Modelpublic

yuhengtu-bytedance/DataDecide-dolma1_6plus-1B-0_2500_5000_7500_10000_weightedavg_merge

sourceHugging Faceupdated 13d agoView on Hugging Face

0likes67downloads

Model Card

DataDecide-dolma16plus-1B-025005000750010000weightedavg_merge

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the Linear merge method using /opt/tiger/PanSafetyBetterMeasurement/mergescalingckptscache/sourceckpts/dolma16plus/step10000 as a base.

Models Merged

The following models were included in the merge:

—/opt/tiger/PanSafetyBetterMeasurement/mergescalingckptscache/sourceckpts/dolma16plus/step2500
—/opt/tiger/PanSafetyBetterMeasurement/mergescalingckptscache/sourceckpts/dolma16plus/step5000
—/opt/tiger/PanSafetyBetterMeasurement/mergescalingckptscache/sourceckpts/dolma16plus/step7500
—/opt/tiger/PanSafetyBetterMeasurement/mergescalingckptscache/sourceckpts/dolma16plus/step0

Configuration

The following YAML configuration was used to produce this model:

yaml

models:
- model: /opt/tiger/Pan_Safety_Better_Measurement/merge_scaling_ckpts_cache/source_ckpts/dolma1_6plus/step0
  parameters:
    weight: 1
- model: /opt/tiger/Pan_Safety_Better_Measurement/merge_scaling_ckpts_cache/source_ckpts/dolma1_6plus/step2500
  parameters:
    weight: 2
- model: /opt/tiger/Pan_Safety_Better_Measurement/merge_scaling_ckpts_cache/source_ckpts/dolma1_6plus/step5000
  parameters:
    weight: 3
- model: /opt/tiger/Pan_Safety_Better_Measurement/merge_scaling_ckpts_cache/source_ckpts/dolma1_6plus/step7500
  parameters:
    weight: 4
- model: /opt/tiger/Pan_Safety_Better_Measurement/merge_scaling_ckpts_cache/source_ckpts/dolma1_6plus/step10000
  parameters:
    weight: 5
base_model: /opt/tiger/Pan_Safety_Better_Measurement/merge_scaling_ckpts_cache/source_ckpts/dolma1_6plus/step10000
merge_method: linear
parameters:
  normalize: true
dtype: float32
out_dtype: bfloat16