save_strategy: False
resume_training: False
use_parallel: False
run_mode: 'predict'

# trainer config
trainer:
  type: CausalLanguageModelingTrainer
  model_name: 'qwen2_7b'

# runner config
runner_config:
  epochs: 5
  batch_size: 1
  sink_mode: True
  sink_size: 2
runner_wrapper:
  type: MFTrainOneStepCell
  scale_sense:
    type: DynamicLossScaleUpdateCell
    loss_scale_value: 65536
    scale_factor: 2
    scale_window: 1000
  use_clip_grad: True

# default parallel of device num = 8 for Atlas 800T A2
parallel_config:
  data_parallel: 1
  model_parallel: 1
  pipeline_stage: 1
  micro_batch_num: 1
  vocab_emb_dp: True
  gradient_aggregation_group: 4
# when model parallel is greater than 1, we can set micro_batch_interleave_num=2, that may accelerate the train process.
micro_batch_interleave_num: 1

model:
  model_config:
    type: LlamaConfig
    batch_size: 1
    seq_length: 8192
    hidden_size: 4096
    num_layers: 32
    num_heads: 32
    vocab_size: 151936
    intermediate_size: 11008
    qkv_has_bias: True
    rms_norm_eps: 1.0e-6
    theta: 1000000.0
    emb_dropout_prob: 0.0
    eos_token_id: [151645, 151643]
    pad_token_id: 151643
    bos_token_id: 151643
    compute_dtype: "bfloat16"
    layernorm_compute_type: "float32"
    softmax_compute_type: "float32"
    rotary_dtype: "bfloat16"
    param_init_type: "float16"
    use_past: True
    use_flash_attention: True
    block_size: 32
    num_blocks: 1024
    offset: 0
    checkpoint_name_or_path: ""
    repetition_penalty: 1.05
    max_decode_length: 512
    temperature: 0.7
    top_k: 20
    top_p: 0.8
    do_sample: True
    is_dynamic: True
    qkv_concat: True
    auto_map:
      AutoTokenizer: [qwen1_5_tokenizer.Qwen2Tokenizer, null]
  arch:
    type: LlamaForCausalLM

processor:
  return_tensors: ms
  tokenizer:
    model_max_length: 32768
    vocab_file: "/path/vocab.json"
    merges_file: "/path/merges.txt"
    unk_token: "<|endoftext|>"
    pad_token: "<|endoftext|>"
    eos_token: "<|im_end|>"
    chat_template: "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"
    type: Qwen2Tokenizer
  type: Qwen2Processor

# mindspore context init config
context:
  mode: 0 #0--Graph Mode; 1--Pynative Mode
  device_target: "Ascend"
  enable_graph_kernel: False
  ascend_config:
    precision_mode: "must_keep_origin_dtype"
  max_call_depth: 10000
  max_device_memory: "59GB"
  save_graphs: False
  save_graphs_path: "./graph"
  device_id: 0

# parallel context config
parallel:
  parallel_mode: 1 # 0-data parallel, 1-semi-auto parallel, 2-auto parallel, 3-hybrid parallel
  gradients_mean: False
  enable_alltoall: False
  full_batch: True
  search_mode: "sharding_propagation"
  enable_parallel_optimizer: False
  strategy_ckpt_config:
    save_file: "./ckpt_strategy.ckpt"
    only_trainable_params: False
  parallel_optimizer_config:
    gradient_accumulation_shard: False
    parallel_optimizer_threshold: 64