sungyub kim

sungyub

1 3 6

https://sungyubkim.github.io/

AI & ML interests

LLM

Recent Activity

authored a paper 26 days ago

KV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache Quantization

authored a paper 26 days ago

LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding

liked a Space about 1 month ago

FineEnvs/multi-harness-rl

View all activity

Organizations

None yet

authored 2 papers 26 days ago

KV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache Quantization

Paper • 2610.14448 • Published Oct 10 • 35

LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding

Paper • 2410.03355 • Published Oct 4, 2024

liked a Space about 1 month ago

The ultimate guide to multi-harness RL

🔀

295

Train open models with RL inside real agent harnesses

updated a dataset 5 months ago

sungyub/deepscaler-preview-verl

Viewer • Updated Jun 26 • 38k • 1.26k

New activity in sungyub/ifbench-verl 7 months ago

Hello, I find datatrove libarary doesn't have datatrove.utils.reward_score.ifeval

#1 opened 7 months ago by

Wuyangqian

liked a Space 8 months ago

The Synthetic Data Playbook: Generating Trillions of the Finest Tokens

📝

294

Explore synthetic data experiments on a bookshelf visualization

upvoted 2 articles 10 months ago

Article

We Got Claude to Build CUDA Kernels and teach open models!

burtenshaw, evalstate, merve, pcuenq

•

Jan 28

• 158

Article

Unlocking Agentic RL Training for GPT-OSS: A Practical Retrospective

•

Jan 27

• 82

updated a collection 10 months ago

VERL QA Datasets

Collection

High-quality QA generation datasets in VERL format: document QA, table reasoning, and multi-hop reasoning tasks. • 6 items • Updated Mar 2

updated a dataset 10 months ago

sungyub/qa-verl-unified

Viewer • Updated Jan 8 • 86.4k • 87

published a dataset 10 months ago

sungyub/qa-verl-unified

Viewer • Updated Jan 8 • 86.4k • 87

updated 2 datasets 10 months ago

sungyub/docqa-rl-verl

Viewer • Updated Jan 8 • 3.6k • 24

sungyub/code-verl-unified

Viewer • Updated Jan 8 • 959k • 1.11k • 3

liked 2 Spaces 11 months ago

FineWeb: decanting the web for the finest text data at scale

🍷

1.47k

Explore the FineWeb dataset and its creation process

Evaluation Guidebook

📝

374

Explore LLM benchmark score trends over time

updated a dataset about 1 year ago

sungyub/codev-r1-verl

Viewer • Updated Nov 11, 2025 • 3.13k • 27

upvoted an article about 1 year ago

Article

Let's talk about LLM evaluation

clefourrier

•

May 23, 2024

• 212

liked 2 Spaces about 1 year ago

The Ultra-Scale Playbook

🌌

4.13k

The ultimate guide to training LLM on large GPU Clusters

The Smol Training Playbook

📚

3.39k

The secrets to building world-class LLMs

updated a dataset about 1 year ago

sungyub/math-verl-unified

Viewer • Updated Nov 9, 2025 • 2.27M • 436 • 3

sungyub kim

AI & ML interests

Recent Activity

Organizations

sungyub's activity

The ultimate guide to multi-harness RL

Hello, I find datatrove libarary doesn't have datatrove.utils.reward_score.ifeval

The Synthetic Data Playbook: Generating Trillions of the Finest Tokens

We Got Claude to Build CUDA Kernels and teach open models!

Unlocking Agentic RL Training for GPT-OSS: A Practical Retrospective

FineWeb: decanting the web for the finest text data at scale

Evaluation Guidebook

Let's talk about LLM evaluation

The Ultra-Scale Playbook

The Smol Training Playbook