base_model: Salesforce/xgen-small-4B-instruct-r

load_in_8bit: false
load_in_4bit: false
strict: false

chat_template: chatml
datasets:
  - path: hardlyworking/HardlyRPv2
    type: chat_template
    split: train
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value
val_set_size: 0.1
output_dir: ./outputs/out
dataset_prepared_path: last_run_prepared
shuffle_merged_datasets: true

hub_model_id: hardlyworking/HoldMy4B
hub_strategy: "all_checkpoints"
push_dataset_to_hub:
hf_use_auth_token: true

plugins:
  - axolotl.integrations.liger.LigerPlugin
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
liger_rope: true
liger_rms_norm: true
liger_layer_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: false
cut_cross_entropy: true

sequence_len: 32768
sample_packing: true
eval_sample_packing: true
pad_to_sequence_len: true

wandb_project: Xgen4B
wandb_entity:
wandb_watch:
wandb_name: Xgen4B
wandb_log_model:

evals_per_epoch: 8
eval_table_size:
eval_max_new_tokens: 128

gradient_accumulation_steps: 2
micro_batch_size: 2
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 1e-5

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: offload
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
s2_attention:

deepspeed:

warmup_ratio: 0.05
saves_per_epoch: 1
debug:
weight_decay: 0.01
fsdp:
fsdp_config:
special_tokens:
   pad_token:

HoldMy4B

This model is a fine-tuned version of Salesforce/xgen-small-4B-instruct-r on the hardlyworking/HardlyRPv2 dataset. It achieves the following results on the evaluation set:

Loss: 2.1637

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: 1e-05
train_batch_size: 2
eval_batch_size: 2
seed: 42
distributed_type: multi-GPU
num_devices: 2
gradient_accumulation_steps: 2
total_train_batch_size: 8
total_eval_batch_size: 4
optimizer: Use OptimizerNames.ADAMW_BNB 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: 24
training_steps: 480

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0	0	2.6420
2.0119	0.125	30	2.2105
1.8963	0.25	60	2.1865
1.8623	0.375	90	2.1787
1.8528	0.5	120	2.1746
1.8784	0.625	150	2.1706
1.9961	0.75	180	2.1686
1.8748	0.875	210	2.1672
2.0385	1.0	240	2.1657
1.9327	1.125	270	2.1646
1.8509	1.25	300	2.1645
1.8279	1.375	330	2.1640
1.8271	1.5	360	2.1638
1.8589	1.625	390	2.1637
1.9824	1.75	420	2.1637
1.8668	1.875	450	2.1637
2.0332	2.0	480	2.1637

Framework versions

Transformers 4.52.3
Pytorch 2.6.0+cu124
Datasets 3.6.0
Tokenizers 0.21.1