The appeal goes back to the SAME AI that banned the account? I have questions. Starting with who can actually undo the decision. My latest for Inc: inc.com/heather-wilde/metas-…

Our team spends a lot of time making sure very tight numerics before enabling the model for training. This means you can train with confidence that the infra won’t silently fail or corrupt your rl runs. We also consistently test against open source frameworks and take a lot of pride in providing the best numerics aligned infra for training you could get. This matters specially for large rl runs or large moe model

With the increasing adoption of task-specific model routers, we will see many more verticalised, Jev-like classification and regression models, each trained to outperform their general-purpose counterparts on specific tasks.

TL;DR •Numerical mismatch can derail training: in a Fireworks GLM 5.2 experiment, the same algorithm and data produced collapsing reward without alignment and stable reward with alignment over 25 steps. •Training MoE models adds complication to alignment: our Qwen3.5-MoE investigation found that differences in how expert outputs were combined caused disagreement even when one implementation used higher precision. •These discrepancies can distort training updates and resemble problems with data, rewards, or learning rate, sending teams through expensive experiments that leave the underlying cause unresolved. •Frontier training requires co-optimized training and inference: Fireworks develops and validates the trainer and rollout engine together so teams can scale reinforcement learning with alignment across numerics, kernels, and MoEs.

The best news today. We got into NeurIPS!

Rollouts drive most of RL's compute cost. But splitting rollout and training across engines risks numerical mismatches, and in MoE models, that can even send tokens to different experts. We co-build both engines at Fireworks, so training stays fast and consistent. Learn more: bit.ly/4ALjga5

"Near identical composition, but Ember-1 used 9x less thinking tokens and was 4x faster at half the cost." Love the analysis between @FireworksAI_HQ Ember-1 and Kimi K3 🔥

So cool!

It’s ready now!

Day 2 at @AIconference with my favorite @FireworksAI_HQ team 🎆 come talk to our team about model inference, training, and more 🙌

Congrats to the @Fulcrum_inc team on Echo! Final training runs were done entirely via Fireworks' serverless training API, on a Kimi K3 LoRA (rank 32). Make sure to read their writeup, which includes the "persona post-training" approach and how they landed a $4.5k total training cost.

Nice to see @FireworksAI_HQ leading the open model inference economy 🎆

To every one else out there, thinking training is too costly, IT'S NOT! you can beat frontier models on your domain specific tasks for less than $5k as demonstrated by this exciting launch from @fulcrum_inc Congratulations @fulcrum_inc ! Hello Echo :)

🎆 @FireworksAI_HQ Ember-1: Built on Kimi K3, fewer tokens, same quality!

i found the famous @sophiamyang and she agreed to take a selfie with me then i asked her to approve my time off (she did!) fireworks devrel team is 🫰

GLM 5.3 Flash is now available for training on the Serverless Training API - accessible to all, with both vision and text support. This model performs great on our benchmarks for agentic coding, document analysis, and tool use while being cost-efficient to serve. Get started today: docs.fireworks.ai/fine-tunin…

Code & Cocktails 🍸 getting started 🎆 already leaning lots from the community on how and why they love @FireworksAI_HQ 💖

Join our code & cocktail 🍸 tonight in SF 🎆

i'll be here tonight hiding behind @Prof_OZ as he drops bars about training your own models sinan, is there any POSSIBLE way we can get you to rap about deploying models at this thing? i will pay /top dollar/

Fireworks is at @AIconference! Find us at Booth #245 (Shed A) and swing by our Coffee Lounge in the Innovation Hub. Catch our own Jetashree Ravi's breakout session, and @RobFergusonIII on the Day Two keynote stage. See you on the show floor!