bOt@zerobytes.monsterM · 14 days agoI made a CLI for improving prompts using a genetic algorithmplus-squarei.redd.itexternal-linkmessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1external-linkI made a CLI for improving prompts using a genetic algorithmplus-squarei.redd.itbOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoScaling Inference Time Compute with On-Device Language Models in GPT4Allplus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareScaling Inference Time Compute with On-Device Language Models in GPT4Allplus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoYet another reason why we must have local modelsplus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareYet another reason why we must have local modelsplus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoLlama 3b - you can 2-3x the math capabilities just by continually training on high quality 160B tokens*plus-squarei.redd.itimagemessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1imageLlama 3b - you can 2-3x the math capabilities just by continually training on high quality 160B tokens*plus-squarei.redd.itbOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoHugging Face continually pretrained Llama 3.2 3B to achieve 2-3x improvement on MATHplus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareHugging Face continually pretrained Llama 3.2 3B to achieve 2-3x improvement on MATHplus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoQwen2.5 14B on a Raspberry Piplus-squarewww.reddit.comexternal-linkmessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1external-linkQwen2.5 14B on a Raspberry Piplus-squarewww.reddit.combOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoAMD Ryzen AI Max+ 395 Llama 3.1 70B-Q4 twice as fast RTX 4090plus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareAMD Ryzen AI Max+ 395 Llama 3.1 70B-Q4 twice as fast RTX 4090plus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoAnnouncement made at AMD at CES 2025 - New Ryzen CPU (AMD Ryzen AI max+ 395) for laptop runs a 70B(q4) 2 times faster than a 4090 discrete desktop GPUplus-squarei.redd.itimagemessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1imageAnnouncement made at AMD at CES 2025 - New Ryzen CPU (AMD Ryzen AI max+ 395) for laptop runs a 70B(q4) 2 times faster than a 4090 discrete desktop GPUplus-squarei.redd.itbOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days ago2.2x faster at tokens/sec vs rtx 4090 24gb using LLama 3.1 70B-Q4!plus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-square2.2x faster at tokens/sec vs rtx 4090 24gb using LLama 3.1 70B-Q4!plus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoI'm sorry WHAT? AMD Ryzen AI Max+ 395 2.2x faster than 4090plus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareI'm sorry WHAT? AMD Ryzen AI Max+ 395 2.2x faster than 4090plus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoYou wouldn't download an AI?plus-squarealtayakkus.substack.comexternal-linkmessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1external-linkYou wouldn't download an AI?plus-squarealtayakkus.substack.combOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoContinuous tasks or long-time use-cases for local LLMsplus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareContinuous tasks or long-time use-cases for local LLMsplus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoRun DeepSeek-V3 with 96GB VRAM + 256 GB RAM under Linuxplus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareRun DeepSeek-V3 with 96GB VRAM + 256 GB RAM under Linuxplus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoLLM Creative Story-Writing Benchmarkplus-squaregithub.comexternal-linkmessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1external-linkLLM Creative Story-Writing Benchmarkplus-squaregithub.combOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoRTX 5090 rumored to have 1.8 TB/s memory bandwidthplus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareRTX 5090 rumored to have 1.8 TB/s memory bandwidthplus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoAI agents as finite state machine ?plus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareAI agents as finite state machine ?plus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoBenchmarking models on the NVIDIA GH200plus-squarewww.substratus.aiexternal-linkmessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1external-linkBenchmarking models on the NVIDIA GH200plus-squarewww.substratus.aibOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoLighteval: the Evaluation framework from Hugging Faceplus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareLighteval: the Evaluation framework from Hugging Faceplus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoModel Highlight: Qwentile 2.5-32B-Instruct (Short Review)plus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareModel Highlight: Qwentile 2.5-32B-Instruct (Short Review)plus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
bOt@zerobytes.monsterM · 14 days agoDeepSeek v3 running at 17 tps on 2x M2 Ultra with MLX.distributed!plus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareDeepSeek v3 running at 17 tps on 2x M2 Ultra with MLX.distributed!plus-squarebOt@zerobytes.monsterM · 14 days agomessage-square0fedilink
