Low-resource safety failures are action failures because the harmfulness representation transfers but the decision calibration does not; this is fixed by recalibrating a high-resource gate with 1-4 target-language examples.
hub Tool reference
Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling
Tool reference. 78% of classified Pith citations use this work as a method, library, or software dependency, not as a substantive claim.
30
Pith papers citing it
Method reference
78% of classified citations
abstract
How do large language models (LLMs) develop and evolve over the course of training? How do these patterns change as models scale? To answer these questions, we introduce \textit{Pythia}, a suite of 16 LLMs all trained on public data seen in the exact same order and ranging in size from 70M to 12B parameters. We provide public access to 154 checkpoints for each one of the 16 models, alongside tools to download and reconstruct their exact training dataloaders for further study. We intend \textit{Pythia} to facilitate research in many areas, and we present several case studies including novel results in memorization, term frequency effects on few-shot performance, and reducing gender bias. We demonstrate that this highly controlled setup can be used to yield novel insights toward LLMs and their training dynamics. Trained models, analysis code, training code, and training data can be found at \url{https://github.com/EleutherAI/pythia}.
hub tools
citation-role summary
dataset 5
background 2
method 2
citation-polarity summary
representative citing papers
Temporarily reducing the learning rate on upper-layer query and key projections during early GPT pretraining prevents premature attention specialization and improves model performance.
The maximum reward gain under KL-regularized LM alignment is a Jeffreys divergence term, estimable as covariance from base samples, with best-of-N approaching the theoretical limit.
Diffusion language models develop early-layer collapse around an indispensable super-outlier due to overtraining, resulting in higher compressibility and reversed optimal sparsity patterns versus autoregressive models.
DDO-RM turns reward scores into a target distribution and applies KL-regularized mirror-descent projection on finite candidates to improve policies, outperforming DPO on Pythia-410M.
A new sequential interaction framework lets LLMs propose questions to forums, with simulations on real Stack Exchange data showing players can reach roughly half the utility of an ideal full-information scenario despite incentive misalignment.
A recurrent-depth architecture enables language models to improve reasoning performance by iterating computation in latent space, achieving gains equivalent to much larger models on benchmarks.
QLoRA finetunes 4-bit quantized LLMs via LoRA adapters to match full-precision performance while using far less memory, enabling 65B-scale training on single GPUs and producing Guanaco models near ChatGPT level.
CDS-trained BabyLMs show earlier and more appropriate production in a new frame-completion task while FineWeb-edu models lead on comprehension benchmarks, indicating current tests underestimate CDS benefits.
Contribution Weights combine attention, value magnitude, and directional alignment to measure token influence more faithfully than attention alone, and show attention sinks actively suppress information via a convex sink-rate to output-norm relationship.
The Shannon Scaling Law treats LLM training as noisy-channel transmission and predicts U-shaped performance degradation when signal-to-noise ratio falls below a threshold, outperforming monotonic scaling laws on Pythia and OLMo2 data.
Causal dimensionality kappa of transformer layers grows sub-linearly with SAE width, remains invariant to model scale, and stays constant across depth while attribution thresholds drop sharply.
Adaptive elastic net SAEs (AEN-SAEs) mitigate feature starvation in SAEs by combining ℓ2 structural stability with adaptive ℓ1 reweighting, producing a Lipschitz-continuous sparse coding map that recovers global feature support under mild assumptions.
A gradient-transport framework with observables D, z, β, δ, v_rel applied to Pico-LM and Pythia datasets shows distinct scaling regimes in duration and efficiency while sharing a near-unity cascade-size backbone.
TOFU loss mitigates the narrowing of generative diversity in LLMs after supervised fine-tuning by addressing neglect of low-frequency patterns and forgetting of prior knowledge.
A unified incentive-score decomposition of preference optimization reveals the disentanglement band condition and reward calibration method that enables suppressing losers while preserving winners in LLM training.
RISE applies CountSketch to dual lexical and semantic channels derived from output-layer gradient outer products, cutting data attribution storage by up to 112x and enabling retrospective and prospective influence analysis on LLMs up to 32B parameters.
Causal interventions reveal that coordination islands block filler-gap mechanisms in Transformers in a gradient way matching humans, yielding the hypothesis that 'and' encodes relational dependencies differently in extractable vs. conjunctive uses.
