myyycroft/Qwen3.5-2B-AmbigQA-oft-block-32-short-form-prompt-member-0

Ensemble member 0 (seed 42) from run qwen35_2b_oft_block_32_small_lr_short_form_prompt.

Fine-tuned from Qwen/Qwen3.5-2B on AmbigQA.

Evaluation note

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):

model:
  name: Qwen/Qwen3.5-2B
  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: oft
  oft:
    target_modules: all-linear
    r: null
    block_size: 32
    module_dropout: 0.05
    use_cayley_neumann: true
    num_cayley_neumann_terms: 5
    coft: false
    eps: 6.0e-05
    block_share: false
    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: 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

Final-step ensemble mean ± std (n=5)

• AmbigQA accuracy: 0.0766 ± 0.0035
• AmbigQA AlignScore: 0.1365 ± 0.0063
• IFEval prompt_level_strict_accuracy: 0.6188 ± 0.0178
• MMLU accuracy: 0.6123 ± 0.0067
• train_loss: 1.4985 ± 0.0034
• steps: 1620
• epochs: 3

Per-member (final step)

member	seed	steps	epochs	train_loss	AmbigQA acc	AmbigQA AlignScore	IFEval strict	MMLU acc
0 (this repo)	42	1620	3	1.4978	0.0781	0.1355	0.6250	0.6096
1	1051	1620	3	1.4999	0.0781	0.1363	0.6250	0.6096
2	2060	1620	3	1.4934	0.0781	0.1424	0.5938	0.6053
3	3069	1620	3	1.4991	0.0703	0.1266	0.6406	0.6140
4	4078	1620	3	1.5025	0.0781	0.1416	0.6094	0.6228

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