While diffusion language models (DLMs) enable fine-grained refinement, their practical controllability remains fragile. We identify and formally characterize a central failure mode--update-forgetting--in which uniform, context-agnostic updates induce token-level fluctuations across timesteps, erasing earlier semantic edits and disrupting the cumulative refinement process, thereby degrading fluency and coherence. As this failure originates in uniform, context-agnostic updates, effective control demands explicit token ordering. We propose Token Timestep Allocation (TTA-DIFFUSION), which realizes soft, semantic token ordering via pertoken timestep schedules: critical tokens are frozen early, while uncertain tokens receive continued refinement. This timestep-based ordering can be instantiated as either a fixed policy or an adaptive policy driven by task signals, thereby supporting a broad spectrum of refinement strategies. Because it operates purely at inference time, it applies uniformly across various DLMs and naturally extends to diverse supervision sources. Empirically, TTA-DIFFUSION improves controllability and fluency: on sentiment control, it yields >20%higher accuracy and nearly halves perplexity using <1/5 the steps; in detoxification, it lowers maximum toxicity (12.2 vs. 14.5) and perplexity (26.0 vs. 32.0). Together, these results demonstrate that softened ordering via timestep allocation is the critical lever for mitigating update-forgetting and achieving stable and controllable diffusion text generation.

In this work, we introduce Erasure of Language Memory (ELM), a principled approach to concept-level unlearning that operates by matching distributions defined by the model's own introspective classification capabilities. Our key insight is that effective unlearning should leverage the model's ability to evaluate its own knowledge, using the language model itself as a classifier to identify and reduce the likelihood of generating content related to undesired concepts. ELM applies this framework to create targeted low-rank updates that reduce generation probabilities for concept-specific content while preserving the model's broader capabilities. We demonstrate ELM's efficacy on biosecurity, cybersecurity, and literary domain erasure tasks. Comparative evaluation reveals that ELM-modified models achieve near-random performance on assessments targeting erased concepts, while simultaneously preserving generation coherence, maintaining benchmark performance on unrelated tasks, and exhibiting strong robustness to adversarial attacks. Our code, data, and trained models are available at elm.baulab.info

More specifically, you will be given the following: 1. An image context: This will describe the contents of an image with sufficient detail to address the instruction.

Foreachprompt, wecompare 6 pairs of models: Quark versus other baselines, as shown in Table 2. These agreement scores are moderate as result of subjectivity involved in ratings of text quality. PPLM (Plug and Play Language Model) uses one or more classifiers to control attributes of model generations. Figure 8: Screenshot of the mechanical turk interfaced used to gather human judgments for the sentimentevaluation. Unlikelihood represents a GPT-2 model fine-tuned with unlikelihoodobjective(Eqn.5)[79].

Generated text may contain offensive or toxic language, contain significant repetition, orbeofadifferent sentiment than desired by the user. We consider thetaskofunlearningthese misalignments byfine-tuning thelanguage model on signals of whatnot to do.

Novel object captioning (NOC) aims to describe images containing objects without observing their ground truth captions during training.

We propose a post-training method for lower-resource languages that preserves fluency of language models even when aligned by disfluent reward models. Preference-optimization is now a well-researched topic, but previous work has mostly addressed models for English and Chinese. Lower-resource languages lack both datasets written by native speakers and language models capable of generating fluent synthetic data. Thus, in this work, we focus on developing a fluent preference-aligned language model without any instruction-tuning data in the target language. Our approach uses an on-policy training method, which we compare with two common approaches: supervised finetuning on machine-translated data and multilingual finetuning. We conduct a case study on Norwegian Bokmål and evaluate fluency through native-speaker assessments. The results show that the on-policy aspect is crucial and outperforms the alternatives without relying on any hard-to-obtain data.