Pretraining data determines loss-to-loss scaling laws in LLMs, while model size, optimization, tokenizer, and architecture have limited impact.
Repeated sampling scales problem coverage log-linearly with sample count, improving SWE-bench Lite performance from 15.9% to 56% using 250 samples.
MiniLLM distills large language models into smaller ones via reverse KL divergence and on-policy optimization, yielding higher-quality responses with lower exposure bias than standard KD baselines.
Properly filtered web data from CommonCrawl alone trains LLMs that significantly outperform models trained on The Pile, with 600 billion tokens and 1.3B/7.5B parameter models released.
Repeating training data up to 4 epochs yields negligible loss increase versus unique data for fixed compute, and a new scaling law accounts for the decaying value of repeated tokens and excess parameters.
UltraChat supplies 1.5 million high-quality multi-turn dialogues that, when used to fine-tune LLaMA, produce UltraLLaMA, which outperforms prior open-source chat models including Vicuna.
citing papers explorer
-
Low-Resource Safety Failures Are Action Failures, Not Representation Failures
Low-resource safety failures are action failures because the harmfulness representation transfers but the decision calibration does not; this is fixed by recalibrating a high-resource gate with 1-4 target-language examples.
-
Learning Less Is More: Premature Upper-Layer Attention Specialization Hurts Language Model Pretraining
Temporarily reducing the learning rate on upper-layer query and key projections during early GPT pretraining prevents premature attention specialization and improves model performance.
-
Theoretical Limits of Language Model Alignment
The maximum reward gain under KL-regularized LM alignment is a Jeffreys divergence term, estimable as covariance from base samples, with best-of-N approaching the theoretical limit.
-
Layer Collapse in Diffusion Language Models
Diffusion language models develop early-layer collapse around an indispensable super-outlier due to overtraining, resulting in higher compressibility and reversed optimal sparsity patterns versus autoregressive models.
-
DDO-RM: Distribution-Level Policy Improvement after Reward Learning
DDO-RM turns reward scores into a target distribution and applies KL-regularized mirror-descent projection on finite candidates to improve policies, outperforming DPO on Pythia-410M.
-
From Competition to Collaboration: Designing Sustainable Mechanisms Between LLMs and Online Forums
A new sequential interaction framework lets LLMs propose questions to forums, with simulations on real Stack Exchange data showing players can reach roughly half the utility of an ideal full-information scenario despite incentive misalignment.
-
Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
A recurrent-depth architecture enables language models to improve reasoning performance by iterating computation in latent space, achieving gains equivalent to much larger models on benchmarks.
-
QLoRA: Efficient Finetuning of Quantized LLMs
QLoRA finetunes 4-bit quantized LLMs via LoRA adapters to match full-precision performance while using far less memory, enabling 65B-scale training on single GPUs and producing Guanaco models near ChatGPT level.
-
Child-directed speech facilitates production, not comprehension, in BabyLMs
CDS-trained BabyLMs show earlier and more appropriate production in a new frame-completion task while FineWeb-edu models lead on comprehension benchmarks, indicating current tests underestimate CDS benefits.
-
Contribution Weights: A Geometrical Analysis of Self-Attention Transformers
Contribution Weights combine attention, value magnitude, and directional alignment to measure token influence more faithfully than attention alone, and show attention sinks actively suppress information via a convex sink-rate to output-norm relationship.
-
LLMs as Noisy Channels: A Shannon Perspective on Model Capacity and Scaling Laws
The Shannon Scaling Law treats LLM training as noisy-channel transmission and predicts U-shaped performance degradation when signal-to-noise ratio falls below a threshold, outperforming monotonic scaling laws on Pythia and OLMo2 data.
-
Causal Dimensionality of Transformer Representations: Measurement, Scaling, and Layer Structure
Causal dimensionality kappa of transformer layers grows sub-linearly with SAE width, remains invariant to model scale, and stays constant across depth while attribution thresholds drop sharply.
-
Feature Starvation as Geometric Instability in Sparse Autoencoders
Adaptive elastic net SAEs (AEN-SAEs) mitigate feature starvation in SAEs by combining ℓ2 structural stability with adaptive ℓ1 reweighting, producing a Lipschitz-continuous sparse coding map that recovers global feature support under mild assumptions.
