Hugging Face's logo

jtatman
/
pythia-160m-storytelling

Text Generation
Transformers
PyTorch
gpt_neox
axolotl
relora
Generated from Trainer
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
pythia-160m-storytelling / README.md
jtatman's picture
jtatman
End of training
a4a408b verified
preview code
|
raw
history blame
3.89 kB
metadata
base_model: EleutherAI/pythia-160m-deduped
library_name: peft
license: apache-2.0
tags:
  - axolotl
  - relora
  - generated_from_trainer
model-index:
  - name: pythia-160m-storytelling
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.4.1

base_model: EleutherAI/pythia-160m-deduped
load_in_8bit: 
datasets:
  - path: jtatman/storywriting_combined_instruct
    type: alpaca
dataset_prepared_path: ds-storytelling
chat_template: inst
val_set_size: 0.01
adapter: lora
lora_model_dir: 
sequence_len: 2048
lora_r: 16
lora_alpha: 32
lora_dropout: 0.05
lora_target_modules:
  - query_key_value
lora_target_linear: true
lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
lora_modules_to_save:
  - embed_in
  - embed_out
  - lm_head
lora_on_cpu: false
# ReLoRA configuration
# # Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed
# relora_steps: # Number of steps per ReLoRA restart
# relora_warmup_steps: # Number of per-restart warmup steps
# relora_anneal_steps: # Number of anneal steps for each relora cycle
# relora_prune_ratio: # threshold for optimizer magnitude when pruning
# relora_cpu_offload:  # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings
relora_steps: 200
relora_warmup_steps: 10
relora_cpu_offload: false
wandb_project: pythia
wandb_entity:
wandb_watch:
wandb_name: pythia-160m-storytelling
wandb_log_model:
output_dir: ./outputs/lora-alpaca-pythia-160m-storytelling
gradient_accumulation_steps: 16
micro_batch_size: 1
num_epochs: 3
learning_rate: 0.0006
lr_scheduler: cosine_with_restarts
#cosine_min_lr_ratio: 0.1
train_on_inputs: false
group_by_length: false
#bf16: auto
#fp16: true
#tf32: false
float16: true
flash_attn: 
xformers_attention: true
optimizer: paged_adamw_8bit
gpu_memory_limit: 8GiB
hub_model_id: jtatman/pythia-160m-storytelling 
early_stopping_patience: 2
#resume_from_checkpoint: outputs/lora-alpaca-pythia-125m/checkpoint-51040
auto_resume_from_checkpoints: true
local_rank:
weight_decay: 0.0
#evals_per_epoch: 4
eval_steps: 200
logging_steps: 1
save_steps: 200
save_total_limit: 5
warmup_steps: 100
tokens:
  - "[INST]"
  - "[/INST]"

pythia-160m-storytelling

This model is a fine-tuned version of EleutherAI/pythia-160m-deduped on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 5.0363

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: 0.0006
  • train_batch_size: 1
  • eval_batch_size: 1
  • seed: 42
  • gradient_accumulation_steps: 16
  • total_train_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine_with_restarts
  • lr_scheduler_warmup_steps: 100
  • num_epochs: 3

Training results

Training Loss Epoch Step Validation Loss
5.5185 0.0012 1 4.8238
4.2012 0.2348 200 4.1556
4.4185 0.4696 400 4.8159
5.0973 0.7043 600 5.0363

Framework versions

  • PEFT 0.11.1
  • Transformers 4.41.2
  • Pytorch 2.3.0+cu121
  • Datasets 2.19.1
  • Tokenizers 0.19.1