Hugging Face's logo

ping98k
/
typhoon-7b-en-to-th-lora

Text Generation
PEFT
Thai
mistral
Generated from Trainer
4-bit precision
bitsandbytes
Model card Files Community
typhoon-7b-en-to-th-lora / README.md
ping98k's picture
ping98k
Update README.md
b3702fd verified
preview code
|
raw
history blame contribute delete
11.9 kB
metadata
library_name: peft
tags:
  - generated_from_trainer
base_model: scb10x/typhoon-7b
model-index:
  - name: work/out
    results: []
datasets:
  - scb_mt_enth_2020
language:
  - th
pipeline_tag: text-generation

Built with Axolotl

See axolotl config

axolotl version: 0.4.0

base_model: ./models/scb10x_typhoon-7b
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true

load_in_8bit: false
load_in_4bit: true
strict: false


datasets:
  - path: ./work/scb-mt-en-th-2020/apdf.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/assorted_government.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/generated_reviews_crowd.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/generated_reviews_translator.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/generated_reviews_yn.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/mozilla_common_voice.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/msr_paraphrase.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/nus_sms.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/paracrawl.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/task_master_1.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/thai_websites.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"
  - path: ./work/scb-mt-en-th-2020/wikipedia.csv
    type:
      system_prompt: ""
      field_system: system
      field_instruction: en_text
      field_output: th_text
      format: "{instruction}<translate>"

dataset_prepared_path: ./work/last_run_prepared
val_set_size: 0.02
output_dir: ./work/out


adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

gpu_memory_limit: 20

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:


wandb_project: typhoon-7b
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 2
num_epochs: 1
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.0004

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

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


warmup_ratio: 0.01
eval_steps: 10
eval_table_size:
eval_table_max_new_tokens: 128
save_steps: 10
save_total_limit: 10
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:

ping98k/typhoon-7b-en-to-th-lora

This model was qlora finetuned on the scb_mt_enth_2020 dataset. It achieves the following results on the evaluation set:

  • Loss: 0.8657

Model description

prompt 

Why can camels survive for long without water?<translate>

output 

ทำไมอูฐสามารถอยู่รอดได้นานโดยไม่มีน้ำ

known issue

model not train with end translate token correctly. some time model will output <translate> or </translate>

Why can camels survive for long without water?<translate>ทำไมอูฐสามารถอยู่รอดได้นานโดยไม่มีน้ำ<translate>

Why can camels survive for long without water?<translate>ทำไมอูฐสามารถอยู่รอดได้นานโดยไม่มีน้ำ</translate>

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: 0.0004
  • train_batch_size: 2
  • eval_batch_size: 2
  • seed: 42
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 90
  • num_epochs: 1

Training results

Training Loss Epoch Step Validation Loss
2.8002 0.01 10 2.7164
2.1186 0.02 20 2.0709
1.717 0.03 30 1.6999
1.5327 0.04 40 1.5332
1.3684 0.04 50 1.4293
1.3992 0.05 60 1.3651
1.3031 0.06 70 1.3198
1.3067 0.07 80 1.2831
1.2685 0.08 90 1.2542
1.2469 0.09 100 1.2293
1.2067 0.1 110 1.2096
1.1458 0.11 120 1.1942
1.1679 0.11 130 1.1732
1.1914 0.12 140 1.1609
1.2329 0.13 150 1.1491
1.1151 0.14 160 1.1365
1.1138 0.15 170 1.1252
1.1607 0.16 180 1.1188
1.083 0.17 190 1.1095
1.1068 0.18 200 1.1016
1.1214 0.18 210 1.0921
1.061 0.19 220 1.0862
1.1072 0.2 230 1.0792
1.0275 0.21 240 1.0739
1.0735 0.22 250 1.0666
1.0549 0.23 260 1.0634
1.0336 0.24 270 1.0561
1.0784 0.25 280 1.0519
1.0313 0.26 290 1.0459
1.0459 0.26 300 1.0415
1.0824 0.27 310 1.0390
1.0543 0.28 320 1.0327
1.0732 0.29 330 1.0287
1.0071 0.3 340 1.0237
1.0336 0.31 350 1.0200
1.0694 0.32 360 1.0155
0.9799 0.33 370 1.0111
1.0025 0.33 380 1.0073
0.9805 0.34 390 1.0044
0.9398 0.35 400 1.0011
1.0133 0.36 410 0.9957
1.0465 0.37 420 0.9916
0.9711 0.38 430 0.9887
0.9786 0.39 440 0.9858
0.9687 0.4 450 0.9835
0.988 0.4 460 0.9810
1.021 0.41 470 0.9770
0.9754 0.42 480 0.9734
0.9677 0.43 490 0.9705
1.0114 0.44 500 0.9667
0.978 0.45 510 0.9643
0.9762 0.46 520 0.9611
0.9795 0.47 530 0.9597
0.9419 0.48 540 0.9558
0.9403 0.48 550 0.9519
0.9408 0.49 560 0.9495
0.9704 0.5 570 0.9460
0.9426 0.51 580 0.9447
0.9288 0.52 590 0.9406
0.9986 0.53 600 0.9394
0.9129 0.54 610 0.9374
0.9797 0.55 620 0.9349
0.9269 0.55 630 0.9317
0.9258 0.56 640 0.9296
0.9041 0.57 650 0.9268
0.9383 0.58 660 0.9240
0.9289 0.59 670 0.9220
0.8906 0.6 680 0.9201
0.9275 0.61 690 0.9171
0.99 0.62 700 0.9150
0.9063 0.62 710 0.9124
0.8757 0.63 720 0.9107
0.9276 0.64 730 0.9087
0.9315 0.65 740 0.9064
0.9442 0.66 750 0.9037
0.8848 0.67 760 0.9015
0.8901 0.68 770 0.8993
0.8714 0.69 780 0.8973
0.8641 0.7 790 0.8956
0.8915 0.7 800 0.8938
0.9069 0.71 810 0.8921
0.8798 0.72 820 0.8901
0.9195 0.73 830 0.8884
0.8936 0.74 840 0.8868
0.8284 0.75 850 0.8851
0.9469 0.76 860 0.8833
0.8854 0.77 870 0.8820
0.8865 0.77 880 0.8809
0.8982 0.78 890 0.8799
0.8683 0.79 900 0.8786
0.9326 0.8 910 0.8773
0.8937 0.81 920 0.8758
0.8995 0.82 930 0.8746
0.9263 0.83 940 0.8735
0.907 0.84 950 0.8725
0.8467 0.84 960 0.8715
0.9037 0.85 970 0.8708
0.833 0.86 980 0.8699
0.878 0.87 990 0.8693
0.8897 0.88 1000 0.8686
0.8931 0.89 1010 0.8681
0.8766 0.9 1020 0.8676
0.839 0.91 1030 0.8672
0.8973 0.92 1040 0.8669
0.8806 0.92 1050 0.8666
0.8683 0.93 1060 0.8664
0.8736 0.94 1070 0.8662
0.8495 0.95 1080 0.8660
0.8364 0.96 1090 0.8659
0.8934 0.97 1100 0.8658
0.8954 0.98 1110 0.8658
0.8783 0.99 1120 0.8657
0.8678 0.99 1130 0.8657

Framework versions

  • PEFT 0.7.1
  • Transformers 4.37.0
  • Pytorch 2.0.1+cu118
  • Datasets 2.16.1
  • Tokenizers 0.15.0