Skip to main content
Link
Menu
Expand
(external link)
Document
Search
Copy
Copied
Marlowe Hackathon 2026
Marlowe Hackathon 2026
Models
The Task and the Data
Key Ideas
Monitoring Training
Beginner
B1. Score Kev-4B before training
B2. Fine-tune Kev-4B
B3. Label filings on one Marlowe GPU
B4. Label on a full node
Intermediate
I1. Fine-tune and profile Kev-4B
I2. Move your data with Globus
I3. Label the test split on a node
I4. Label the corpus on two nodes
Advanced
A1. Score Kev-27B before training
A2. Fine-tune Qwen 27B
A3. Fine-tune on two nodes
A4. Inference at scale with Qwen 235B
A5. Distill a student
Moving Data with Globus
Label with Your Model
Take It Home
Reference
Search Marlowe Hackathon 2026
GitHub repository
GSB RCpedia
HAI Marlowe Docs
Intermediate
I1. Fine-tune and profile Kev-4B
I1. Fine-tune and profile Kev-4B
This walkthrough is still being written.
Where
Yens, 1 GPU
What you do
Fine-tune Kev-4B and measure time per step.
Data
train.jsonl
, 54,426 filings (2004 to 2018) with Jev’s soft labels
Model
Kev-4B, merged to full weights