OpenHermes-7B

image/png

Model description

OpenHermes 7B is the first fine tune of the Hermes dataset that has a fully open source dataset!

What is unique about this 7B model is that it used sample packing, which speeds up training by many multiples if the dataset token averages arent near the max sequence length.

OpenHermes was trained on 242,000 entries of primarily GPT-4 generated data, from open datasets across the AI landscape, including:

  • GPTeacher - General Instruct, Roleplay v1, Roleplay v2, and Code Instruct Datasets, by Teknium
  • WizardLM (v1, evol_instruct 70k), by WizardLM Team/nlpxucan
  • Airoboros GPT-4 (v1.0), by JonDurbin
  • Camel-AI's domain expert datasets, by the Camel-AI Team
  • CodeAlpaca, by Sahil2801
  • GPT4-LLM and Unnatural Instructions, by Microsoft

Filtering included removal of OpenAI refusals, disclaimers, and "As an AI" type examples and more

The base dataset mix the model was trained on is identical to Nous-Hermes', minus the Nous-Instruct and PDACTL datasets which were private datasets.

The WANDB Project is public and can be examined at this link: https://wandb.ai/teknium1/openhermes/runs/openhermes-v2-qlora-7b-packed

Huge thank you to main_horse for compute access and a16z for sponsoring my work, and all the dataset creators and other people who's work has contributed to this project!

Benchmark Information

Benchmark Results

GPT-4All Benchmark Set

|    Task     |Version| Metric |Value |   |Stderr|
|-------------|------:|--------|-----:|---|-----:|
|arc_challenge|      0|acc     |0.4727|±  |0.0146|
|             |       |acc_norm|0.4957|±  |0.0146|
|arc_easy     |      0|acc     |0.7862|±  |0.0084|
|             |       |acc_norm|0.7643|±  |0.0087|
|boolq        |      1|acc     |0.7801|±  |0.0072|
|hellaswag    |      0|acc     |0.5789|±  |0.0049|
|             |       |acc_norm|0.7654|±  |0.0042|
|openbookqa   |      0|acc     |0.3480|±  |0.0213|
|             |       |acc_norm|0.4500|±  |0.0223|
|piqa         |      0|acc     |0.7867|±  |0.0096|
|             |       |acc_norm|0.7938|±  |0.0094|
|winogrande   |      0|acc     |0.7048|±  |0.0128|

Average: 0.679

BigBench:

|                      Task                      |Version|       Metric        |Value |   |Stderr|
|------------------------------------------------|------:|---------------------|-----:|---|-----:|
|bigbench_causal_judgement                       |      0|multiple_choice_grade|0.5000|±  |0.0364|
|bigbench_date_understanding                     |      0|multiple_choice_grade|0.5908|±  |0.0256|
|bigbench_disambiguation_qa                      |      0|multiple_choice_grade|0.3023|±  |0.0286|
|bigbench_geometric_shapes                       |      0|multiple_choice_grade|0.1003|±  |0.0159|
|                                                |       |exact_str_match      |0.0000|±  |0.0000|
|bigbench_logical_deduction_five_objects         |      0|multiple_choice_grade|0.2520|±  |0.0194|
|bigbench_logical_deduction_seven_objects        |      0|multiple_choice_grade|0.1871|±  |0.0148|
|bigbench_logical_deduction_three_objects        |      0|multiple_choice_grade|0.3833|±  |0.0281|
|bigbench_movie_recommendation                   |      0|multiple_choice_grade|0.2500|±  |0.0194|
|bigbench_navigate                               |      0|multiple_choice_grade|0.5000|±  |0.0158|
|bigbench_reasoning_about_colored_objects        |      0|multiple_choice_grade|0.4370|±  |0.0111|
|bigbench_ruin_names                             |      0|multiple_choice_grade|0.2679|±  |0.0209|
|bigbench_salient_translation_error_detection    |      0|multiple_choice_grade|0.2495|±  |0.0137|
|bigbench_snarks                                 |      0|multiple_choice_grade|0.5249|±  |0.0372|
|bigbench_sports_understanding                   |      0|multiple_choice_grade|0.5406|±  |0.0159|
|bigbench_temporal_sequences                     |      0|multiple_choice_grade|0.2470|±  |0.0136|
|bigbench_tracking_shuffled_objects_five_objects |      0|multiple_choice_grade|0.1944|±  |0.0112|
|bigbench_tracking_shuffled_objects_seven_objects|      0|multiple_choice_grade|0.1509|±  |0.0086|
|bigbench_tracking_shuffled_objects_three_objects|      0|multiple_choice_grade|0.3833|±  |0.0281|
Average: 0.3367

AGI Eval

|             Task             |Version| Metric |Value |   |Stderr|
|------------------------------|------:|--------|-----:|---|-----:|
|agieval_aqua_rat              |      0|acc     |0.2441|±  |0.0270|
|                              |       |acc_norm|0.2402|±  |0.0269|
|agieval_logiqa_en             |      0|acc     |0.2458|±  |0.0169|
|                              |       |acc_norm|0.2965|±  |0.0179|
|agieval_lsat_ar               |      0|acc     |0.2522|±  |0.0287|
|                              |       |acc_norm|0.2130|±  |0.0271|
|agieval_lsat_lr               |      0|acc     |0.2745|±  |0.0198|
|                              |       |acc_norm|0.2686|±  |0.0196|
|agieval_lsat_rc               |      0|acc     |0.2900|±  |0.0277|
|                              |       |acc_norm|0.2379|±  |0.0260|
|agieval_sat_en                |      0|acc     |0.4466|±  |0.0347|
|                              |       |acc_norm|0.3738|±  |0.0338|
|agieval_sat_en_without_passage|      0|acc     |0.3738|±  |0.0338|
|                              |       |acc_norm|0.3301|±  |0.0328|
|agieval_sat_math              |      0|acc     |0.2318|±  |0.0285|
|                              |       |acc_norm|0.1864|±  |0.0263|
Average: 0.2683

TruthfulQA:

hf-causal-experimental (pretrained=teknium/OpenHermes-7B,dtype=float16), limit: None, provide_description: False, num_fewshot: 0, batch_size: 8
|    Task     |Version|Metric|Value |   |Stderr|
|-------------|------:|------|-----:|---|-----:|
|truthfulqa_mc|      1|mc2   |0.4542|±  |0.0148|

Training procedure

image/png

Downloads last month
475
Inference Examples
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.

Model tree for teknium/OpenHermes-7B

Finetuned
(63)
this model
Quantizations
1 model

Dataset used to train teknium/OpenHermes-7B

Collection including teknium/OpenHermes-7B