I am SO impressed by this model! It's cheap as deepseek v4 flash, better than minimax m3 I would say, and almost a glm 5.2. 120b params only for such a good result, with 258k context. This is amazing.
Yesterday I was making experiments running laguna xs 2.1 in my local machine (a smaller version), with a ryzen 5700x, 64 RAM and a 4060 ti 8gb nvidia
It was able to run locally with a context of 128k using llama-cpp and I tried using it for local development in a project that used uv, python, tool calls, file system explorations, websearches - all worked as a charm... in my own hardware
I was SO impressed. Then I saw open router is letting people try laguna S 2.1 for free - it's a model with 120b params and 258k context - if you top up at least 10 usd, it gives you 1000 requests per day for free so you can try it out. but the paid version is super cheap as well, specially the cache.
Yep! I loved Qwen 3.5 and 3.6, jumped to Minimax 2.5 and 2.7, but Laguna S2.1 feels like I have SOTA at home.
We’re going to get to the point where the enormous size of models will not be able to keep being updated with the huge amount of content being generated online, and so I think a strong SOTA-esque thinking model with a good harness, fast web crawler, and large MCP capabilities will be the future tool of choice, rather than larger and larger static models (unless someone creates upgradable but compact embeddings)
- How it compares to Cassandra regarding latency?
- How is the client API? Does it support lock free communication? Can you send things to nodes in batches?
Please, could you share the uses you thoughts for it? This is what we are most interested in. We did the platform for ourselves, and it seems powerful now, we would like to know what could interest others...
Scheduling is done by any external tool like cron, a java app running quartz or control-m. However, the processor job knows the last time it was run and know which data it should process to update the time. You might also choose to run the job overriding the last time it ran, to reprocess data it you want.
This is interesting, indeed.
As your platform is agnostic to the runtime, wouldn't it be easy to use ours instead of hadoop or storm? Or you don't see any benefit? About the tuples, you still could use them, if I understood it well...
Yesterday I was making experiments running laguna xs 2.1 in my local machine (a smaller version), with a ryzen 5700x, 64 RAM and a 4060 ti 8gb nvidia
It was able to run locally with a context of 128k using llama-cpp and I tried using it for local development in a project that used uv, python, tool calls, file system explorations, websearches - all worked as a charm... in my own hardware
I was SO impressed. Then I saw open router is letting people try laguna S 2.1 for free - it's a model with 120b params and 258k context - if you top up at least 10 usd, it gives you 1000 requests per day for free so you can try it out. but the paid version is super cheap as well, specially the cache.