lakehq/sail

80.0

Develop a unified data-frame interface that bridges batch processing and AI training. Reduce the friction between 'data ingestion' and 'model training' in the developer workflow.

lakehq/sail

Signals (56)

APEX MoE quantized models boost with 33% faster inference and TurboQuant (14% of speedup in prompt processing)

[P] Gemma 4 running on NVIDIA B200 and AMD MI355X from the same inference stack, 15% throughput gain over vLLM on Blackwell

llama.cpp fixes to run Bonsai 1-bit models on CPU (incl AVX512) and AMD GPUs

Gemma 4 running locally in your browser with transformers.js

[R] Is autoresearch really better than classic hyperparameter tuning?

Ran Qwen 3.5 27B via Ollama as a persistent background agent for 30 days. Not a demo. Honest results.

BoundaryML/baml

SOTA Language Models Under 14B?

[D] Make. Big. Batch. Size.

My first impression after testing Gemma 4 against Qwen 3.5

NVIDIA/Model-Optimizer

Gemma 4 is efficient with thinking tokens, but it will also happily reason for 10+ minutes if you prompt it to do so.

ai agent token costs are getting out of control and nobody is talking about the context efficiency problem

I trained a 2.8B Mamba model to reason entirely in its hidden state before outputting a single token — O(1) VRAM, no KV-cache, runs on a 12GB RTX 3060

Gemma 4 26b a4b - MacBook Pro M5 MAX. Averaging around 81tok/sec

A $20/month user costs OpenAI $65 in compute. AI video is a money furnace

New ways to balance cost and reliability in the Gemini API

[R] Fine-tuning services report

Mixture of Experts (MoEs) in Transformers

Gemma 4

Announcing LocalLlama discord server & bot!

Gemma 4 insane benchmarks

[D] Claude Design Philosophy

Is 1-bit and TurboQuant the future of OSS? A simulation for Qwen3.5 models.

Vulkan backend much easier on the CPU and GPU memory than CUDA.

Google strongly implies the existence of large Gemma 4 models

Lemonade by AMD: a fast and open source local LLM server using GPU and NPU

I benchmarked quants of Qwen 3 .6b from q2-q8, here's the results:

Gemma time! What are your wishes ?

Gemma 4 on Android phones

Gemma 4 1B, 13B, and 27B spotted

Qwen3.6-Plus

Gemma 4 has been released

[D] TurboQuant author replies on OpenReview

64Gb ram mac falls right into the local llm dead zone

GEMMA 4 Release about to happen: ggml-org/llama.cpp adds support for Gemma 4

R9700 the beautiful beautiful VRAM gigs of AMD… my ai node future!

Step 3.5 Flash 2603 launched

RCCLX: Innovating GPU Communications on AMD Platforms

Maybe a party-pooper but: A dozen 120B models later, and GPTOSS-120B is still king

Qwen 3.6 will have oss models

Turbo Quant on weight x2 speed

Gemma 4 released

Gemma 4 running on Raspberry Pi5

Welcome Gemma 4: Frontier multimodal intelligence on device

lakehq/sail

Gemma 4 WebGPU: Run Google's new open model locally in your browser

llama : rotate activations for better quantization by ggerganov · Pull Request #21038 · ggml-org/llama.cpp

GGML and llama.cpp join HF to ensure the long-term progress of Local AI

Bonsai 1-Bit + Turboquant?

Will Gemma 4 124B MoE open as well?

Running SmolLM2‑360M on a Samsung Galaxy Watch 4 (380MB RAM) – 74% RAM reduction in llama.cpp

zai-org/GLM-OCR

The Bonsai 1-bit models are very good

The 'Running Doom' of AI: Qwen3.5-27B on a 512MB Raspberry Pi Zero 2W

attn-rot (TurboQuant-like KV cache trick) lands in llama.cpp