llama.cpp version b9675 enables FP16 support for operations like SQR, SQRT, LOG, SIN, COS, and CLAMP. The release includes binaries for macOS, Linux, Android, Windows, and openEuler across multiple architectures, with support for Vulkan, ROCm, OpenVINO, SYCL (FP16 and FP32), and CUDA 12.4 and 13.3.
Local models transitioned from being primarily privacy-focused toys to practical tools for coding, private document management, and local workflows within a year. While they still fall short of replacing top closed models for complex tasks requiring planning and error correction, the overall improvement in usability and performance is evident.
llama.cpp releases version b9680 with updated Vulkan support and new binaries for macOS, Linux, Android, Windows, and openEuler. The release includes CPU and GPU variants for multiple architectures, with support for Vulkan, CUDA, OpenVINO, SYCL, and ROCm.
A user is seeking a locally hosted, self-contained app to generate English subtitles (in .srt or .ass format) from video files. They consider Qwen-ASR and Whisper as strong options but report poor subtitle timing in ComfyUI implementations and unreliable performance with older models like those in storytoolkitAI. They ask for recommendations that work well on Windows and can handle multiple languages.
GLM-5.2, a 744B parameter model from Z.ai, has been evaluated as the top frontend coding model globally, outperforming all Opus versions including Opus 4.8. This achievement is highlighted in third-party evaluations that validate official offline tests, marking a significant milestone for a model of its size, particularly in the competitive frontend coding domain.
LoopCoder-v2, a parallel loop Transformer model, achieves superior code generation and reasoning performance with two loops, improving SWE-bench Verified from 43.0 to 64.4 points and Multi-SWE from 14.0 to 31.0 points. Variants with three or more loops perform worse, indicating a non-monotonic loop-count effect due to growing positional mismatch and diminishing returns.
ScaFE repositions large language models as feature engineers for scar classification, generating executable Python code from clinical criteria to extract interpretable features. The framework achieves superior performance with limited data, preserves privacy by processing images locally, and produces clinically grounded features aligned with established scoring systems like the Vancouver Scar Scale.
This paper outlines a systems-oriented workflow for embedded machine learning on microcontroller-class devices. It details key engineering decisions such as data sampling, feature extraction, class imbalance validation, model-runtime co-design, and streaming deployment, using inertial motion recognition and keyword spotting as case studies. The work provides practical design rules for robust on-device inference, including data curation, quantization, thresholding, scheduling, and field monitoring.
ReproRepo introduces a scalable framework using GitHub issues to evaluate ML paper reproducibility. It shows that LLM agents like Codex with GPT-5.5 identify at least one human-reported blocker in 90% of 1,149 ML papers, highlighting their ability to detect visible failures and semantic issues, though exact localization remains limited.
A modded V620 flashed with W6800 firmware enables mini-DisplayPort output but disables some compute cores, despite higher boost clocks. Benchmarks show Vulkan outperforms ROCm in tensor group (TG) tasks, while ROCm is faster in prompt processing (PP) at longer context depths, with significant performance gains in Q4_K_XL settings.
A user explores cost-effective methods to run GLM 5.x locally using 4-bit quantization, such as IQ4_XS, without relying on unified memory. Options include CPU-only setups like Sapphire Rapids ES with DDR5, multi-GPU offloading, or similar-sized models. The user runs a 5900X + 128GB DDR4 + 7900XT 20GB system, successfully handling Minimax 2.7 at Q4_K_S and Qwen 3.6 27B at IQ4_XS.
ReproRepo introduces a scalable framework using GitHub issues to evaluate ML paper reproducibility. It shows that LLM agents like Codex with GPT-5.5 identify at least one semantically related blocker in 90% of paper-repository pairs without executing code.
LoopCoder-v2, a parallel loop Transformer model, achieves significant gains in code generation and reasoning with two loops, improving SWE-bench Verified from 43.0 to 64.4 points and Multi-SWE from 14.0 to 31.0 points. Variants with three or more loops perform worse, showing diminishing returns and oscillatory updates due to positional mismatches from cross-loop offsets.
ScaFE proposes using large language models as feature engineers to transform medical images into clinically interpretable representations. By generating deterministic Python code from established scar assessment criteria, it extracts features aligned with clinical scoring systems like the Vancouver Scar Scale. The method achieves superior performance under limited data, with advantages in data efficiency, privacy preservation, and interpretability.
The ALeRCE astronomical database introduces a text-to-SQL system using large language models, enabling natural language queries to generate executable SQL. The system, evaluated on 110 NL/SQL pairs, uses a step-by-step framework that outperforms direct-inference baselines, with Claude Opus 4.6 achieving high precision on simple queries and among the best overall performance across evaluated models.
This paper outlines a systems-oriented workflow for embedded machine learning on microcontroller-class devices. It details key engineering decisions such as data sampling, feature extraction, class imbalance validation, model-runtime co-design, and streaming deployment, using inertial motion recognition and keyword spotting as case studies. The work provides practical design rules for robust on-device inference, including data curation, quantization, thresholding, scheduling, and field monitoring.
An empirical study of 86,156 test-file patches from 33,596 agent-authored PRs reveals that 80.2% of test patches contain weak or no explicit oracle signals. Strong-oracle test files significantly improve merge likelihood (OR = 1.28, p < 0.001) after adjusting for multiple factors, indicating test file presence alone overestimates verification strength.
ReproRepo introduces a scalable framework using GitHub issues to evaluate ML paper reproducibility. It shows that LLM agents like Codex with GPT-5.5 identify at least one blocker in 90% of paper-repository pairs without executing code, though exact localization remains challenging.
SwiftTrans addresses runtime efficiency gaps in LLM-based code translation by introducing Multi-Perspective Exploration and Difference-Aware Selection. The framework extends CodeNet, F2SBench, and introduces SwiftBench to evaluate runtime performance, showing consistent improvements in both correctness and efficiency across benchmarks.
Current coding benchmarks were designed before agentic software engineering and fail to capture the complexity of real-world systems. They conflate model performance with the entire harness, ignore valid alternative solutions, and lack feedback signals at individual component levels, making iterative improvement difficult.