Metrics below are computed on small fixed subsets, not full benchmarks. Exact subset sizes: AmbigQA (128), IFEval (64), MMLU (228). Reported values are from the final training step (step=1620, epoch=3).

Dataset

Hub dataset: sewon/ambig_qa (config_name=light)
Revision: 6e667596df70f17ba3c8e7be4b7361f6be8b60f8
Splits: train=train, validation=validation
Dev holdout: dev_fraction=0.1, split_seed=1729
Bootstrap resampling: False

System prompt

No system prompt (dataset.system_prompt: null).

Hyperparameters

Exact resolved settings used for this run (also attached as resolved_config.yaml):

yaml

model:
  name: google/gemma-4-E2B-it
  revision: main
  dtype: bfloat16
  enable_thinking: false
dataset:
  name: sewon/ambig_qa
  config_name: light
  revision: 6e667596df70f17ba3c8e7be4b7361f6be8b60f8
  train_split: train
  validation_split: validation
  dev_fraction: 0.1
  split_seed: 1729
  bootstrap: false
  max_train_examples: null
  max_dev_examples: null
  system_prompt_set: false
adaptation:
  method: lora
  lora:
    r: 8
    alpha: 16
    dropout: 0.0
    target_modules: all-linear
    bias: none
training:
  num_train_epochs: 3.0
  max_steps: -1
  learning_rate: 5.0e-05
  weight_decay: 0.01
  warmup_ratio: 0.03
  per_device_train_batch_size: 16
  per_device_eval_batch_size: 32
  gradient_accumulation_steps: 2
  max_seq_length: 512
  logging_steps: 10
  save_every_steps: 150
  save_total_limit: 3
  gradient_checkpointing: false
  bf16: true
  tf32: true
  max_grad_norm: 1.0
  dataloader_num_workers: 2
  adam_beta1: 0.9
  adam_beta2: 0.999
  adam_epsilon: 1.0e-08
ensemble:
  size: 5
  base_seed: 42
  this_member_index: 2
  this_member_seed: 2060
evaluation_subsets:
  ambigqa:
    subset_size: 128
    seed: 1001
  ifeval:
    subset_size: 64
    seed: 1002
  mmlu:
    subset_size: 228
    seed: 1004

Metrics breakdown

The metrics in this section use small fixed subsets, not full benchmarks: AmbigQA (128), IFEval (64), MMLU (228).

Final-step ensemble mean ± std (n=5)

AmbigQA (128) accuracy: 0.1094 ± 0.0078
AmbigQA (128) AlignScore: 0.1979 ± 0.0157
IFEval (64) prompt_level_strict_accuracy: 0.4781 ± 0.0450
MMLU (228) accuracy: 0.5351 ± 0.0215
train_loss: 1.7738 ± 0.0048
steps: 1620
epochs: 3

Per-member (final step)

member	seed	steps	epochs	train_loss	AmbigQA (128) acc	AmbigQA (128) AlignScore	IFEval (64) strict	MMLU (228) acc
0	42	1620	3	1.7787	0.1094	0.2244	0.4688	0.5395
1	1051	1620	3	1.7768	0.1016	0.1831	0.4219	0.5219
2 (this repo)	2060	1620	3	1.7753	0.1016	0.1954	0.4531	0.5088
3	3069	1620	3	1.7663	0.1172	0.1960	0.5312	0.5658
4	4078	1620	3	1.7720	0.1172	0.1907	0.5156	0.5395

Files

Model weights / adapter files from members/member_002/final/
resolved_config.yaml — full resolved training config
ensemble_metrics.png — train/eval curves for the whole ensemble
run_artifacts/ — ensemble-level manifests, status, and captured environment
run_artifacts/members/member_002/ — this member's manifests, metadata, and status
run_artifacts/members/member_002/predictions/ — per-dataset JSONL predictions from every intermediate evaluation step