tavtav/eros-7B-BETA

See axolotl config

axolotl version: 0.4.0

base_model: Alignment-Lab-AI/Alignment-Lab-AIlonger
load_in_8bit: false
load_in_4bit: false
strict: false
tokenizer_type: LlamaTokenizer

datasets:
  - path: PygmalionAI/spice
    type: sharegpt
    conversation: chatml

  - path: PygmalionAI/NYROS
    type: sharegpt
    conversation: chatml

chat_template: chatml

dataset_prepared_path: /workspace/disk2/2prepath2
val_set_size: 0.05
output_dir: /workspace/disk2/Eros2-b
eval_sample_packing: true
sequence_len: 16384
sample_packing: true
pad_to_sequence_len: true
torch_compile: true
hf_use_auth_token: true
hub_strategy: all_checkpoints
hub_model_id: PygmalionAI/Eros-BETA
hub_private_repo: true
push_to_hub: true
wandb_project: Erosium-b
wandb_entity:
wandb_watch: all
overwrite_output_dir: true
wandb_name:
wandb_log_model:
save_safetensors: true
gradient_accumulation_steps: 4
micro_batch_size: 1
num_epochs: 4
optimizer: adamw_bnb_8bit
amsgrad: true
max_grad_norm: 1
lr_scheduler: 'cosine'
lr_scheduler_kwargs:
  num_cycles: 6
learning_rate: 0.00005
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
train_on_inputs: false
group_by_length: true
neftune_noise_alpha: 6
bf16: auto
fp16:
tf32: false
seed: 314159
early_stopping_patience:
local_rank:
logging_steps: 1
log_level: debug
xformers_attention:
flash_attention: true
warmup_steps:
eval_per_epoch: 0.25
save_steps: 0.20
debug:
deepspeed: ./deepspeed_configs/zero2.json
weight_decay: 0.05
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"

Eros-BETA

This model is a fine-tuned version of Alignment-Lab-AI/Alignment-Lab-AIlonger on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.1394

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: 5e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 314159
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 3
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
1.267	1.02	224	1.3057
1.1657	2.02	448	1.2184
1.062	3.02	672	1.1664
0.8812	3.94	880	1.1394

Framework versions

• Transformers 4.39.0.dev0
• Pytorch 2.1.2+cu118
• Datasets 2.18.0
• Tokenizers 0.15.0