Open Source LLM

Qualcomm Open-Sources The ELD Linker Embedded Linker

While LLVM LLD and Mold are two popular linkers on Linux systems with the latter being known for its impressive speed and much faster than GNU LD and Gold, this past week Qualcomm open-sourced a new linker: ELD.

ELD is a new linker open-sourced by Qualcomm focused on the needs of embedded developers. ELD is derived from mclinker and tailored for ELF files and the needs of embedded systems/developers. ELD can pair nicely with LLVM-based toolchains.

Qualcomm's ELD only targets AArch32,

Open Source LLM

OpenAI宣佈4月底停用GPT-4模型 全面升級至GPT-4o

OpenAI宣佈,GPT-4退役後將完全由新型原生多模態模型GPT-4o取代。目前官網頁面中GPT-4已新增「即將於4月30日下線」的提示信息,但仍可在API中使用。

根據官方評估報告,GPT-4o在寫作質量、編程輔助、STEM領域全面超


Open Source LLM

Chinese AI firm SenseTime bets on multimodal models to stand out from rivals

SenseTime CEO Xu Li speaks at the launch event of the SenseNova V6 models in Shanghai on Thursday. Photo: Handout

SenseTime , an artificial intelligence (AI) pioneer in China, has launched new models that it claims surpass OpenAI products in reasoning capabilities, as it bets on multimodal models to secure its position in the competitive AI landscape.
The company on Thursday unveiled SenseNova V6 and V6 Reasoner, new iterations of its self-developed AI model series. V6 outperformed OpenAI’s GPT-4o across several metrics, including fact-checking,
Continue reading...

Open Source LLM

Nvidia公佈Llama-3.1 Nemotron Ultra 效能超越DeepSeek R1 但僅一半參數量

Nvidia本週 開源Llama-3.1-Nemotron-Ultra-253B-v1模型 ,宣稱效能和DeepSeek R1相當,但僅一半的參數量。

Llama-3.1-Nemotron-Ultra是Nvidia 三月GTC大會上宣佈的 Llama Nemotron系列。Nano可在PC和邊緣裝置執行、Super可在單一GPU上執行,而Ultra模型則適用於


Open Source LLM

Craig Federighi’s leadership has already resulted in this major Siri pivot, per report

Today a revealing new look into Apple’s recent Siri struggles was published . That report contains myriad details on internal drama and conflicts, but it also ends with a big piece of news: under Craig Federighi’s leadership, for the first time Apple engineers can now use third-party LLMs to build Siri features.

:

Federighi has already shaken things up. In a departure from previous policy, he has instructed Siri’s machine-learning engineers to do whatever it takes to build the best AI features, even


Open Source LLM

Llama 4問世反應平平,開發人員認效能表現言過其實

Meta本週公佈旗艦模型Llama 4家族,不過一些開發人員試用後,發現標竿測試結果經過操弄。也有人覺得Llama 4模型效能並不如Meta説的那麼高。

Llama 4是Meta第一個以混合專家(mixture of experts,MoE)架構訓練的模型家族 。Meta並已開源二個Llama 4模型


Open Source LLM

GMKtec EVO-X2 Powered By Ryzen AI Max+ 395 To Launch For $2,052: The First AI+ Mini PC With 70B LLM Support

The EVO-X2 will be the first AI Mini PC with 70B LLM support to perform operations locally and will be significantly faster than RTX 4090 in AI performance. GMKtec EVO-X2 to Launch Next Week for Pre-orders, Starting at 14,999 Yuan; Features AMD Ryzen AI Max+ 395 The AI PC boom has compelled many PC hardware makers to produce dedicated AI PCs and GMKtech has come up with its own mini AI PC that it calls the

Open Source LLM

Mozilla Builders' LocalScore: An Interesting Local AI LLM Benchmark

Via Mozilla's Mozilla Builders initiative for fostering open-source AI projects is LocalScore, an interesting local AI large language model (LLM) benchmark for Windows and Linux systems. LocalScore has a lot of potential and also builds off the Mozilla Ocho Llamafile project as an easy-to-distribute LLM framework . LocalScore is still in its early stages but is already working well and will also be used in future hardware reviews on Phoronix.

Llamafile 0.9.2 was released this past week and brings with it LocalScore into
Continue reading...

Open Source LLM

Meta's Llama 4 Can Process 10 Million Tokens as Input, Lives in Native Multimodality

Meta has prepared a leap-forward update for its Llama model series with the v4 release, entering an era of native multimodality within the company's AI models. At the forefront is Llama 4 Scout, a model boasting 17 billion active parameters distributed across 16 experts in a mixture-of-experts (MoE) configuration. With FP4 precision, this model is engineered to run entirely on a single NVIDIA H100 GPU. Scout now supports an industry-leading input context window of up
Continue reading...

Open Source LLM

Meta公佈MoE架構開發的Llama 4 開源4000億、1090億參數的Maverick

回應新崛起的DeepSeek, Meta上週末公佈第一個混合專家(mixture of experts,MoE)模型家族Llama 4 ,並同時開源4000億參數的Maverick及1090億參數的Scout,此外預覽高達2兆參數量的Behemoth。

Llama 4是Meta第一個以混合專家(mixture of experts,MoE)架構訓練的模型

Menu