TinyLlama-1.1B-intermediate-step-715k-1.5T finetuned using airoboros-3.1-no-mathjson-max-1k dataset.
Qlora is used. Adapter is merged.
SFT code: https://github.com/habanoz/qlora.git
Command used:
accelerate launch $BASE_DIR/qlora/train.py \
--model_name_or_path $BASE_MODEL \
--working_dir $BASE_DIR/$OUTPUT_NAME-checkpoints \
--output_dir $BASE_DIR/$OUTPUT_NAME-peft \
--merged_output_dir $BASE_DIR/$OUTPUT_NAME \
--final_output_dir $BASE_DIR/$OUTPUT_NAME-final \
--num_train_epochs 1 \
--logging_steps 1 \
--save_strategy steps \
--save_steps 75 \
--save_total_limit 2 \
--data_seed 11422 \
--evaluation_strategy steps \
--per_device_eval_batch_size 4 \
--eval_dataset_size 0.01 \
--eval_steps 75 \
--max_new_tokens 1024 \
--dataloader_num_workers 3 \
--logging_strategy steps \
--do_train \
--do_eval \
--lora_r 64 \
--lora_alpha 16 \
--lora_modules all \
--bits 4 \
--double_quant \
--quant_type nf4 \
--lr_scheduler_type constant \
--dataset habanoz/airoboros-3.1-no-mathjson-max-1k \
--dataset_format airoboros_chat \
--model_max_len 1024 \
--per_device_train_batch_size 4 \
--gradient_accumulation_steps 4 \
--learning_rate 1e-5 \
--adam_beta2 0.999 \
--max_grad_norm 0.3 \
--lora_dropout 0.0 \
--weight_decay 0.0 \
--seed 11422 \
--gradient_checkpointing \
--use_flash_attention_2 \
--ddp_find_unused_parameters False
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
Metric | Value |
---|---|
Avg. | 34.98 |
AI2 Reasoning Challenge (25-Shot) | 30.72 |
HellaSwag (10-Shot) | 54.32 |
MMLU (5-Shot) | 24.78 |
TruthfulQA (0-shot) | 41.67 |
Winogrande (5-shot) | 57.62 |
GSM8k (5-shot) | 0.76 |
- Downloads last month
- 723
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Dataset used to train habanoz/TinyLlama-1.1B-intermediate-step-715k-1.5T-lr-5-1epch-airoboros3.1-1k-instruct-V1
Evaluation results
- normalized accuracy on AI2 Reasoning Challenge (25-Shot)test set Open LLM Leaderboard30.720
- normalized accuracy on HellaSwag (10-Shot)validation set Open LLM Leaderboard54.320
- accuracy on MMLU (5-Shot)test set Open LLM Leaderboard24.780
- mc2 on TruthfulQA (0-shot)validation set Open LLM Leaderboard41.670
- accuracy on Winogrande (5-shot)validation set Open LLM Leaderboard57.620
- accuracy on GSM8k (5-shot)test set Open LLM Leaderboard0.760