Auxiliary Rewards Improve Small-Language-Model Self-Improvement
UID-Conj. auxiliary rewards boost online adaptation and offline SFT by 2.08 and 1.93 points over baselines on AIME math tasks.
Stories per day · 30 days
3 RSS
Relatedagent-training1aime1auxiliary-rewards1bitnet1code-generation1distillation1embeddings1fine-tuning1gemma1grpo1inference-cost1language-model-agents1lfm31liquid-ai1malware1multimodal1open-source1open-weights1
13 stories, newest first · one card per story
Stories tagged #small-language-models
Liquid AI releases LFM3, six open-weight 2B models
The available text is only the headline; licenses, benchmarks, and release details are not included.
Fine-tuned small models predict x86-64 register inputs from outputs
GRPO fine-tuning lifts Nemotron-3-Nano knowable-register accuracy from 63.3% to 95.2%, within four points of frontier scores.
Specializing SLM vocabularies can cut inference cost by 77%
A Hugging Face post claims vocabulary specialization cut small-model inference cost by 77% in real applications; no further details were provided.
0.5B model detects prompt injection in local RAG pipelines
Authors report 0.994 in-distribution recall at a 0.026 false-positive rate, but near-zero recall on PoisonedRAG.
Progress-Aware Distillation targets stagnation in small agent models
Tests on eight models from 0.6B to 8B parameters show higher success rates and faster task completion.
Small models generate post-exploitation code in lab tests
Qwen2.5-Coder 1.5B reached Pass@5 of 100% on directory enumeration in Python and Lua.
Frozen 16 MB codes replace learned embeddings in small models
A Gemma-shaped ternary model fits in a 46 MB file and runs 615–740 tokens per second on a laptop CPU.
EdgeWise: Bringing Long-Horizon Agents to the Edge via Selective Cloud Delegation
EdgeWise delegates only key agent steps to the cloud, cutting inference cost by up to 62%.
Harness-Aware Distillation for Small Language Model Agents
Harness-Aware Distillation trains smaller agents on skills the fixed software harness cannot supply.
Keyword Harnesses Fail Open: A Cheap Diagnostic Ladder for Tool-Use Claims in Small Language Models
Keyword benchmarks falsely credit a 1.1B model with tool use; cheap diagnostics and targeted SFT expose and repair it.
Keyword Harnesses Fail Open: A Cheap Diagnostic Ladder for Tool-Use Claims in Small Language Models
Keyword tool-use scores falsely credit small Spanish security models; a cheap diagnostic ladder separates real tool calls.
openbmb/MiniCPM5-2B — new model trending #4 on Hugging Face
OpenBMB released MiniCPM5-2B, a dense 2B open-weights Transformer claiming 2B-class open-source SOTA for on-device deployment.