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

You answer open-domain factoid questions with a short-form answer only.

Reply with just the answer string: typically a name, place, organization, title, date, number, or brief factual phrase (often about 1–4 words). Do not write a sentence, explanation, or preamble.

Examples of the expected style:
- Who painted The Persistence of Memory? → Salvador Dalí
- When was the first iPhone released in the United States? → June 29, 2007
- What is the capital of New Zealand? → Wellington
- Who wrote Pride and Prejudice? → Jane Austen
- What is the chemical symbol for gold? → Au
- Who directed the 1994 film Pulp Fiction? → Quentin Tarantino

Formatting rules:
- Output the answer alone on a single line.
- Do not restate the question, add reasoning, markdown, bullets, or quotation marks around the answer.
- Prefer a canonical short form over a descriptive sentence.
- Give one answer only; do not list alternatives or hedge.
- Use natural short forms for dates and names (e.g. June 29, 2007; Jane Austen).

Hyperparameters

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

yaml

model:
  name: google/gemma-4-E4B-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: true
adaptation:
  method: full
training:
  num_train_epochs: 3.0
  max_steps: -1
  learning_rate: 2.0e-05
  weight_decay: 0.1
  warmup_ratio: 0.03
  per_device_train_batch_size: 4
  per_device_eval_batch_size: 16
  gradient_accumulation_steps: 8
  max_seq_length: 512
  logging_steps: 10
  save_every_steps: null
  save_total_limit: null
  gradient_checkpointing: true
  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: 0
  this_member_seed: 42
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: 0.1484 ± 0.0124
AmbigQA (128) AlignScore: 0.2178 ± 0.0149
IFEval (64) prompt_level_strict_accuracy: 0.8500 ± 0.0489
MMLU (228) accuracy: 0.7465 ± 0.0192
train_loss: 1.1823 ± 0.0039
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 (this repo)	42	1620	3	1.1850	0.1641	0.2424	0.8594	0.7456
1	1051	1620	3	1.1866	0.1484	0.2113	0.8594	0.7193
2	2060	1620	3	1.1763	0.1406	0.2094	0.7656	0.7719
3	3069	1620	3	1.1820	0.1562	0.2208	0.8750	0.7412
4	4078	1620	3	1.1819	0.1328	0.2051	0.8906	0.7544

Files

Model weights / adapter files from members/member_000/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_000/ — this member's manifests, metadata, and status
run_artifacts/members/member_000/predictions/ — per-dataset JSONL predictions from every intermediate evaluation step