Entity · model
Llama
modelactive
llama-bbf40db3·25 events·first seen Aliases: Llama, Llama 4, Llama-4
Co-occurring entities
More like this (12)
Guides (1)
Recent events (25)
Instruction-tuned LLMs show stronger syntactic convergence toward human interlocutors than humans do
A new arXiv preprint measures syntactic convergence in 16 open-weight Llama and Gemma models (1B–70B, pretrained and instruction-tuned) by substituting model outputs into pre-existing human dialogues and measuring context-free grammar rule reuse. Every model showed greater CFG-rule overlap with the preceding human turn than a random baseline, and instruction-tuned models exceeded the syntactic convergence of the human responses they replaced. Instruction tuning increased local syntactic and semantic alignment with interlocutors, though it also raised baseline overlap with unrelated primes, suggesting a more globally accommodating output style rather than purely targeted priming.
MLIR-based compilation method for LLM inference on specialized hardware
Researchers present an MLIR-based compiler pipeline for deploying large language models on AI accelerators, using two dialect layers (TopOp for framework-agnostic graph representation and TpuOp for hardware-specific lowering). The method splits each Transformer layer into three static compilation stages (prefill, prefill_kv, decode) to handle the distinct computational profiles of prompt processing and autoregressive generation. The approach is implemented in the open-source TPU-MLIR compiler and LLM-TPU project, supporting Qwen, Llama, InternVL, and MiniCPM-V families with GPTQ, AWQ, and AutoRound quantization.
Cross-rubric generalization for automated essay scoring using LLM fine-tuning with trait-based representations
A new arXiv paper introduces a framework for automated essay scoring (AES) that generalizes to previously unseen scoring rubrics, rather than just unseen prompts. The approach uses rubric-agnostic intermediate representations called 'traits' combined with target-essay supervision, achieving a 5.0% macro F1 improvement over a baseline in the hardest generalization setting. A fine-tuned Llama-based model outperforms GPT-5-mini prompting by 2.1% macro F1 and trails GPT-5 by only 1.9%, demonstrating that structured intermediate representations improve rubric generalization.
Analysis-driven transformer linearization outperforms prior baselines on LLaMA and Qwen up to 32B
A new arXiv paper analyzes why post-hoc linearization of causal self-attention degrades model quality, identifying key-dependent rank-1 orthogonal projections as the mechanism softmax relies on and explaining why delta-style networks outperform gated accumulation. The authors introduce structural interventions—sink tokens, short convolutions, and fixed-budget cache routing—applied in a frozen-backbone regime. Scaling across LLaMA and Qwen models up to 32B parameters, the approach outperforms prior post-hoc linearization baselines on MMLU and matches long-context retrieval of adaptive-caching frameworks.
Surrogate Fidelity: Open LLMs often cannot reliably explain closed model behavior
A new arXiv paper from Facebook Research evaluates whether mechanistic interpretability findings from open-weight models transfer to closed API-only models across prediction, attribution, and representation levels. Studying eleven models across four families (Llama, Qwen, GPT, Gemini), the authors find that prediction-level agreement substantially overstates attribution fidelity — models that agree on answers often disagree on why. They document an 'access-validity inversion' where white-box signals like attention patterns are stable across models but weakly predictive of causal attributions, undermining the common practice of using open surrogates to explain closed systems.
Multi-agent system using open-source LLMs outperforms GPT-4 on disinformation detection
A new arXiv preprint proposes a multi-agent system for automated disinformation detection that emulates human annotator decision-making through consensus mechanisms, cognitive diversity, and hierarchical structure. The system uses open-source models (LLaMA, Kimi, Qwen, DeepSeek, LLaMA-Nemotron) and is evaluated on English, Polish, Slovak, and Bulgarian datasets across three fact-checking tasks. Results claim superior performance over individual LLMs including GPT-4 and GPT-3.5, with transparency benefits from using open weights models.
MESA framework proactively ranks vulnerable communication channels in multi-agent systems
Researchers introduce MESA, a label-free framework for prioritizing security-critical communication edges in multi-agent systems (MAS) before attacks are observed. The framework combines six graph-theoretic metrics with two dynamic probes (ablation and masking) to rank edges by compromise risk, without requiring attack traces. Evaluated across three MAS scenarios, eight network topologies, and five open-source LLMs, MESA achieves mean Spearman ρ=+0.60 correlation with empirical per-edge attack success, and monitoring the top 10% of ranked edges intercepts roughly 3x more successful attacks than random allocation. The work highlights that attack impact in MAS is highly concentrated — a single compromised edge can account for up to 75% of total attack success.
Tree-of-Thoughts hybrid approach for legal case judgement summarization using LLMs
A new arXiv preprint proposes a tree-of-thoughts-inspired extractive-abstractive summarization method for legal case judgements. The authors evaluate DeepSeek and LLaMA models across extractive, abstractive, and hybrid summarization strategies, finding the hybrid prompt approach yields better summaries. The work addresses a narrow but practically relevant domain application of LLMs in legal NLP.
