File size: 1,570 Bytes
c0e3f47 4cb348b c0e3f47 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 |
---
license: apache-2.0
datasets:
- OpenAssistant/oasst_top1_2023-08-25
language:
- en
pipeline_tag: text-generation
---
TinyLlama-1.1B-intermediate-step-715k-1.5T finetuned using OpenAssistant/oasst_top1_2023-08-25 dataset.
SFT code:
https://github.com/jzhang38/TinyLlama/tree/main/sft
Evaluation Results at:
https://huggingface.co/datasets/open-llm-leaderboard/details_habanoz__tinyllama-oasst1-top1-instruct-full-lr1-5-v0.1_public/blob/main/results_2023-11-23T17-25-53.937618.json
Command used:
```bash
accelerate launch finetune.py \
--model_name_or_path TinyLlama/TinyLlama-1.1B-intermediate-step-715k-1.5T \
--output_dir ./output/1_5T_FT_lr1e-5_ep5_top1_2023-08-25 \
--logging_steps 10 \
--save_strategy epoch \
--data_seed 42 \
--save_total_limit 2 \
--evaluation_strategy epoch \
--eval_dataset_size 512 \
--max_eval_samples 1000 \
--per_device_eval_batch_size 1 \
--max_new_tokens 32 \
--dataloader_num_workers 3 \
--group_by_length=False \
--logging_strategy steps \
--remove_unused_columns False \
--do_train \
--do_eval \
--warmup_ratio 0.05 \
--lr_scheduler_type constant \
--dataset OpenAssistant/oasst_top1_2023-08-25 \
--dataset_format oasst1 \
--source_max_len 1 \
--target_max_len 1023 \
--per_device_train_batch_size 2 \
--gradient_accumulation_steps 8 \
--max_steps 0 \
--num_train_epochs 5 \
--learning_rate 1e-5 \
--adam_beta2 0.999 \
--max_grad_norm 1.0 \
--weight_decay 0.0 \
--seed 0 \
--trust_remote_code
``` |