SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.05448 · 2024

Task Diversity Shortens the ICL Plateau

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 12 functions out of this paper's own repositories and ran 11 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
sehyunkwon/task-diversity-icl canonical 11 of 12
FunctionStatusWhere it lives
accuracy Ran sehyunkwon/task-diversity-icl/src/bool_task.py
pointer only (licence: NONE) · get_code("d7ae3e05768f422d")
adjust_batch_sizes Ran sehyunkwon/task-diversity-icl/task_configs/generate.py
pointer only (licence: NONE) · get_code("420674d96c9c88ae")
cross_entropy Ran sehyunkwon/task-diversity-icl/src/bool_task.py
pointer only (licence: NONE) · get_code("41065e1207ba1b0d")
cross_entropy Ran sehyunkwon/task-diversity-icl/src/continuous_task.py
pointer only (licence: NONE) · get_code("797281bf91c00aef")
generate_idx_list Ran sehyunkwon/task-diversity-icl/src/bool_task.py
pointer only (licence: NONE) · get_code("1ee7db3dafc63aa0")
get_optimal_ncl_acc Ran sehyunkwon/task-diversity-icl/src/ncl.py
pointer only (licence: NONE) · get_code("71292b149b12d917")
get_optimal_ncl_loss Ran sehyunkwon/task-diversity-icl/src/ncl.py
pointer only (licence: NONE) · get_code("0157552fb1f390d1")
mean_squared_error Ran sehyunkwon/task-diversity-icl/src/continuous_task.py
pointer only (licence: NONE) · get_code("fc94f27a0973e093")
nn_scoring_step Ran sehyunkwon/task-diversity-icl/src/attention_analysis.py
pointer only (licence: NONE) · get_code("a81285c7684656ca")
prefix_scoring_step Ran sehyunkwon/task-diversity-icl/src/attention_analysis.py
pointer only (licence: NONE) · get_code("211a2ef589b849d4")
squared_error Ran sehyunkwon/task-diversity-icl/src/continuous_task.py
pointer only (licence: NONE) · get_code("348d63a78eb4cb94")
get_final_var Not yet run sehyunkwon/task-diversity-icl/src/curriculum.py
pointer only (licence: NONE) · get_code("3c24ac252fef0868")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

In-context learning (ICL) describes a language model's ability to generate outputs based on a set of input demonstrations and a subsequent query. To understand this remarkable capability, researchers have studied simplified, stylized models. These studies have consistently observed long loss plateaus, during which models exhibit minimal improvement, followed by a sudden, rapid surge of learning. In this work, we reveal that training on multiple diverse ICL tasks simultaneously shortens the loss plateaus, making each task easier to learn. This finding is surprising as it contradicts the natural intuition that the combined complexity of multiple ICL tasks would lengthen the learning process, not shorten it. Our result suggests that the recent success in large-scale training of language models may be attributed not only to the richness of the data at scale but also to the easier optimization (training) induced by the diversity of natural language training data.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2410.05448")
get_code_for_paper("2410.05448")
have("2410.05448")

Connect an agent — have() is free.