# Automatically upload checkpoint and final model to HF
# hub_model_id: username/custom_model_name
# 是否以 8-bit 精度加载模型
load_in_8bit: false
# 是否以 4-bit 精度加载模型（与QLoRA绑定, 强制使用）
load_in_4bit: false
# 是否严格匹配模型结构，关闭表示可加载少部分差异结构（如以适配 adapter）
# strict: false
base_model: /workspace/xuyiwei/merged_agent_model_0908
# 数据集设置
chat_template: qwen3
datasets:
  - path: /workspace/jiangqiyi/file_manage_agent_train/data/retrieval_agent_train_data.json # - 表示列表（list）中的一项, 即可以同时使用多个数据集
    type: chat_template # chat_template(自定义格式) alpaca
    roles_to_train: ["assistant"]
    field_messages: messages # 标识的字段
    message_property_mappings:  # message_property_mappings={'role':'role', 'content':'content'})
      role: role
      content: content
dataset_prepared_path:
val_set_size: 0.05
output_dir: models/0910
sequence_len: 8192 # 模型所能处理的最大上下文长度(默认2048)
pad_to_sequence_len: true
# context_parallel_size: 2 # 长序列拆分至多个GPU(强制要求 mirco_batch_size: 1)
sample_packing: false # 在训练时将多个样本拼接（packing）成一个长序列(sequence_len)输入到模型中，以提高训练效率。
eval_sample_packing: false # 评估时拼接多个样本
# 训练超参数
adapter: lora  # lora qlora
lora_model_dir:
lora_r: 16 # lora_r默认首选 16，平衡精度与显存
lora_alpha: 64 # 缩放系数，用于控制 LoRA 的影响力, 一般设为 2*r 或 4*r
lora_dropout: 0.05
lora_target_linear: true
micro_batch_size: 1 # 微批次大小 94G的H100可以设为4(Token为1w)
gradient_accumulation_steps: 16 # 梯度累积: 将多个微批次的梯度(micro_batch_size)累积起来，然后更新模型权重 有效 Batch 常取 16: 小于 8 训练会抖，大于 32 只会更耗时、收益有限
auto_find_batch_size: false # 允许Axolotl不断调整batch_size  ⚠️Zero-3不适用
num_epochs: 1
optimizer: adamw_torch_fused
lr_scheduler: cosine
learning_rate: 2e-5
# bf16: auto + tf32: true，可获得更好的稳定性和性能。
bf16: auto
tf32: true
# early_stopping_patience:
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
# auto_resume_from_checkpoints: true #自动从output_dir寻找最新checkpoint断点恢复
logging_steps: 1
flash_attention: true
# xformers_attention: true
warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
weight_decay: 0.0
# deepspeed: /workspace/deepspeed_configs/zero2.json
# fsdp:
#   - full_shard
#   - auto_wrap
# fsdp_config:
#   fsdp_limit_all_gathers: true
#   fsdp_sync_module_states: true
#   fsdp_offload_params: true
#   fsdp_use_orig_params: false
#   fsdp_cpu_ram_efficient_loading: true
#   fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
#   fsdp_transformer_layer_cls_to_wrap: Qwen3DecoderLayer
#   fsdp_state_dict_type: FULL_STATE_DICT
#   fsdp_sharding_strategy: FULL_SHARD
# special_tokens:
# wandb_project:
# wandb_entity:
# wandb_watch:
# wandb_name:
# wandb_log_model:

models/0910

This model was trained from scratch on the /workspace/jiangqiyi/file_manage_agent_train/data/retrieval_agent_train_data.json dataset. It achieves the following results on the evaluation set:

Loss: 0.0452
Memory/max Mem Active(gib): 23.87
Memory/max Mem Allocated(gib): 23.87
Memory/device Mem Reserved(gib): 26.56

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 1
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 16
total_train_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 10
training_steps: 279

Training results

Training Loss	Epoch	Step	Validation Loss	Mem Active(gib)	Mem Allocated(gib)	Mem Reserved(gib)
No log	0	0	0.0491	19.7	19.7	19.8
0.0547	0.2515	70	0.0482	23.87	23.87	26.56
0.0423	0.5030	140	0.0461	23.87	23.87	26.56
0.0449	0.7545	210	0.0452	23.87	23.87	26.56

Framework versions

PEFT 0.17.0
Transformers 4.55.2
Pytorch 2.6.0+cu124
Datasets 4.0.0
Tokenizers 0.21.4

cjkasbdkjnlakb/agent-0910

models/0910

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

cjkasbdkjnlakb/
agent-0910