Aura-4B

Aura-4B

Introduction

Aura-4B is a state of the art dedicated roleplaying model designed to fulfill your every desire.

This finetune has seen several hundreds of millions of tokens of completion, instruction and roleplaying data. A Kahneman-Tversky Optimization was applied to give this model a unique output style.

Developed by Aura Industries, with contributions from Anthracite Org

Model Details

• Model Name: Aura-4B
• Base Model: IntervitensInc/Llama-3.1-Minitron-4B-Width-Base-chatml
• Model Type: Chat Completions
• Prompt Format: ChatML
• License: Apache-2.0
• Language: English
• Max Context: 8,192+ tokens

License

This model is licensed under the Apache 2.0 License.

Quantizations

Static GGUF

Imatrix GGUF

EXL2

Open LLM Leaderboard Evaluation Results

Coming soon...

Training Configuration

<details><summary>Click here for Axolotl configs</summary>

Completion SFT

base_model: IntervitensInc/Llama-3.1-Minitron-4B-Width-Base-chatml
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

hub_model_id: jeiku/completion4B
hub_strategy: "all_checkpoints"
push_dataset_to_hub:
hf_use_auth_token: true

datasets:
  - path: Mielikki/Erebus-87k
    type: completion
    field: body

shuffle_merged_datasets: true
val_set_size: 0.0025
output_dir: ./outputs/out

adapter:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

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

wandb_project: EXP4B
wandb_entity:
wandb_watch:
wandb_name: EXP4B
wandb_log_model:

gradient_accumulation_steps: 12
micro_batch_size: 3
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.00001
weight_decay: 0.05

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

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

warmup_ratio: 0.1
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1

debug:
deepspeed: deepspeed_configs/zero3_bf16.json
fsdp:
fsdp_config:

special_tokens:
  pad_token: <|finetune_right_pad_id|>

Instruct SFT

base_model: jeiku/completion4B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

hub_model_id: jeiku/instructered4B
hub_strategy: "all_checkpoints"
push_dataset_to_hub:
hf_use_auth_token: true

datasets:
  - path: FourOhFour/Instruct_Phase
    type: sharegpt
    conversation: chatml

chat_template: chatml

shuffle_merged_datasets: true
val_set_size: 0.0025
output_dir: ./outputs/out

adapter:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

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

wandb_project: EXP4B
wandb_entity:
wandb_watch:
wandb_name: EXP4B
wandb_log_model:

gradient_accumulation_steps: 12
micro_batch_size: 3
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.00001
weight_decay: 0.05

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

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

warmup_ratio: 0.1
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 2

debug:
deepspeed: deepspeed_configs/zero3_bf16.json
fsdp:
fsdp_config:

special_tokens:
  pad_token: <|finetune_right_pad_id|>

Roleplaying SFT

base_model: jeiku/instructered4B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

hub_model_id: jeiku/TheBest4B
hub_strategy: "all_checkpoints"
push_dataset_to_hub:
hf_use_auth_token: true

datasets:
  - path: FourOhFour/RP_Phase
    type: sharegpt
    conversation: chatml

chat_template: chatml

shuffle_merged_datasets: true
val_set_size: 0.0025
output_dir: ./outputs/out

adapter:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

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

wandb_project: EXP4B
wandb_entity:
wandb_watch:
wandb_name: EXP4B
wandb_log_model:

gradient_accumulation_steps: 12
micro_batch_size: 3
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.00001
weight_decay: 0.05

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

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

warmup_ratio: 0.1
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 2

debug:
deepspeed: deepspeed_configs/zero3_bf16.json
fsdp:
fsdp_config:

special_tokens:
  pad_token: <|finetune_right_pad_id|>

KTO

base_model: FourOhFour/Crispy_Crab_4B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

hub_model_id: jeiku/aura4bkto
hub_strategy: "all_checkpoints"
push_dataset_to_hub:
hf_use_auth_token: true

chat_template: chatml

rl: kto
rl_beta: 0.2
kto_desirable_weight: 0.2

datasets:
  - path: anthracite-core/full-opus-chosen-hermes-rejected-kto-v1
    type: chatml.argilla

shuffle_merged_datasets: true
val_set_size: 0.0
output_dir: ./outputs/out

sequence_len: 8192
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: false

wandb_project: Aura-4B
wandb_entity:
wandb_watch:
wandb_name: Aura-4B
wandb_log_model:

gradient_accumulation_steps: 16
micro_batch_size: 2
num_epochs: 2
max_steps: 500

optimizer: adamw_8bit
lr_scheduler: cosine
learning_rate: 0.00001
weight_decay: 0.05

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

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

warmup_steps: 10
evals_per_epoch: 2
eval_table_size:
eval_max_new_tokens: 
saves_per_epoch: 1

debug:
deepspeed: 
fsdp:
fsdp_config:
fsdp:
fsdp_config:

special_tokens:
  pad_token: <|finetune_right_pad_id|>

</details><br>