README.md

metadata

language:
  - en
license: apache-2.0
datasets:
  - OpenAssistant/oasst_top1_2023-08-25
pipeline_tag: text-generation
model-index:
  - name: tinyllama-oasst1-top1-instruct-full-lr1-5-v0.1
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 32.85
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=habanoz/tinyllama-oasst1-top1-instruct-full-lr1-5-v0.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 58.16
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=habanoz/tinyllama-oasst1-top1-instruct-full-lr1-5-v0.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 25.96
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=habanoz/tinyllama-oasst1-top1-instruct-full-lr1-5-v0.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 38.35
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=habanoz/tinyllama-oasst1-top1-instruct-full-lr1-5-v0.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 57.7
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=habanoz/tinyllama-oasst1-top1-instruct-full-lr1-5-v0.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 0.45
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=habanoz/tinyllama-oasst1-top1-instruct-full-lr1-5-v0.1
          name: Open LLM Leaderboard

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:

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

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	35.58
AI2 Reasoning Challenge (25-Shot)	32.85
HellaSwag (10-Shot)	58.16
MMLU (5-Shot)	25.96
TruthfulQA (0-shot)	38.35
Winogrande (5-shot)	57.70
GSM8k (5-shot)	0.45