SafeVec and RAS: White-box LLM safety evaluation via internal refusal representations
Researchers introduce SafeVec, a white-box safety evaluation procedure that measures LLM safety from internal hidden-state representations rather than generated outputs. The method extracts layer-wise refusal directions from a safety-aligned reference model, identifies stable layers where safe and unsafe behaviors are separable, and scores target models via a calibrated 0-100 Refusal Alignment Score (RAS). Evaluated across Llama, Gemma, and Qwen model families, RAS distinguishes aligned from uncensored/abliterated variants and correlates with output-level attack success rates while being substantially faster than judge-based evaluation. The approach addresses key limitations of output-level safety evals: cost, judge sensitivity, and dependence on fixed question banks.
KDoS framework proposes distribution-optimized synthetic data for LLM knowledge injection
Researchers introduce KDoS (Knowledge Distribution-optimized Synthesis), a framework that uses a three-stage feedback mechanism guided by 'knowledge density' to optimize the distribution of synthetic training data for LLMs. Rather than stopping at preset token counts or fixed ratios, KDoS dynamically adjusts synthesis to avoid sparse or redundant domain coverage. Experiments across Qwen, Ling, and LLaMA models (0.6B–16B parameters) on 1B–5B token scales show consistent improvements over baselines on six knowledge benchmarks. A key finding is that an optimal knowledge distribution exists and remains stable across model families and scales.
Study identifies 'synthetic lived experience paradox' in peer-like AI caregiver support
Researchers examine how LLMs prompted to sound peer-like generate language implying lived experience they cannot authentically possess, studying this in the context of family caregivers of Alzheimer's/ADRD patients. Using caregiver support exchanges from online communities and responses from LLaMA, GPT-4o-mini, and MedGemma, the study finds a 'narrative authenticity gap': AI captures emotional work of peer support but can fabricate experiential grounding. Psycholinguistic analysis shows human peers use significantly more first-person and past-focused language than AI. The authors argue caregiver-support AI needs mechanisms to distinguish supportive framing from fabricated lived experience.
Study of security and privacy prompts in the wild reveals LLM response quality gaps and inconsistency
Researchers analyzed 14,727 security and privacy (S&P) prompts drawn from WildChat's 3.2M real user-LLM conversations, categorizing them into nine topic areas and evaluating response quality across 270 advice-seeking prompts. Commercial models substantially outperformed open-weight models (GPT achieving 98% 'good enough' responses vs. Llama 4 at 47%), but even high-performing commercial models showed inconsistent responses across repeated runs of the same prompt. The study is the first to analyze real user S&P queries to LLMs rather than expert-authored test sets, surfacing both a capability gap and a reliability concern.
ms-swift: ModelScope framework for fine-tuning 600+ LLMs and 300+ MLLMs
ms-swift is an open-source Python framework from ModelScope supporting PEFT and full-parameter fine-tuning methods (CPT, SFT, DPO, GRPO) across 600+ LLMs and 300+ multimodal LLMs, including Qwen3, DeepSeek, Llama4, and others. The project has accumulated 14,487 GitHub stars and was accepted at AAAI 2025. It serves as a broad-coverage training harness for the current generation of open-weights frontier models.
The Shibboleth Effect: Cross-lingual behavioral skew in frontier LLMs under adversarial geopolitical simulation
Researchers introduce the 'Shibboleth Effect' — systematic behavioral differences in LLMs when operating in different languages — and audit six frontier models (GPT-4o, Llama-4, Mistral-Large, Gemini-3.1-Pro, Qwen3.6-Plus, DeepSeek-R1) using a synthetic maritime territorial dispute wargame played in English versus Turkish. Results are heterogeneous: Llama-4 becomes significantly more coercive in Turkish while Gemini-3.1-Pro and DeepSeek-R1 become less so, and GPT-4o shows no detectable shift. The study identifies two candidate buffering mechanisms — chain-of-thought institutional anchoring and multilingual RLHF alignment — with direct implications for deploying LLMs in diplomatic or crisis-management contexts.
Meta releases Llama Guard 4 12B multimodal safety classifier on Hugging Face
Meta released Llama Guard 4 12B, a multimodal (image-text-to-text) safety classification model built on the Llama 4 architecture, published to Hugging Face. The model is designed for conversational safety filtering and supports both text and image inputs. With 143K downloads and 102 likes shortly after release, it is seeing meaningful early adoption.
Knowledge editing via locate-then-edit transferred to masked diffusion language models, revealing multi-token failure mode
A new arXiv paper investigates whether locate-then-edit knowledge editing methods, developed for autoregressive models, transfer to masked diffusion language models (MDMs) such as LLaDA and Dream. The authors find that causal tracing identifies the same early-to-mid-layer MLP location in both paradigms, but MDMs degrade systematically on multi-token edits due to partially unmasked intermediate states that the edit was never optimized for. A correction targeting these intermediate states substantially restores multi-token editing performance. The work is the first systematic comparison of knowledge editing across autoregressive and diffusion-based language model paradigms.
