Hugging Face's logo

stillerman
/
starcoder-mathematica

PEFT
Safetensors
gpt_bigcode
Generated from Trainer
8-bit precision
bitsandbytes
Model card Files Community
starcoder-mathematica / README.md
stillerman's picture
stillerman
Upload folder using huggingface_hub
076b539 verified
preview code
|
raw
history blame contribute delete
3.55 kB
metadata
license: bigcode-openrail-m
library_name: peft
tags:
  - generated_from_trainer
base_model: bigcode/starcoder
model-index:
  - name: lora-out
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.3.0

base_model: bigcode/starcoder
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
is_llama_derived_model: false

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: /workspace/axolotl-mdel/mathematica.txt
    type: completion

lora_modules_to_save:
  - embed_tokens
  - lm_head

dataset_prepared_path:
val_set_size: 0.05
output_dir: ./lora-out

sequence_len: 2048
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: starcoder-mathematica
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 1
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

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
s2_attention:

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: "[PAD]"
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

lora-out

This model is a fine-tuned version of bigcode/starcoder on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 3.0968

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: True
  • load_in_4bit: False
  • 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: fp4
  • bnb_4bit_use_double_quant: False
  • bnb_4bit_compute_dtype: float32

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0002
  • train_batch_size: 1
  • eval_batch_size: 1
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 2
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 10
  • num_epochs: 1

Training results

Training Loss Epoch Step Validation Loss
2.8529 0.0 1 3.1576
0.4365 0.25 127 3.1416
2.953 0.5 254 3.1146
0.35 0.75 381 3.0968

Framework versions

  • PEFT 0.7.0
  • Transformers 4.37.0.dev0
  • Pytorch 2.0.1+cu118
  • Datasets 2.16.1
  • Tokenizers 0.15.0