Rae-15B-PT-Ckpts

<!-- 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: 0.9.0

## model
base_model: SillyTilly/ServiceNow-AI-Apriel-Nemotron-15b-Thinker-Chatml
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

## qlora COPE
load_in_8bit: false
load_in_4bit: false
strict: false

## data 
datasets:
datasets:
  - path: Delta-Vector/Ursa-Erebus-16K
    type: completion
    field: body
  - path: Delta-Vector/Ursa-Books-Light-Novels-V1
    type: completion
    field: text
  - path: NewEden/Orion-LIT
    type: completion
    field: text
  - path: Delta-Vector/Ursa-Asstr-V2-18k
    type: completion
    field: content
  - path: Delta-Vector/Ursa-Books-V2
    type: completion
    field: text
  - path: Delta-Vector/Ursa-Scribblehub-7k
    type: completion
    field: text
#  - path: Delta-Vector/Ursa-SCP-wiki-1.9K
#    type: completion
#    field: text
  - path: Delta-Vector/Ursa-Orion-EA-Comp-Filtered
    type: completion
    field: Text
  - path: Delta-Vector/Ursa-HoneyFeed
    type: completion
    field: text
  - path: Delta-Vector/Ursa-Falling-through-the-world
    type: completion
    field: content
shuffle_merged_datasets: true
dataset_prepared_path: dataset_preparedss
val_set_size: 0.0
output_dir: ./Rae-15B-Pretrain

## Liger + CCE
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

## CTX settings
sequence_len: 16384
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

## max grad norm
max_grad_norm: 0.001


## WandB
wandb_project: Rae
wandb_entity:
wandb_watch:
wandb_name: Pretrain-15B
wandb_log_model:

## evals
#evals_per_epoch: 0
#eval_table_size:
#eval_max_new_tokens: 128

## hoe params
gradient_accumulation_steps: 2
micro_batch_size: 2
num_epochs: 1
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 1e-5

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

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
s2_attention:

warmup_steps: 50
saves_per_epoch: 2
debug:
deepspeed: ./deepspeed_configs/zero3_bf16.json
weight_decay: 0.0001
fsdp:
fsdp_config:
special_tokens:
   pad_token: <pad>

</details><br>

Rae-15B-Pretrain

This model is a fine-tuned version of SillyTilly/ServiceNow-AI-Apriel-Nemotron-15b-Thinker-Chatml on the Delta-Vector/Ursa-Erebus-16K, the Delta-Vector/Ursa-Books-Light-Novels-V1, the NewEden/Orion-LIT, the Delta-Vector/Ursa-Asstr-V2-18k, the Delta-Vector/Ursa-Books-V2, the Delta-Vector/Ursa-Scribblehub-7k, the Delta-Vector/Ursa-Orion-EA-Comp-Filtered, the Delta-Vector/Ursa-HoneyFeed and the Delta-Vector/Ursa-Falling-through-the-world datasets.

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: 6
• gradient_accumulation_steps: 2
• total_train_batch_size: 24
• total_eval_batch_size: 12
• 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: 50
• num_epochs: 1.0

Training results

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1