Gemma

Timeline

1. Oct 18, 2026 · Opinion / analysis · 1 source
   Embedding model can be used to search text inside image
   Just something I've been testing over few week since Embedding Gemma 2 released which is about 330MB at Q8.
2. Oct 17, 2026 · Opinion / analysis · 1 source
   llama.cpp has come a long way
   I started out back in March 2025 running Mistral Nemo Q4KM with 8K F16 on an AMD Radeon RX 7800 XT 16GB, with first ollama and then lmstudio.
3. Oct 16, 2026 · Research paper · 1 source
   LocalIntent: Outcome-Specific Structured Generation for Backend-Compatible Smart Building Automation with Local LLMs
   Locally deployed large language models (LLMs) can reduce reliance on remote cloud services for natural language smart building automation, but backend-facing processing requires correct outcome selection, alignment with local capabilities, and valid structured outputs.
4. Oct 16, 2026 · Open-source release · 1 source
   axolotl-ai-cloud/axolotl v0.21.0
   This release is headlined by Reflection's Beam, a 501B-parameter MoE you can QLoRA-tune on a single node.
5. Oct 16, 2026 · Research paper · 1 source
   HydraLLM: Adaptive Width Slicing for Elastic Text and Vision-Language Models
   We introduce HydraLLM, an elastic MQA-based architecture.
6. Oct 15, 2026 · Research paper · 1 source
   Pearl-Chat: Engineering a native Luganda language model from first principles in JAX and Flax NNX
   We present Pearl-Chat, an 8,473,122-parameter ($\sim$8.5M) causal decoder-only transformer engineered for Luganda (Oluganda, ISO 639-2: lg), a low-resource Niger-Congo Bantu language spoken by over 10 million people in East Africa.
7. Oct 15, 2026 · Opinion / analysis · 1 source
   What a £300, 35W Snapdragon mini PC can actually do 👀: 35B coder at ~30 tok/s, and an 80B model that doesnt even fit in RAM running at 17 tok/s
   £299.99 Lenovo IdeaCentre Mini (Snapdragon X1-26-100, 32GB, CPU only):
8. Oct 15, 2026 · Model release · 12 sources
   Gemini 4 Argon: Built to solve your hardest problems
   Gemini 4 Argon: Built to solve your hardest problems
9. Oct 15, 2026 · Funding / M&A · 1 source
   [AINews] Modal’s $1B ARR, $16B Series D makes it NYC’s newest AI Infra Decacorn, 3x usage and 3x valuation in 5 months, 15x in 1 yr
   Today Modal switches from our infra unicorn category to infra decacorn.
10. Oct 15, 2026 · Opinion / analysis · 1 source
    I tested Qwen3.8 27B KLD for 21 Models at 9 different KV quants that fit on 16GB VRAM (189 benchmarks).
    I calculated KLD and perplexity for Qwen3.8 27B models that fit 16GB VRAM:
11. Oct 15, 2026 · Research paper · 1 source
    Calibration Language Shapes Block-Influence Pruning in Multilingual Language Models
    We study whether English calibration transfers to Indian languages in Qwen2.5-1.5B-Instruct, Llama-3.2-1B, and Gemma-2-2B.
12. Oct 15, 2026 · Opinion / analysis · 1 source
    4x R9700 builds with $127 used EPYCs: $10.7k all-in per machine. I built three (384GB VRAM)
    I saw the 4x R9700 build for a 10-person startup here a few days ago and wanted to share my results, since I opted to spend most of the money on the cards.
13. Oct 15, 2026 · Research paper · 1 source
    Embeddings Are Overpriced? A Frozen 16 MB Vocabulary for Small and Ternary Language Models
    In Gemma 3 270M the embedding table holds 168M of 268M parameters, and because quantisation methods leave the embedding in 16-bit, it is also most of the file once the rest of the model is compressed.
14. Oct 14, 2026 · Research paper · 1 source
    INFUSE: A Unified Surface Benchmark for Indirect Prompt Injection in Agentic Systems
    We introduce INFUSE, an end-to-end benchmark for evaluating IPI throughout the lifecycle of long-horizon agent workflows.
15. Oct 14, 2026 · Research paper · 2 sources
    Sparse Autoencoders Are a Shared Coordinate System for Symbolic Matching
    Our method, SAE-IDF, encodes a symbol in its native context into a sparse code with a sparse autoencoder (SAE), re-weights each dimension by its rarity and matches across systems.
16. Oct 14, 2026 · Research paper · 1 source
    Are You Reading the Right Thing? Rethinking Internal Signals for Prompt-Injection Detection
