sllama-6

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>

<details><summary>See axolotl config</summary>

axolotl version: 0.13.0.dev0

base_model: /home/alex/Workspace/sllama/out_5/checkpoint-1722000
trust_remote_code: true
resize_token_embeddings_to_32x: true
plugins:
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_layer_norm: true

unfrozen_parameters:
  - ^(?![\s\S]*embed_tokens)[\s\S]+$

datasets:
  - path: lima.jsonl
    type: chat_template

dataloader_num_workers: 0
group_by_length: false
dataset_prepared_path: data_prep
output_dir: ./out_6_lima
dataloader_pin_memory: true
shuffle_merged_datasets: true

sequence_len: 2048
sample_packing: true
eval_sample_packing: true
pad_to_sequence_len: true

use_tensorboard: true
use_wandb: true
wandb_project: sllama

gradient_accumulation_steps: 1

micro_batch_size: 1
num_epochs: 4
#max_steps: 100000
save_steps: 100
save_total_limit: 2
save_only_model: true
optimizer: sgd
optim_args:
  momentum: 0.98
lr_scheduler: cosine
learning_rate: 0.1
#embedding_lr: 5e-7
cosine_constant_lr_ratio: 0.1
max_grad_norm: 1.0

bf16: auto
fp8: true

gradient_checkpointing: false
gradient_checkpointing_kwargs:
  use_reentrant: false

logging_steps: 10
torch_compile: true
torch_compile_backend: inductor
torch_compile_mode: default
flash_attention: true

warmup_ratio: 0.05
weight_decay: 0.01

</details><br>

out_6_lima

This model was trained from scratch on the lima.jsonl dataset.

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: 0.1
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• optimizer: Use OptimizerNames.SGD and the args are: momentum=0.98
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 64
• training_steps: 1288

Training results

Framework versions

• Transformers 4.56.1
• Pytorch 2.7.1+cu128
• Datasets 4.0.0
• Tokenizers 0.22.1