GSM8K
PulseAugur coverage of GSM8K — every cluster mentioning GSM8K across labs, papers, and developer communities, ranked by signal.
- instance of DagsHub 90%
- instance of mathematics-dataset 90%
- instance of Gotit.pub 90%
- used by Grpo 90%
- instance of ScienceCast 70%
- instance of alphaXiv 70%
- used by alphaXiv 70%
- instance of CatalyzeX 70%
- used by mathematics-dataset 70%
- instance of Math-500 70%
- used by Qwen3-4B 70%
- instance of MMLU-Pro 70%
25 day(s) with sentiment data
-
New LLM techniques: Byte-level conversion and efficient decoding strategies
Researchers have developed a method to convert a 27B parameter hybrid-attention LLM, Qwen3.8, into a pure byte-level model. This byte-level model, trained on self-generated data, shows competitive performance on several…
-
New model optimizes compute for multistage AI training pipelines
Researchers have developed an analytical model to understand and optimize compute allocation in multistage foundation model training. The model, which combines unsupervised pretraining, supervised fine-tuning (SFT), and…
-
New method boosts small LLM precision with adaptive weight bits
Researchers have developed a novel method for enhancing the precision of small local language models during decoding. This technique involves initially using 4-bit weights and then escalating to 8-bit weights only for s…
-
Concentrated dropout degrades LLM reasoning, study finds
A new arXiv paper investigates the impact of concentrated versus diffuse attention head dropout on the multi-step reasoning abilities of autoregressive Large Language Models. The study found that concentrated bursts of …
-
New BARS method improves LLM accuracy with refusal mechanism
Researchers have developed BARS, a novel sequential sampling method designed to improve the accuracy and efficiency of large language models in identifying majority answers, particularly for complex tasks like mathemati…
-
AI models perform worse with maximum reasoning effort, research finds
New research indicates that forcing AI models to use maximum reasoning effort on every request can lead to negative outcomes, including increased latency, higher costs, and diminished answer quality for many common task…
-
AI model evaluation evolves from token prediction to complex workflows
The evaluation of frontier AI models has evolved through distinct eras, moving from simple token prediction and log-likelihood grading to assessing instruction following, then to verifying math, code, factuality, and sa…
-
NVIDIA Dory 4B model runnable via llama.cpp tutorial
This tutorial guides users through running NVIDIA's Dory 4B model using llama.cpp. It details how to build a specific fork of llama.cpp, download the Dory 1.0 4B model in GGUF format, and analyze its architecture. The g…
-
New LLM engine boosts Qwen3.6 performance by up to 95% over vLLM
A new engine for running large language models (LLMs) has demonstrated significant performance improvements over vLLM, particularly with the Qwen3.6 model. Benchmarks show the engine can achieve up to 95% more output fr…
-
Weight edit on Qwen3.5-0.8B shows marginal gains in replication attempt
A replication attempt of a weight edit on the Qwen3.5-0.8B model revealed that the reported improvements in MATH-500 and GSM8K scores may be influenced by evaluation methodology. The author found that the unedited model…
-
8B model shows strong continual learning, retaining news knowledge over days
A researcher has developed a continual learning method for an 8B parameter model, named Apertus-8B, that allows it to retain information from previous days' news. After four days of learning, the model maintained 90.7% …
-
FILO framework unifies low-rank adaptation for LLMs, improving efficiency
Researchers have introduced FILO, a novel Fisher-Geometric framework designed to unify and improve low-rank adaptation (LoRA) techniques for large language models. FILO addresses the challenge of singular operators in f…
-
New research defines 'behavioral rate' for fine-tuning language models
A new research paper introduces the concept of a "behavioral rate" to quantify the information needed for fine-tuning language models. This rate measures the smallest compressed size of an adapter that retains a specifi…
-
New research explores expert deactivation in LLMs, revealing performance impacts
Researchers have investigated methods for deactivating experts in large language models during inference, finding that the approach significantly impacts performance. Their study compared four deactivation techniques—re…
-
New GIST method enhances offline RLVR by selecting informative data
Researchers have developed GIST (Gibbs-Targeted Hierarchical Data Selection), a novel method for improving offline reinforcement learning with verifiable rewards (RLVR). GIST focuses on selecting more informative respon…
-
New PACT framework and PACT-Check tool explore cross-model transfer in LLMs
Researchers have introduced PLatour, an open-source framework designed to facilitate Prefill-to-Autoregressive Cross-model Transfer (PACT). This method allows a larger language model to pass its prompt representations t…
-
MIT researchers enable AI to fix model weights directly, boosting performance
Researchers from MIT CSAIL have introduced a novel technique called Sparks of Weight-Space Programming, which allows an AI model, GPT-6 Astra, to directly read and modify the weights of a neural network without traditio…
-
New theory explains Chain-of-Thought generalization in LLMs
A new paper explores the theoretical underpinnings of Chain-of-Thought (CoT) reasoning in Large Language Models (LLMs), focusing on its out-of-distribution (OOD) generalization capabilities. The research posits that CoT…
-
New 'Hinted Scaling Laws' Improve AI Model Training Predictions
Researchers have introduced "hinted scaling laws" to improve the accuracy of predictions for large model training runs. This method uses privileged information to create easier versions of benchmarks, ensuring even smal…
-
New method steers LLM reasoning by editing activation explanations
Researchers have developed a new method called VERS (Verbalize, Edit, Reconstruct, Steer) to guide reasoning language models by editing their activation explanations. This technique aims to reduce the number of generate…