Edit model card

Built with Axolotl

See axolotl config

axolotl version: 0.4.1

base_model: Fischerboot/LLama3-Lexi-Aura-3Some-SLERP-SLERP-ql-merge
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

chat_template: llama3
datasets:
  - path: Fischerboot/newnewdataset-sophie
    type: sharegpt
  - path: PJMixers/grimulkan_theory-of-mind-ShareGPT
    type: sharegpt
    conversation: llama3
dataset_prepared_path: last_run_prepared
val_set_size: 0.1
output_dir: ./outputs/256-rank-2-datasets

adapter: qlora
lora_model_dir:

sequence_len: 128
sample_packing: false
pad_to_sequence_len: true

lora_r: 256
lora_alpha: 128
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 8
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

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

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

loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3

eval_sample_packing: false
warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<|begin_of_text|>"
  eos_token: "<|end_of_text|>"
  pad_token: "<|end_of_text|>"

outputs/256-rank-2-datasets

This model is a fine-tuned version of Fischerboot/LLama3-Lexi-Aura-3Some-SLERP-SLERP-ql-merge on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.4406

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: 0.0002
  • train_batch_size: 1
  • eval_batch_size: 1
  • 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: 8

Training results

Training Loss Epoch Step Validation Loss
6.1066 0.0034 1 6.0671
0.2818 0.2526 74 0.6377
0.5131 0.5051 148 0.4845
0.4376 0.7577 222 0.4022
0.3492 1.0102 296 0.3534
0.3393 1.2628 370 0.3735
0.4548 1.5154 444 0.4114
0.5944 1.7679 518 0.4224
0.3372 2.0205 592 0.3720
0.5017 2.2730 666 0.4280
0.2303 2.5256 740 0.3827
0.2586 2.7782 814 0.3366
0.1802 3.0307 888 0.3832
0.3442 3.2833 962 0.3267
0.1136 3.5358 1036 0.3215
0.144 3.7884 1110 0.3031
0.2701 4.0410 1184 0.2889
0.1677 4.2935 1258 0.3019
0.2538 4.5461 1332 0.3172
0.1108 4.7986 1406 0.2773
0.1633 5.0512 1480 0.2980
0.1342 5.3038 1554 0.2953
0.1557 5.5563 1628 0.3226
0.2 5.8089 1702 0.3255
0.0051 6.0614 1776 0.3302
0.0164 6.3140 1850 0.3817
0.0249 6.5666 1924 0.4169
0.1055 6.8191 1998 0.4275
0.0044 7.0717 2072 0.4351
0.0156 7.3242 2146 0.4392
0.0524 7.5768 2220 0.4393
0.0382 7.8294 2294 0.4406

Framework versions

  • PEFT 0.11.1
  • Transformers 4.41.1
  • Pytorch 2.1.2+cu118
  • Datasets 2.19.1
  • Tokenizers 0.19.1
Downloads last month
0
Inference API
Unable to determine this model’s pipeline type. Check the docs .

Model tree for Fischerboot/ll3-sophie-2-sets-256