-
Finite-Size Gradient Transport in Large Language Model Pretraining: From Cascade Size to Intensive Transport Efficiency
A gradient-transport framework with observables D, z, β, δ, v_rel applied to Pico-LM and Pythia datasets shows distinct scaling regimes in duration and efficiency while sharing a near-unity cascade-size backbone.
-
Diversity in Large Language Models under Supervised Fine-Tuning
TOFU loss mitigates the narrowing of generative diversity in LLMs after supervised fine-tuning by addressing neglect of low-frequency patterns and forgetting of prior knowledge.
-
Towards Disentangled Preference Optimization Dynamics: Suppress the Loser, Preserve the Winner
A unified incentive-score decomposition of preference optimization reveals the disentanglement band condition and reward calibration method that enables suppressing losers while preserving winners in LLM training.
-
Sketching the Readout of Large Language Models for Scalable Data Attribution and Valuation
RISE applies CountSketch to dual lexical and semantic channels derived from output-layer gradient outer products, cutting data attribution storage by up to 112x and enabling retrospective and prospective influence analysis on LLMs up to 32B parameters.
-
Causal Drawbridges: Characterizing Gradient Blocking of Syntactic Islands in Transformer LMs
Causal interventions reveal that coordination islands block filler-gap mechanisms in Transformers in a gradient way matching humans, yielding the hypothesis that 'and' encodes relational dependencies differently in extractable vs. conjunctive uses.
-
LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws
Pretraining data determines loss-to-loss scaling laws in LLMs, while model size, optimization, tokenizer, and architecture have limited impact.
-
Large Language Monkeys: Scaling Inference Compute with Repeated Sampling
Repeated sampling scales problem coverage log-linearly with sample count, improving SWE-bench Lite performance from 15.9% to 56% using 250 samples.
-
MiniLLM: On-Policy Distillation of Large Language Models
MiniLLM distills large language models into smaller ones via reverse KL divergence and on-policy optimization, yielding higher-quality responses with lower exposure bias than standard KD baselines.
-
The RefinedWeb Dataset for Falcon LLM: Outperforming Curated Corpora with Web Data, and Web Data Only
Properly filtered web data from CommonCrawl alone trains LLMs that significantly outperform models trained on The Pile, with 600 billion tokens and 1.3B/7.5B parameter models released.
-
Scaling Data-Constrained Language Models
Repeating training data up to 4 epochs yields negligible loss increase versus unique data for fixed compute, and a new scaling law accounts for the decaying value of repeated tokens and excess parameters.
-
Enhancing Chat Language Models by Scaling High-quality Instructional Conversations
UltraChat supplies 1.5 million high-quality multi-turn dialogues that, when used to fine-tune LLaMA, produce UltraLLaMA, which outperforms prior open-source chat models including Vicuna.
-
Features have life history. And we should care
Language model features form an early stable carrier scaffold of about 50 sparse features that is load-bearing, predictable from onset firing, and recruits most later features.
-
Failure Modes of Maximum Entropy RLHF
Derives SimPO from MaxEnt RL and reports that MaxEnt RL in online RLHF exhibits frequent overoptimization and unstable KL dynamics across scales, unlike stable KL-constrained baselines.
-
Similarity Field Theory: A Mathematical Framework for Intelligence
Similarity Field Theory defines a similarity field over entities, concepts as superlevel-set fibers, and intelligence as a generative operator that preserves fiber membership under evolution.
-
StarCoder: may the source be with you!
StarCoderBase matches or beats OpenAI's code-cushman-001 on multi-language code benchmarks; the Python-fine-tuned StarCoder reaches 40% pass@1 on HumanEval while retaining other-language performance.
-
Baichuan 2: Open Large-scale Language Models
Baichuan 2 presents 7B and 13B LLMs trained on 2.6T tokens that match or exceed similar open models on MMLU, CMMLU, GSM8K, HumanEval and excel in medicine and law.
-
A Survey of Large Language Models
This survey reviews the background, key techniques, and evaluation methods for large language models, emphasizing emergent abilities that appear at large scales.