Scott Condron @_ScottCondron

Helping build AI/ML dev tools at @weights_biases. I post about machine learning, data visualisation, software tools. scottcondron.com Dublin, Ireland Joined April 2018

Tweets

3K
Followers

6K
Following

2K
Likes

6K

Scott Condron @_ScottCondron

a week ago

Great to see people trying out using weave traces in W&B model training runs to inspect agent rollouts. Would love to chat to more people trying this out and to hear any other rollout visualizations you’d like to see on top of this

Fred Jonsson @enginoid

2 weeks ago

0 1 4 600 1

1 0 0 464 2

W&B Weave @weave_wb

2 weeks ago

Your RL run just spiked at step 89! But, do you know why? We’re fixing that. Today we’re launching W&B Weave Traces to give you a step by step look into your agent’s decisions. This is the first drop from our fresh new integration with @OpenPipeAI. More RL magic is incoming.

2 31 355 288K 44

Download Video

Boris Dayma 🖍️ @borisdayma

2 weeks ago

Interesting Muon experiment 🤓 Learning rate of Adam parameters (embeddings/gains) does not matter so much (here from 1e-4 to 1e-2). It’s more about Muon LR

1 2 13 1K 2

Download Image

will brown @willccbb

2 weeks ago

rollout logging for verifiers is now natively supported in @weave_wb :) huge thanks to @weights_biases for driving the integration, and congrats on the release!!

Shawn Lewis @shawnup

2 weeks ago

rollout logging for verifiers is now natively supported in @weave_wb :) huge thanks to @weights_biases for driving the integration, and congrats on the release!!

15 27 160 36K 111

Download Video

7 12 149 15K 39

David Corbitt @dvdcrbt

2 weeks ago

This is one of the first things I asked for when joining @weights_biases. I had this whole argument prepared for why it would be so useful to RL devs when debugging. Only to learn the entire feature set had been built and was ready to ship 😂

Shawn Lewis @shawnup

2 weeks ago

15 27 160 36K 111

Download Video

0 3 10 542 1

Shawn Lewis @shawnup

2 weeks ago

Announcing "look at your damn data" for Reinforcement Learning! You now get the full power of @weave_wb traces in the @weights_biases workspace. For any wandb run that calls an LLM, just "pip install weave" and "import weave" in your code. We also have out-of-the-box…

15 27 160 36K 111

Download Video

Alex Volkov (Thursd/AI) @altryne

2 weeks ago

Folks, this really is a big deal for everyone who does RL and wants a nicer way to see the rollouts! Combining @weave_wb and @weights_biases Models has finally happened, and is super easy to set up! We're leaning into RL, and this is just the start 👀 Supports @willccbb…

Shawn Lewis @shawnup

2 weeks ago

15 27 160 36K 111

Download Video

1 8 30 8K 19

Download Image

Weights & Biases @weights_biases

3 weeks ago

Evaluating LLMs in @weave_wb just got a whole lot easier! Now you can run your evals right in the UI, without writing a single line of code. Select datasets, pick models, include LLM-as-judge scorers, and hit Run. Simple as that.

3 7 99 96K 22

Download Image

kalomaze @kalomaze

3 weeks ago

if you can't handle me at my worst of n then you don't deserve me at my best of n

5 2 93 4K 10

Download Image

Weights & Biases @weights_biases

3 weeks ago

Our W&B Inference service just got a major upgrade. @Zai_org's GLM-4.5 is now live on our platform! This ~300B param open model excels in reasoning, coding & agentic tasks via multi-token prediction & RL. See below to find out how you can get $50 in credits to try it out!

13 16 253 141K 26

Download Image

Shawn Lewis @shawnup

3 weeks ago

Coming soon to a W&B workspace near you: the full power of Weave traces. Especially useful for RL!

1 3 18 2K 5

Download Image

Kyle Corbitt @corbtt

4 weeks ago

Ok, some big news that I've been sitting on for a minute: @OpenPipeAI is getting acquired by @CoreWeave!

98 15 553 48K 56

Shawn Lewis @shawnup

4 weeks ago

🧵 We acquired OpenPipe! OpenPipe's ART framework makes it easy to beat foundation models with smaller open models on real problems, using reinforcement learning.

22 39 677 82K 398

Download Image

Scott Condron @_ScottCondron

4 weeks ago

Is there a library that creates a completion-compatible API for a given python function?

1 0 0 519 1

Kyle Corbitt @corbtt

4 weeks ago

🚨 We’ve just published a recipe to train a frontier-level deep research agent using RL. With just 30 hours on an H200, any developer can now beat Sonnet-4 on DeepResearch Bench using open-source tools. (Thread 🧵)

38 175 1K 211K 2K

Download Image

Julia Neagu @JuliaANeagu

a month ago

fine tuning hasn’t taken off (yet) because (1) you need to have a good idea of what you want to improve, and (2) you need a way to label data at scale. most builders don’t know enough about (1), mostly because they have little proper monitoring. until recently, (2) meant time…