Measuring the Serving Stack Instead of the Model: Hidden Confounds in Local Tool-Use Evaluation
Local serving stacks can distort tool-use scores, including false 0% fidelity for rejected models.
43
AI model
Gemma-3
- Mentions · 7d0
- Mentions · 30d0
- All mentions5
- First seenSep 9, 2026
- Last mentionedSep 22, 2026
Timeline
5 articles · newest firstA Zeroth-Order Paradigm for LLM Preference Alignment
Researchers propose ComPO, a zeroth-order comparison-based preference alignment method with convergence guarantees that mitigates likelihood displacement in LLMs.
32
A Zeroth-Order Paradigm for LLM Preference Alignment
ComPO is a zeroth-order preference alignment method using comparison oracles to mitigate likelihood displacement across Mistral, Llama, Gemma, and Qwen3 models.
38
Google Research Releases ToolGrad: Answer-First Framework Hits 99.8% Pass Rate for Tool-Use Data Generation
Google Research and partners introduce ToolGrad, a verified tool-chain-first data generation framework reaching 99.8% pass rate and boosting Gemma-3-12B to 83.1 on BFCL.
58
Active Adaptation, Not Static Defense: Temporal Dynamics of Preventative Steering in Adversarial Fine-Tuning
Researchers analyze why Preventative Steering protects LLMs against malicious fine-tuning, finding active adaptation drives protection, and propose Progressive Intensity Scheduling.
32
Appears with
- Qwen3ai model · 2
- Gemma-2ai model · 2
- Mistralai model · 2
- Llamaai model · 2
- llama.cppproduct · 1
- Ollamaproduct · 1
- ToolBenchproduct · 1
- Claude 4.5 Opusai model · 1
- GPT-5ai model · 1
- Gemini 2.5 Proai model · 1
- RIKEN AIPorganization · 1
- ComPOproduct · 1
- University of Tokyoorganization · 1
- Llama-3.2ai model · 1
- Qwen2.5ai model · 1
- Gemini 2.5 Flash-Liteai model · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.