pmahdavi/Olmo-3.1-7B-Math-Pruned-50

merged-model

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

Merge Details

Merge Method

This model was merged using the TIES merge method using allenai/Olmo-3-1025-7B as a base.

Models Merged

The following models were included in the merge:

• allenai/Olmo-3.1-7B-RL-Zero-Math

Configuration

The following YAML configuration was used to produce this model:

# Magnitude pruning on OLMo-3.1-7B-RL-Zero-Math task vector
#
# Logic:
#   1. Compute task vector: delta = finetuned - base
#   2. Keep top 50% of delta by absolute magnitude, zero the rest
#   3. Output = base + pruned_delta
#
# Usage:
#   modal run modal_merge.py --config examples/olmo-math-magnitude-prune.yaml --hf-repo pmahdavi/Olmo-3.1-7B-Math-Pruned-50

merge_method: ties
base_model: allenai/Olmo-3-1025-7B

models:
  - model: allenai/Olmo-3.1-7B-RL-Zero-Math
    parameters:
      weight: 1.0
      density: 0.5  # Keep top 50% by magnitude

parameters:
  normalize: false  # Don't normalize (single model, no effect anyway)

dtype: bfloat16