SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2403.06963 · ICML · 2024

The Pitfalls of Next-Token Prediction

Gregor Bachmann, Vaishnavh Nagarajan

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 7 functions out of this paper's own repositories and ran 6 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
gregorbachmann/next-token-failures canonical 6 of 7
FunctionStatusWhere it lives
accuracy Ran gregorbachmann/next-token-failures/utils/training_utils.py
pointer only (licence: NONE) · get_code("c1f4b077a9b230f4")
gelu Ran gregorbachmann/next-token-failures/models/lib.py
pointer only (licence: NONE) · get_code("272ef5f104a11d59")
get_lr Ran gregorbachmann/next-token-failures/utils/training_utils.py
pointer only (licence: NONE) · get_code("1b9a2fc53c54c9be")
get_run_name Ran gregorbachmann/next-token-failures/utils/training_utils.py
pointer only (licence: NONE) · get_code("e64b971fffaa3f52")
load_pythia Ran gregorbachmann/next-token-failures/utils/load.py
pointer only (licence: NONE) · get_code("d8ecc47b9ccc6855")
remap Ran gregorbachmann/next-token-failures/utils/load.py
pointer only (licence: NONE) · get_code("ca87fc1303002ce5")
load_gpt Not yet run gregorbachmann/next-token-failures/utils/load.py
pointer only (licence: NONE) · get_code("33eb4da2120207ac")

Repositories linked to this paper

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

Abstract

Can a mere next-token predictor faithfully model human intelligence? We crystallize this emerging concern and correct popular misconceptions surrounding it, and advocate a simple multi-token objective. As a starting point, we argue that the two often-conflated phases of next-token predictionautoregressive inference and teacher-forced training -must be treated distinctly. The popular criticism that errors can compound during autoregressive inference, crucially assumes that teacherforcing has learned an accurate next-token predictor. This assumption sidesteps a more deep-rooted problem we expose: in certain classes of tasks, teacher-forcing can simply fail to learn an accurate next-token predictor in the first place. We describe a general mechanism of how teacherforcing can fail, and design a minimal planning task where both the Transformer and the Mamba architecture empirically fail in that manner -remarkably, despite the task being straightforward to learn. Finally, we provide preliminary evidence that this failure can be resolved using teacherless training, a simple modification using dummy tokens that predicts multiple tokens in advance. We hope this finding can ground future debates and inspire explorations beyond the next-token prediction paradigm. We make our code available under https://github.com/gregorbachmann/ Next-Token-Failures * Equal contribution 1 ETH Zürich, Switzerland 2 Google Research, US.

For agents

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

get_harvested_code_for_paper("2403.06963")
get_code_for_paper("2403.06963")
have("2403.06963")

Connect an agent — have() is free.