Continuous language diffusion works by entering high-margin decoder basins where frozen T5 embeddings recover 93-96% of native decisions and linear readouts reach 97.9% agreement, implying models should be evaluated as representation-decoder systems.
T5gemma 2: Seeing, reading, and understanding longer
9 Pith papers cite this work. Polarity classification is still indexing.
9
Pith papers citing it
citation-role summary
background 2
method 1
citation-polarity summary
years
2026 9representative citing papers
KaLM-Reranker-V1 uses encoder–decoder FBNL with Matryoshka pooling to match Qwen3-class reranking quality at substantially lower online cost.
Introduces V-RAGBench benchmark and CARVE method that selects per-chunk retrieval configurations via parallel retrievers and adaptive reranking, outperforming eight VideoRAG baselines.
BoostLLM trains sequential PEFT adapters in a boosting framework with tree path inputs to improve LLM performance on few-shot tabular classification, matching or exceeding XGBoost.
Attention-based models can retrieve evidence intrinsically by using decoder attention to score and reuse their own pre-encoded chunks, outperforming separate retrieval pipelines on QA benchmarks.
LinguDistill recovers approximately 10% of lost performance on language benchmarks in VLMs by selectively distilling from a frozen LM teacher using KV-cache sharing, while preserving vision performance.
With matched data and sizes, CTC+LLM shallow fusion beats tight speech-LLM integration on in-domain ASR, while prefix LLMs win average WER on out-of-domain HuggingFace sets.
S5-TTS introduces a streaming T5-TTS variant with lookahead-causal masking and interleaved multi-source distillation that achieves comparable quality to full-context models while cutting end-to-end latency.
Motif-Video 2B reaches 83.76% on VBench, outperforming a 14B-parameter model with 7x fewer parameters and far less training data through shared cross-attention and a three-part backbone.
citing papers explorer
-
Continuous Language Diffusion as a Decoder-Interface Problem
Continuous language diffusion works by entering high-margin decoder basins where frozen T5 embeddings recover 93-96% of native decisions and linear readouts reach 97.9% agreement, implying models should be evaluated as representation-decoder systems.
-
KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking
KaLM-Reranker-V1 uses encoder–decoder FBNL with Matryoshka pooling to match Qwen3-class reranking quality at substantially lower online cost.
-
Rethinking RAG in Long Videos: What to Retrieve and How to Use It?
Introduces V-RAGBench benchmark and CARVE method that selects per-chunk retrieval configurations via parallel retrievers and adaptive reranking, outperforming eight VideoRAG baselines.
-
BoostLLM: Boosting-inspired LLM Fine-tuning for Few-shot Tabular Classification
BoostLLM trains sequential PEFT adapters in a boosting framework with tree path inputs to improve LLM performance on few-shot tabular classification, matching or exceeding XGBoost.
-
Retrieval from Within: An Intrinsic Capability of Attention-Based Models
Attention-based models can retrieve evidence intrinsically by using decoder attention to score and reuse their own pre-encoded chunks, outperforming separate retrieval pipelines on QA benchmarks.
-
LinguDistill: Recovering Linguistic Ability in Vision-Language Models via Selective Cross-Modal Distillation
LinguDistill recovers approximately 10% of lost performance on language benchmarks in VLMs by selectively distilling from a frozen LM teacher using KV-cache sharing, while preserving vision performance.
-
LLMs and Speech: Integration vs. Combination
With matched data and sizes, CTC+LLM shallow fusion beats tight speech-LLM integration on in-domain ASR, while prefix LLMs win average WER on out-of-domain HuggingFace sets.
-
Streaming T5-based Text-to-Speech Synthesis with Limited Lookahead
S5-TTS introduces a streaming T5-TTS variant with lookahead-causal masking and interleaved multi-source distillation that achieves comparable quality to full-context models while cutting end-to-end latency.
-
Motif-Video 2B: Technical Report
Motif-Video 2B reaches 83.76% on VBench, outperforming a 14B-parameter model with 7x fewer parameters and far less training data through shared cross-attention and a three-part backbone.