README_backup.md

metadata

library_name: transformers
tags:
  - generated_from_trainer
datasets:
  - ./dataset/0510train_gpt+synthetic+whole_cat_changed_prompt_7000.jsonl
model-index:
  - name: >-
      mnt/nas2/eval-paradox/model/checkpoints/hub/0510qwen_7b_r2o_0425train_gpt+synthetic+whole_cat_changed_prompt
    results: []

See axolotl config

axolotl version: 0.6.0

base_model: /mnt/nas2/eval-paradox/model/checkpoints/hub/models--Qwen--Qwen2.5-7B-Instruct
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true

load_in_8bit: false
load_in_4bit: false
strict: false

output_dir: /mnt/nas2/eval-paradox/model/checkpoints/hub/0510qwen_7b_r2o_0425train_gpt+synthetic+whole_cat_changed_prompt
chat_template: qwen_25
datasets:
  - path: ./dataset/0510train_gpt+synthetic+whole_cat_changed_prompt_7000.jsonl
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content
    roles:
      system:
        - system
      user:
        - user
      assistant:
        - assistant

dataset_prepared_path: last_run_prepared
val_set_size: 0.005
output_dir: /mnt/nas2/eval-paradox/model/checkpoints/hub/0510qwen_7b_r2o_0425train_gpt+synthetic+whole_cat_changed_prompt
eval_sample_packing: False

sequence_len: 4096
sample_packing: False
pad_to_sequence_len: False

wandb_project: mergedbench
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:


plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_swiglu: true
liger_fused_linear_cross_entropy: false

gradient_accumulation_steps: 8
micro_batch_size: 4
eval_batch_size: 4
num_epochs: 3
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint: 
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 30
evals_per_epoch: 3
eval_max_new_tokens: 128
eval_table_size:
saves_per_epoch: 3
debug:
deepspeed: deepspeed_configs/zero3_bf16.json
offload_optimizer: true
offload_param: true
weight_decay: 0.01
fsdp:
fsdp_config:
special_tokens:

mnt/nas2/eval-paradox/model/checkpoints/hub/0510qwen_7b_r2o_0425train_gpt+synthetic+whole_cat_changed_prompt

This model was trained from scratch on the ./dataset/0510train_gpt+synthetic+whole_cat_changed_prompt_7000.jsonl dataset. It achieves the following results on the evaluation set:

• Loss: 0.0031

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 16
• optimizer: Use paged_adamw_8bit 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: 30
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss
1.8321	0.0186	1	1.7002
0.025	0.3349	18	0.0112
0.0189	0.6698	36	0.0119
0.0344	1.0186	54	0.0133
0.0074	1.3535	72	0.0080
0.0042	1.6884	90	0.0056
0.002	2.0372	108	0.0044
0.0042	2.3721	126	0.0030
0.0007	2.7070	144	0.0031

Framework versions

• Transformers 4.48.1
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.1