Papers
Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: What is the impact of model scaling (e.g., 7B vs. 70B parameters) on LLaMA's language understanding capabilities as measured by GLUE or SuperGLUE benchmarks. 0 claims were extracted from source literature; 0 were…
Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How do alignment techniques (e.g., RLHF, DPO) affect GPT-4's performance on code generation tasks, as evaluated by HumanEval or MBPP benchmarks. 10 claims were extracted from source literature; 9 were…
Abstract: This report synthesises findings from 3 peer-reviewed papers addressing the following research question: How does Qwen2-VL's accuracy on GSM8K-V compare to LLaVA-1.6 and InternVL when evaluating robustness to visual noise in grade school math problems. 0 claims were extracted from source literature; 0 were…
Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What is the performance gap between Claude-Sonnet-3.5 and distilled mobile models on context-dependent language understanding tasks specific to mobile interaction patterns. 11 claims were extracted from source…
Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: What is the impact of chain-of-thought prompting on Qwen2-VL's performance trajectory across the MathVista benchmark compared to baseline zero-shot inference. 11 claims were extracted from source literature; 10…
Abstract: This report synthesises findings from 2 peer-reviewed papers addressing the following research question: What is the accuracy degradation of Llama-3.1-70B on MMSU emotional intent classification when acoustic features are perturbed. 10 claims were extracted from source literature; 10 were independently verified…
Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: What is the accuracy-throughput tradeoff of ARS when applied to InternVL3-8B on code generation tasks, as measured by HumanEval or MBPP benchmarks. 7 claims were extracted from source literature; 7 were…
Abstract: This report synthesises findings from 1 peer-reviewed paper addressing the following research question: What is the failure rate of Gemini-2.5-Flash on edge-case coding tasks when evaluated for five-nines reliability standards. 8 claims were extracted from source literature; 7 were independently verified against…
Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does LongVA-7B perform on diagram-based code generation tasks compared to LLaVA-1.6 and Qwen-VL on the HumanEval-V benchmark. 13 claims were extracted from source literature; 1 was independently verified…
Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does Video-LLaVA-8B compare to LLaVA-NeXT on the HumanEval-V benchmark for diagram-based code generation accuracy. 13 claims were extracted from source literature; 1 was independently verified against…
Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How does the alignment of Foundation-Sec-8B with human preferences or safety constraints affect its performance on reasoning tasks in visual contexts, measured by accuracy and bias metrics across. 16 claims were…
Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How do Large Multimodal Models compare in accuracy on HumanEval-V tasks when evaluated against other multimodal benchmarks like MMBench or DAVIS. 12 claims were extracted from source literature; 1 was…
Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How does the performance of gemma-2-2B compare to other 2B-parameter models on the Mobile-MMLU benchmark, particularly in low-resource settings with limited storage and computational constraints. 9 claims were…
Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How do multimodal models compare in performance on HumanEval-V tasks when evaluated for low-resource diagram understanding with limited training data. 0 claims were extracted from source literature; 0 were…
Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does the performance of open-source multimodal models scale with model size on HumanEval-V benchmarks compared to proprietary models. 9 claims were extracted from source literature; 9 were independently…
Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: Can fine-tuning Phi-4 on additional synthetic visual-math datasets improve its robustness on out-of-distribution GSM8K-V problems. 10 claims were extracted from source literature; 9 were independently verified…
Abstract: This report synthesises findings from 3 peer-reviewed papers addressing the following research question: What is the impact of varying image resolution and text complexity on Phi-4's reasoning performance in grade school math word problems with visual contexts. 9 claims were extracted from source literature; 9 were…
Abstract: This report synthesises findings from 3 peer-reviewed papers addressing the following research question: What is the accuracy drop of codellambda-7b-hf-float16 on RoundTripCodeEval when subjected to adversarial code perturbations, and how does this robustness compare to Llama-2-13b and WizardCoder-13b. 8 claims were…
Abstract: This report synthesises findings from 6 peer-reviewed papers addressing the following research question: How does the iterative refinement process in Self-Refine affect the response quality of codegen-2b on the HELM benchmark for language understanding tasks, and what is the quantitative difference in. 10 claims were…
Abstract: This report synthesises findings from 1 peer-reviewed paper addressing the following research question: What is the comparative reasoning accuracy of codellama-7b-hf-float16 on binary analysis tasks versus other specialized LLMs (e.g., BinGPT, assemblyLLM) using the BinMetric benchmark. 10 claims were extracted from…
Abstract: This report synthesises findings from 7 peer-reviewed papers addressing the following research question: How do different quantization techniques (e.g., 8-bit, 4-bit) affect the performance of codellama-7b-hf on binary analysis tasks compared to the float16 baseline. 9 claims were extracted from source literature; 9…
Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does Lugha-Llama-8B-wura perform on African language reasoning benchmarks compared to base Llama 8B. 12 claims were extracted from source literature; 8 were independently verified against retrieved documents.…
Abstract: This report synthesises findings from 2 peer-reviewed papers addressing the following research question: What is the performance gap of XGLM-564M between Indonesian and English language understanding benchmarks across different educational difficulty levels. 6 claims were extracted from source literature; 6 were…
Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What are the comparative MT-bench conversation quality scores between Phi-3-mini and Mistral-7B-v0.1 across diverse dialogue domains. 14 claims were extracted from source literature; 14 were independently…
Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: What are the comparative MT-bench conversation quality scores between Phi-3-mini and InternVL2-8B when evaluated on multi-turn instruction following tasks. 14 claims were extracted from source literature; 14 were…