See axolotl config

axolotl version: 0.8.0

base_model: Qwen/Qwen2.5-7B-Instruct
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: false

load_in_8bit: false
load_in_4bit: false
strict: false

output_dir: ./outputs/out
chat_template: qwen_25
datasets:
  - path: mb_base.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: ./outputs/out
eval_sample_packing: False

sequence_len: 8192
sample_packing: False
pad_to_sequence_len: False

wandb_project: mergedbench
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
hub_model_id: amphora/merged-bench-train-base

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

gradient_accumulation_steps: 2
micro_batch_size: 8
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: auto
fp16:
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: 1
debug:
deepspeed: deepspeed_configs/zero3_bf16.json
weight_decay: 0.01
fsdp:
fsdp_config:
special_tokens:

merged-bench-train-base

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the mb_base.jsonl dataset. It achieves the following results on the evaluation set:

• Loss: 0.3173

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: 8
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 16
• optimizer: Use OptimizerNames.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
1.0382	0.0059	1	1.0319
0.3455	0.3373	57	0.3270
0.3169	0.6746	114	0.3173
0.2116	1.0118	171	0.3009
0.2064	1.3491	228	0.3020
0.1871	1.6864	285	0.2955
0.1069	2.0237	342	0.2880
0.1014	2.3609	399	0.3192
0.0955	2.6982	456	0.3173

Framework versions

• Transformers 4.51.0
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1