@GithubProjects
Account based inSouth Asia
About this account
- Account based in
- South Asia
- Connected via
- South Asia App Store
Account-level information from X, not a live location or the device used for a specific post.
We're sharing/showcasing best of @github projects/repos. Follow to stay in loop. Promoting Open-Source Contributions. UNOFFICIAL, but followed by github
San Francisco, CA
Joined October 2020
- Tweets13.6K
- Following66
- Followers342K
- Likes23.9K
New open-source embodied AI project: SenseNova-RoboRSI.
It includes a unified evaluation harness, a study covering 12 simulation benchmarks, Dockerized dependencies, a common HTTP API, research tooling, and SenseNova-RoboRSI-27B-Preview.
SenseTimeโs report concludes that GPT-6-Astra has crossed the threshold of ๐ฒ๐บ๐ฏ๐ผ๐ฑ๐ถ๐ฒ๐ฑ ๐ฐ๐ผ๐บ๐ฝ๐ฒ๐๐ฒ๐ป๐ฐ๐ฒ, but ๐ฝ๐ฟ๐ฎ๐ฐ๐๐ถ๐ฐ๐ฎ๐น ๐ฒ๐บ๐ฏ๐ผ๐ฑ๐ถ๐ฒ๐ฑ ๐ถ๐ป๐๐ฒ๐น๐น๐ถ๐ด๐ฒ๐ป๐ฐ๐ฒ is not here yet.
Report: arxiv.org/abs/2610.22471
We release the most comprehensive ๐ฒ๐บ๐ฝ๐ถ๐ฟ๐ถ๐ฐ๐ฎ๐น ๐๐๐๐ฑ๐ ๐ผ๐ณ ๐๐ฃ๐ง-๐ฒ-๐๐๐๐ฟ๐ฎโ๐ ๐ฒ๐บ๐ฏ๐ผ๐ฑ๐ถ๐ฒ๐ฑ ๐ถ๐ป๐๐ฒ๐น๐น๐ถ๐ด๐ฒ๐ป๐ฐ๐ฒ across 12 simulation benchmarks and 8 real-robot platforms, together with ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ-๐ฅ๐ผ๐ฏ๐ผ๐ฅ๐ฆ๐: an open-source harness framework and a 27B model that runs at ~1/50 the cost of GPT-6-Astra.
Here's what we found:
๐น ๐ง๐ต๐ฒ ๐ฐ๐ฎ๐ฝ๐ฎ๐ฏ๐ถ๐น๐ถ๐๐ ๐ฐ๐ฒ๐ถ๐น๐ถ๐ป๐ด ๐ถ๐ ๐ฎ๐ฐ๐๐๐ฎ๐น๐น๐ ๐บ๐๐ฐ๐ต ๐ต๐ถ๐ด๐ต๐ฒ๐ฟ. Under a unified protocol (ICL + up to 10ร the interaction budget), GPT-6-Astra achieves clear SOTA across all 12 benchmarks: RoboDojo: 27.6% at 1ร โ 81.9% at 10ร, DexJoCo: 87.3% vs. 40.3% prior SOTA, while the LIBERO benchmarks are pushed to saturation.
๐น ๐ง๐ผ๐ผ ๐บ๐๐ฐ๐ต ๐ฏ๐๐ฐ๐ธ ๐ณ๐ผ๐ฟ ๐๐ต๐ฒ ๐ฏ๐ฎ๐ป๐ด. Even a simple task like placing pens in a holder costs ~$55 in inference.
๐น ๐ง๐ต๐ฒ ๐๐ผ๐ฟ๐น๐ฑ ๐ฑ๐ผ๐ฒ๐๐ปโ๐ ๐๐ฎ๐ถ๐. Time matters as real robots overheat, and an egg burned while the model was still thinking.
Our takeaway: GPT-6-Astra has largely crossed the threshold of ๐ฒ๐บ๐ฏ๐ผ๐ฑ๐ถ๐ฒ๐ฑ ๐ฐ๐ผ๐บ๐ฝ๐ฒ๐๐ฒ๐ป๐ฐ๐ฒ, but ๐ฝ๐ฟ๐ฎ๐ฐ๐๐ถ๐ฐ๐ฎ๐น ๐ฒ๐บ๐ฏ๐ผ๐ฑ๐ถ๐ฒ๐ฑ ๐ถ๐ป๐๐ฒ๐น๐น๐ถ๐ด๐ฒ๐ป๐ฐ๐ฒ is not here yet.
๐ Project page: sensenova.ai/has-gpt-6-astraโฆ
๐ Report: sensenova.ai/gpt-6/assets/doโฆ
๐ป Code & model: github.com/OpenSenseNova/Senโฆ
You can run your own model-sharing site locally tonight: accounts, uploads, browsing, and comments on customizations people share:
- Users create accounts, upload models, and browse shared ones
- Hosts textual inversions, hypernetworks, VAEs, and more
- Next.js and tRPC stack, Apache-2.0 licensed
osp.fyi/civitai
What if an LLM could carry useful history forward with a fraction of the storage?
MemT 1.0 brings persistent, updatable memory into the modelโs own computation, with Mixture-of-Storage to scale native memory capacity.
At 128K tokens of history, the team reports that MemT 1.0-Nano delivers approximately:
1.86ร faster end-to-end inference
66ร lower persistent storage
Compared with full-context KV cache. For developers, that means less waiting and a smaller persistent memory footprint in this evaluation, a compelling direction for applications with long interaction histories.
The model family includes Nano and Flash. Explore the release:
huggingface.co/collections/Mโฆ
Can large language models retain information from past interactions and reuse it in the next task?
Introducing MemT-1.0: Towards Memory-Native Evolution.
This time, we've built memory directly into the model architecture.
Our MemT-1.0-Nano and MemT-1.0-Flash make persistent memory part of model computation, carrying information across interactions.
