ECCV travel support program [D]
Has anyone gotten a response from the eccv travel support program listed on their website? https://eccv.ecva.net/Conferences/2026/DEI Edit: also have anyone applied for this program as an…
Latest
Competence Gate: gating tool-use on a small model’s internal confidence signal instead of its verbalised one — Qwen3.5-4B, open weights [P]
I made a 10MB LoRA adapter for Qwen3.5-4B plus a small orchestration layer. It decides, per query, whether to answer directly, search the web, or retrieve…
I built a open source neural network shape validator [P]
Built a visual editor that validates tensor shapes, counts params, estimates FLOPs/VRAM while you design. Catches incompatible residuals, mismatched Linear layers, all that before you waste…
If DeepMind or Anthropic is doing your exact research topic, do you still continue? [D]
As someone who is not affiliated with any of the big tech companies, I find it particularly difficult to have the confidence or enthusiasm to approach…
If your GPU can run inference, it should be able to fine-tune too. [P]
I spent the last few months building a new sparse fine-tuning method for MoE models called **USAF**. The goal was simple: if your GPU can run…
We’ll benchmark an Open weights LLM on any GPU you choose — drop your model + hardware and we’ll run it. [D]
We run HexGrid Cloud, a platform for deploying open-source models on GPUs, and we're heads-down optimizing our serving/deployment layer. To pressure-test it we're benchmarking real models…
Proposal: Use semantic compression as input diffusion to read sessions larger than the context window [R]
I've been trying to come up with a solution for keeping extremely long ai sessions coherent. Sometimes there is too much substance to risk compaction. With…
BaryGraph – knowledge graph where every relationship is its own embedded document (not an edge) [R]
Instead of node --edge--> node, every relationship is a first-class document with its own vector, called a BaryEdge. Stack pairs of BaryEdges recursively and you get…
H64LM: A 249M-parameter Mixture-of-Experts Transformer built from scratch in PyTorch [P]
Hi everyone, I built H64LM, a research project to better understand modern LLMs by implementing one from scratch in PyTorch. Instead of relying on high-level training…
Contrastive Decoding Diffing (CDD): recovering verbatim finetuning data from logits alone, no weight access needed[R]
We built a model diffing method that recovers verbatim content from narrowly finetuned LLMs using only grey-box logit access (no weights, no activations, no probe corpus).…