🏷️ Fine-Tuning

LLM NVIDIA and the 4B BF16 Model: The Specific Weight of Local Inference 2026-10-17

Hardware ORL pushes diffusion LLMs to 2,788 tokens/s on a single RTX 5090 2026-10-16

Hardware Bolt Graphics seeks $100 million to challenge Nvidia and AMD with the Zeus GPU 2026-10-16

Frameworks PRISM: privileged information for multi-agent topology design 2026-10-16

Frameworks DeepUnfolding Agent: Adaptive Control Becomes an External Service 2026-10-16

LLM Meta Opens Weights for Muse Spark Open: 816B Parameters and a 1M Token Context 2026-10-15

LLM Nemotron 3.5 Bolt: A 74B VLM Compressed to 24GB for On-Premise 2026-10-15

LLM Mr.Chatterbox-2.7B-Chat: an LLM trained from scratch on Victorian texts 2026-10-15

LLM Bengali shatters English LLM benchmark certainties 2026-10-15

OnPremise Following the Flow the Law Requires: Voice Agents and Regulated Debt Collection 2026-10-15

LLM Inverse Models Turn Answer-Only Data Into Reasoning Distillation 2026-10-15

LLM Lotus: forecasting as a pretraining principle for time-series LLMs 2026-10-15

LLM China's open-weight models trail US closed models by 7 months, and data transparency matters more 2026-10-14

Altro Apple researchers turn idle Macs into a virtual GPU cloud with ‘Minions’ 2026-10-14

LLM 51 offline labs to learn AI from neurons to agents 2026-10-14

Hardware Reverse engineering CUDA: the ironic challenge from an AMD user 2026-10-13

Hardware Intel's Crescent Island Becomes D-Series with Up to 480 GB LPDDR5X 2026-10-12

Hardware From Crescent Island to D Series: Intel Targets Enterprise Inference 2026-10-12

Hardware Intel D-Series D270 Targets Inference with FP8 and 1.5 TB/s 2026-10-12

LLM MiniCPM-V 4.7: the first open-source MoE with smooth on-device inference 2026-10-12

Hardware A 'broken' RTX 4090 for $350 hides a fake GPU and swapped memory 2026-10-12

Altro Strata: Building from Source and the Missing Documentation Problem 2026-10-12

Altro On-device receipt scanning with a 0.8B vision model 2026-10-12

LLM AutoIndexer: When Models Rewrite Their Own Output Like in a Text Editor 2026-10-12

Hardware A GPU in the 'wrong' place? Hotspot 19 degrees cooler 2026-10-11

Hardware Counterfeit RTX 4090 cards hit the market: the hidden risk for on-premise LLM inference 2026-10-11

Hardware Nvidia reportedly halts GeForce RTX 5090 production to prioritize AI data center GPUs 2026-10-10

Altro KNOD advances in-kernel GPU offloading for packet processing on AMD GPUs 2026-10-10

LLM Vega, an open-source 800M parameter decision model with 73k token context 2026-10-10

LLM Open-source GLM 5.3 Flash tops Artificial Analysis Cyber Index over Claude 2026-10-09

Altro Diffu-LoRA: Fine-Tuning That Allocates Parameters Where Needed 2026-10-09

LLM Freezing the Decoder, Healing the Encoder: The Shared Learning Rate Trap 2026-10-09

Frameworks Synthesis Through Simulation: SAP's framework for schema-free enterprise data 2026-10-09

Frameworks Running decision models on an RTX 4090: the runtime, not just size, wins 2026-10-08

Altro COSMIC Epoch 1.10 brings GPU-accelerated rendering to the launcher 2026-10-08

LLM LittleBit and the race toward ultra-low-bit quantization for tiny models 2026-10-08

LLM Hirundo removes CCP alignment from Qwen, cutting censorship to 2.8% 2026-10-08

LLM Child ASR adaptation: tuning without forgetting adults 2026-10-08

Altro llama.cpp on stage: self-hosted inference shifts phase 2026-10-07

Frameworks Agent Lightning 1.0: a 3,500-line RL framework for training agents with real harnesses 2026-10-07

LLM When External Guidance Helps LLM Reasoning: The Bias-Variance Theory of GA-GRPO 2026-10-07

Altro TEMPEST: Angular Embeddings for Scalable Driver Identification Without Retraining 2026-10-07

Altro Ubuntu 27.04 to default to ntpd-rs: Canonical confirms Rust time daemon 2026-10-06

Altro Outlook blocks .msix packages: Microsoft tightens email attachment security 2026-10-06

Hardware Vulkan and RADV: Valve Extends the Life of AMD Polaris GPUs on Linux 2026-10-06

Altro IRIS and Bpifrance back Vates with €30M for global expansion 2026-10-06

Frameworks ROCm 10.1: AMD's six-week release cycle is the real story 2026-10-05

Frameworks PyTorch consolidates media I/O in TorchCodec and refocuses vision/audio on transforms 2026-10-05

LLM Agens Volundr 32B: hybrid architecture keeps KV cache on 18 of 72 layers 2026-10-05

Frameworks Raman and Machine Learning: A Hybrid Framework Identifies Six Pharmaceuticals 2026-10-05

LLM AliceAI-80B-A3B: local fine-tuning stumbles on data, not on V100s 2026-10-04

LLM Apex-2: a 3.87B MoE trained from scratch on 86.5 billion tokens 2026-10-04

Hardware A modder fixes melting RTX 5090 power connectors with a custom distributor 2026-10-04

Hardware AMD Adds New GPU IP Enablement and Bug Fixes for Linux 7.4 2026-10-04

LLM Aleph Alpha brings Kolibri-1: 78B total, 3.46B active, 1M token context 2026-10-03

Hardware Biren prepares BR20X and sharpens China's AI GPU challenge to Nvidia 2026-10-03

Hardware RTX 5090: Micro Center introduces no-export declaration 2026-10-02

Frameworks Helion in vLLM's linear backend: one GEMM, per-shape tuning, up to 10% throughput gains 2026-10-02

Frameworks PyTorch Certification Turns Into a Structured Operational Pathway 2026-10-02

LLM Qwen3.8-27B-Humanlike-Chat 2.0: Human Tone Now Meets Tool Calls and Instruction Following 2026-10-02