Hacker News: srush

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Thu, 30 Oct 2025 01:17:40 +0000

Awesome to hear, I will share with the team.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 21:11:28 +0000

Roughly, we had Cursor software engineers record real questions they were asking models, and then had them record the PR that they made that contained the result. We then cleaned these up. That is the benchmark.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 21:04:41 +0000

A lot of people use it! It scores very well on our benchmarks, significantly better than Composer-1.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 20:58:52 +0000

We use Ray data for our map-style processing jobs. For example one tool have runs over all the rollouts from the RL system and collects qualitative statistics to understand which type of agent trajectories are being reward, and what types of searches and terminal commands are being made.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 20:29:51 +0000

Oh good question. Actually speaking at the Ray Summit next week in SF so we will talk more about it. We used Ray throughout the pipeline for running evals, for the RL controller, for data collation, and for visualizations. One tool we found helpful was Ray Data which let us easily scale over data and run logs.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 19:25:41 +0000

We train with a single agent. is that the question?

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 19:24:54 +0000

neat!

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 19:24:00 +0000

Our view is that there is a now a minimal amount of intelligence that is necessary to be productive, and that if you can pair that with speed that is awesome.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 18:10:18 +0000

Let us know how you like it.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 18:09:56 +0000

Straight up untrue.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 18:08:02 +0000

There are lots of good models we like here. But we agree that getting the right point on the smart+fast graph can make agentic coding feel really good.

(Cursor researcher)

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 18:06:25 +0000

Thanks! Yeah, been working here for 9 months now. Fascinated byt agentic coding both as a researcher and user.

Totally agree that "smart model" is the table stakes for usefulness these days.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 18:03:19 +0000

We will do our best. Luckily I don't think there are major telecom companies called Composer-2.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 18:02:14 +0000

Unfortunately not, as we used our own internal code for the benchmark. We would also like to see more benchmarks that reflect the day-to-day agentic coding use.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 17:48:35 +0000

The blog talks about the training process. Specifically we trained with RL post-training on coding examples.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 17:47:27 +0000

Thanks! I do like the labs blog posts as well though, OpenAI and Anthropic have some classics.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 17:44:47 +0000

Our primary focus is on RL post-training. We think that is the best way to get the model to be a strong interactive agent.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 17:34:25 +0000

We did a lot of internal testing and thought this model was already quite useful for release.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 17:32:39 +0000

Cheetah was an earlier (and dumber) version of this model that we used to test production speed. They are both developed in-house. If you liked Cheetah, give this model a try.

New comment by srush in "Composer: Building a fast frontier model with RL"

srush — Wed, 29 Oct 2025 17:23:34 +0000

We like the name Composer and were sad to see it go. Excited to bring it back. (Agree Cheetah is a cool name too.)