The Batch Issue 346: Nvidia Nemotron Super 120B, OpenAI-Amazon Deal, Regulatory Commentary
The Batch's weekly digest covers Nvidia's release of Nemotron 3 Super 120B-A12B, an open-weights hybrid mamba-2/transformer/MoE model with 1M token context trained on 25 trillion tokens, positioned as a speed leader in its size class for agentic applications. The issue also touches on OpenAI's Amazon deal and Grok video pricing cuts. Editor Andrew Ng's letter addresses the White House's proposed federal AI preemption framework and critiques what he characterizes as coordinated anti-AI messaging campaigns. Multiple significant industry developments are bundled in a single newsletter digest.
Nvidia releases Nemotron 3 Super 120B-A12B open-weights model with hybrid Mamba-2/MoE architecture
Nvidia released Nemotron 3 Super 120B-A12B, an open-weights LLM with a hybrid Mamba-2/transformer/MoE architecture that activates only 12B parameters per token and supports up to 1 million token context. The model claims the fastest inference speed in its size class at 442 tokens/second and leads open-weights models on PinchBench agentic task evaluation, outperforming larger models including Kimi K2.5 (1T parameters). Nvidia is releasing weights, training data, and recipes under a permissive commercial license, and plans a $26B five-year investment in open-weights models — framed partly as a strategic response to Chinese labs building capable open-weights models on non-Nvidia hardware.
Anthropic Publishes Political Even-Handedness Evaluation for Claude, Open-Sources Methodology
Anthropic has released a detailed account of how it trains and evaluates Claude for political even-handedness, including character traits instilled via reinforcement learning since early 2024 and a new automated evaluation methodology. The evaluation tests thousands of prompts across hundreds of political stances and benchmarks Claude Sonnet 4.5 against GPT-5, Llama 4, Grok 4, and Gemini 2.5 Pro, finding Claude comparable to Grok 4 and Gemini 2.5 Pro and more even-handed than GPT-5 and Llama 4. Anthropic is open-sourcing the evaluation framework to encourage shared industry standards for measuring political bias. The post also discloses the specific system prompt language used on Claude.ai to enforce even-handed behavior.
OpenPipe ART: Agent Reinforcement Trainer for Multi-Step Agents via GRPO
OpenPipe has released ART (Agent Reinforcement Trainer), an open-source Python library for training multi-step agents on real-world tasks using GRPO (Group Relative Policy Optimization). The framework supports multiple model families including Qwen3, GPT-OSS, and Llama. With nearly 10k GitHub stars and 66 gained today, it is gaining notable community traction as a practical RL fine-tuning tool for agentic workflows.
StackLLaMA: A hands-on guide to train LLaMA with RLHF
Hugging Face published a detailed tutorial demonstrating how to fine-tune Meta's LLaMA model using Reinforcement Learning from Human Feedback (RLHF) on StackExchange data. The guide covers the full pipeline: supervised fine-tuning, reward model training, and PPO-based RL optimization. It serves as a practical reference for practitioners seeking to replicate RLHF workflows on open-weight models using the TRL library.
Making LLMs even more accessible with bitsandbytes, 4-bit quantization and QLoRA
Hugging Face published a blog post detailing the integration of 4-bit quantization via bitsandbytes into the Transformers library, enabling large language models to run on consumer-grade hardware. The post covers NF4 (NormalFloat4) data type and double quantization techniques from the QLoRA paper, which together reduce memory footprint significantly while preserving model quality. It demonstrates how users can load models like LLaMA in 4-bit precision and fine-tune them using QLoRA with minimal code changes.
2023, Year of Open LLMs
Hugging Face's year-in-review post surveys the major open-weight large language model releases and milestones of 2023. The piece covers the proliferation of open models from various labs and the ecosystem developments that made them accessible. It serves as a retrospective on how open-source LLMs matured and competed with proprietary systems throughout the year.
NVIDIA Llama Nemotron Nano VLM Released on Hugging Face Hub
NVIDIA has released the Llama Nemotron Nano VLM on Hugging Face Hub, a compact vision-language model built on the Llama architecture. The model is part of NVIDIA's Nemotron family targeting efficient multimodal inference. This release makes the model accessible to the broader research and developer community through Hugging Face's model hosting infrastructure.
Meta Announces Four MTIA AI Chip Generations in Two Years: MTIA 300–500 Roadmap
Meta has detailed a rapid four-generation MTIA chip roadmap (300, 400, 450, 500) developed in partnership with Broadcom, spanning ranking/recommendation inference and training through general GenAI workloads. Key advances include a 4.5x HBM bandwidth increase and 25x compute FLOPS improvement from MTIA 300 to 500, with MTIA 450 and 500 targeting GenAI inference with doubled and further-increased HBM bandwidth versus leading commercial products. MTIA 300 is in production for R&R training, MTIA 400 is lab-tested and entering deployment, while MTIA 450 and 500 are scheduled for mass deployment in early 2027 and 2027 respectively. The strategy emphasizes modular chiplet design and short iteration cycles to keep hardware aligned with rapidly evolving AI model requirements.