CCTest · Blog
DiffusionGemma: Discrete Diffusion Takes Aim at LLM Decoding Bottlenecks
Diffusion Models
cctest.ai
Diffusion Models

DiffusionGemma: Discrete Diffusion Takes Aim at LLM Decoding Bottlenecks

DiffusionGemma is an experimental open-weight language model that uses discrete diffusion to generate text in parallel blocks rather than token by token. Built by fine-tuning Gemma 4, it explores whether diffusion-style decoding can deliver a better speed-capability trade-off for LLMs.

Read more
CCTest · Blog
DAPD Reframes Policy Distillation Around Information Alignment
Reinforcement Learning
cctest.ai

DAPD Reframes Policy Distillation Around Information Alignment

DAPD argues that a key failure mode in on-policy self-distillation comes from information asymmetry: the teacher can rely on privileged signals that the student will not have at inference time. Its dual-anchored design aims to prevent this “privilege illusion” from being distilled into the final policy.

Read more
CCTest · Blog
GradCuit Optimizes LLM Reasoning Inside the Transformer at Test Time
Large Language Models
cctest.ai

GradCuit Optimizes LLM Reasoning Inside the Transformer at Test Time

GradCuit introduces a test-time latent optimization method that inserts trainable continuous states into an intermediate Transformer layer. By routing reward-weighted gradients from the generated continuation back to those states, it aims to make latent reasoning more robust and interpretable.

Read more
CCTest · Blog
DEFT-RLVR tackles future-trajectory leakage in autonomous driving VLM reasoning
Robotics & Physical AI
cctest.ai

DEFT-RLVR tackles future-trajectory leakage in autonomous driving VLM reasoning

The paper argues that many autonomous-driving VLM training pipelines let teacher models see the ground-truth future trajectory too early, encouraging post-hoc rationalization rather than causal reasoning. AD-MCQ and DEFT-RLVR recast planning as verifiable trajectory selection.

Read more