    Several recent studies propose detecting prompt injections from a language model's internal state.
17. Oct 14, 2026 · Product / feature launch · 1 source
    Google DeepMind and Singapore Government announce new AI tool to protect citizens against scams
    Stage 1: A lightweight, efficient classifier screens inbound messages, using minimal power, focused on catching potential threats.
18. Oct 14, 2026 · Research paper · 1 source
    Diminishing Returns of Data in Sparse Autoencoders: Functional Behavior does not Guarantee Feature Reproducibility
    Sparse autoencoders (SAEs) are typically trained on hundreds of millions to billions of activation tokens, yet little work has tested whether this scale is necessary.
19. Oct 13, 2026 · Benchmark result · 1 source
    SuperLocalMemory
    Agent memory with Laya, Jev + Google Gemma
20. Oct 13, 2026 · Research paper · 1 source
    Cost-Efficient Privacy Enforcement for Email-Generating LLM Agents under Strategic Information Requests
    We propose local contract-conditioned rewriting as a cost-efficient enforcement layer for email-generating agents.
21. Oct 13, 2026 · Research paper · 2 sources
    Training LLMs for Multi-Agent Collaboration with Verifiable Information Rewards
    Effective collaboration among large language model (LLM) agents in multi-agent systems requires sharing relevant evidence, requesting missing information, and incorporating teammates' contributions into decisions.
22. Oct 12, 2026 · Research paper · 1 source
    Soft Teacher Forcing
    We introduce Soft Teacher Forcing (Soft-TF), a simple training algorithm that combines teacher-forced embeddings with soft tokens through random convex combinations.
23. Oct 12, 2026 · Opinion / analysis · 1 source
    U-Space: Uncovering When and Why Uncertainty Arises in Language Models [R][D]
    If you build on open-weight reasoning models, you have probably wanted a "how sure is the model about this answer?" signal that does not cost 5x sampling, a judge model, or a labeled dataset.
24. Oct 12, 2026 · Research paper · 1 source
    Separating Prompt Harmfulness from Interaction Safety via Partial Generation
    LLM-based information systems need safety guardrails.
25. Oct 12, 2026 · Research paper · 2 sources
    HexaBee-4B: Six-Language Translation with Multilingual Answering and Bidirectional Supervision
    We present HexaBee-4B, a translator adapted from Gemma 3 4B IT through three stages: Wikipedia continued pretraining, joint multilingual answering and answer-text translation, and bidirectional translation specialization.
26. Oct 12, 2026 · Opinion / analysis · 1 source
    Compressed a RoBERTa AI-text detector from ~500 MB to ~97 MB so it runs client-side in the browser, plus a 26 MB image detector [P]
    I wanted an AI text and image detector that doesn't need you to upload anything, so I took two existing open detectors and got them running fully client-side with ONNX Runtime Web.
27. Oct 12, 2026 · Opinion / analysis · 1 source
    QLoRA on ExLlamaV3 Quants - Train @ 2-8bpw
    A fork of ExLlamaV3, this is a native ExLlamaV3 training implementation with plain PyTorch, without transformers.
28. Oct 12, 2026 · Research paper · 1 source
    PAIR: Memory-Efficient and Fast Inference with Reading--Reasoning Fusion
    For context-based question answering, using one large model for both evidence extraction and answer reasoning requires substantial memory and computation.
29. Oct 11, 2026 · Opinion / analysis · 1 source
    Taming the AI Hardware Cambrian Explosion — Abdul Dakkak, Modular
    Abdul Dakkak, chief scientist at Modular, explains why today's AI software stack is a mess, with vLLM, SGLang, TensorRT-LLM and llama.cpp each patching around different hardware, and how Modular rebuilt the stack from the ground up.
30. Oct 11, 2026 · Research paper · 1 source
    Lightweight Scheduling Policies for Continuously Batched LLLM Serving Across GPU Architectures
    Large language model serving is increasingly limited not only by model size and GPU speed, but also by scheduling during autoregressive decoding, where continuous batching interleaves requests at token granularity as sequences arrive and finish.

Often appears with