CCL-Bench — Benchmark Results

Collective Communication & LLM Performance Metrics

CCL-Bench is a collaborative benchmarking project studying how distributed LLM workloads behave across different hardware, frameworks, and communication libraries.

… workloads
🔍
🏆 Leaderboard

⇧ Upload Trace & Workload Card

① Contributor
② Model & Data
③ Hardware
④ Framework
⑤ Files & Upload
Workload name is required
Contributor name is required
Valid email is required
Model family is required
Phase is required
Batch size is required
Sequence length is required
Hardware type is required
Hardware model is required
Total count is required
Framework is required
📦
Drag & drop trace files here
or browse your computer
.nsys-rep, .json, .tar, .gz, .tgz, .zip, .bz2, .xz, .zst  ·  Max 500 MB
Uploading…
Select 1 more to compare