Hugging Face's logo

jeiku
/
Theory_of_Mind_128_StableLM

PEFT
stablelm_epoch
Generated from Trainer
custom_code
4-bit precision
bitsandbytes
Model card Files Community
Theory_of_Mind_128_StableLM / README.md
jeiku's picture
jeiku
Upload 8 files
6cf9a7b verified
preview code
|
raw
history blame contribute delete
3.74 kB
metadata
license: cc-by-sa-4.0
library_name: peft
tags:
  - generated_from_trainer
base_model: stabilityai/stablelm-3b-4e1t
model-index:
  - name: qlora-out-2
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.3.0

base_model: stabilityai/stablelm-3b-4e1t
base_model_config: stabilityai/stablelm-3b-4e1t
trust_remote_code: true
model_type: AutoModelForCausalLM
tokenizer_type: GPTNeoXTokenizerFast

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: theory_of_mind_airoboros_fixed.json
    type: alpaca

dataset_prepared_path:
val_set_size: 0.005
output_dir: ./qlora-out-2

adapter: qlora

wandb_project: theoryofmind
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

sequence_len: 1024
sample_packing: false
pad_to_sequence_len: true
save_safetensors: false

lora_r: 128
lora_alpha: 256
lora_dropout: 0.05
lora_target_linear: false
lora_fan_in_fan_out:
lora_modules_to_save:
  - embed_tokens
  - lm_head
lora_target_modules:
  - q_proj
  - v_proj

gradient_accumulation_steps: 1
micro_batch_size: 16
num_epochs: 5
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 0.00005

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

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

warmup_steps: 10
evals_per_epoch: 1
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<|endoftext|>"
  eos_token: "<|im_end|>"
  unk_token: "<|endoftext|>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"

qlora-out-2

This model is a fine-tuned version of stabilityai/stablelm-3b-4e1t on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.9864

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

The following bitsandbytes quantization config was used during training:

  • quant_method: bitsandbytes
  • load_in_8bit: False
  • load_in_4bit: True
  • llm_int8_threshold: 6.0
  • llm_int8_skip_modules: None
  • llm_int8_enable_fp32_cpu_offload: False
  • llm_int8_has_fp16_weight: False
  • bnb_4bit_quant_type: nf4
  • bnb_4bit_use_double_quant: True
  • bnb_4bit_compute_dtype: bfloat16

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 5e-05
  • train_batch_size: 16
  • eval_batch_size: 16
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 10
  • num_epochs: 5

Training results

Training Loss Epoch Step Validation Loss
1.928 0.05 1 1.7816
1.2231 1.0 22 1.1896
0.8273 2.0 44 1.0456
0.517 3.0 66 0.9905
1.0244 4.0 88 0.9915
0.6749 5.0 110 0.9864

Framework versions

  • PEFT 0.7.0
  • Transformers 4.37.0.dev0
  • Pytorch 2.0.1+cu117
  • Datasets 2.16.1
  • Tokenizers 0.15.0