gflow Quick Reference

Daemon

gflowd init
gflowd up
gflowd status
gflowd down
gflowd restart

Inspect + Monitor

# GPUs (availability + allocations)
ginfo

# Jobs (default: active jobs)
gqueue
gqueue -a              # include completed
gqueue -P ml-research  # filter by project

# Useful formats
gqueue -f JOBID,NAME,ST,TIME,NODES,NODELIST(REASON)
gqueue -f JOBID,NAME,PROJECT,ST,TIME,NODES,NODELIST(REASON)
gqueue -s Running -f JOBID,NAME,ST,NODES,NODELIST(REASON)

# Dependency tree
gqueue -t

Example gqueue -t output:

JOBID  NAME   ST  TIME      NODES  NODELIST(REASON)
1      prep   CD  00:02:15  0      -
├─2    train  R   00:10:03  1      0
└─3    eval   PD  -         0      (WaitingForDependency)

Submit Jobs (gbatch)

# Command
gbatch python train.py --epochs 100

# Script
gbatch train.sh

# Common options
gbatch --gpus 1 --time 2:00:00 --name train-resnet python train.py
gbatch --gpus 1 --shared --gpu-memory 20G python train.py
gbatch --priority 50 python urgent.py
gbatch --conda-env myenv python script.py
gbatch --project ml-research python train.py
gbatch --dry-run --gpus 1 python train.py

Notes:

• --memory (--max-mem) limits host RAM.
• --gpu-memory (--max-gpu-mem) limits per-GPU VRAM.
• --shared requires --gpu-memory.

Script Directives

Only these are parsed from scripts:

#!/bin/bash
# GFLOW --gpus=1
# GFLOW --shared
# GFLOW --time=2:00:00
# GFLOW --memory=4G
# GFLOW --gpu-memory=20G
# GFLOW --priority=20
# GFLOW --conda-env=myenv
# GFLOW --depends-on=123
# GFLOW --project=ml-research

CLI flags override script directives.

Dependencies

# Single dependency
gbatch --depends-on <job_id|@|@~N> python next.py

# Multiple dependencies
gbatch --depends-on-all 1,2,3 python merge.py     # AND
gbatch --depends-on-any 4,5 python fallback.py    # OR

# Shorthands: @ = most recent job, @~N = Nth most recent

# Dependency failure behavior
gbatch --depends-on 123 --no-auto-cancel python next.py

Arrays

gbatch --array 1-10 python process.py --task '$GFLOW_ARRAY_TASK_ID'

Params (--param)

gbatch --param lr=0.001,0.01 --param bs=32,64 python train.py --lr {lr} --batch-size {bs}
gbatch --param-file params.csv --name-template 'run_{id}' python train.py --id {id}

Control

# Cancel (use --dry-run to see dependent jobs)
gcancel <job_id>
gcancel --dry-run <job_id>

# Hold/release
gjob hold <job_id>
gjob release <job_id>

# Details / redo / update
gjob show <job_id>
gjob redo <job_id>
gjob redo <job_id> --cascade
gjob update <job_id> --gpus 2 --time-limit 4:00:00

Runtime Control (gctl)

# Restrict which GPUs the scheduler can allocate (new allocations only)
gctl show-gpus
gctl set-gpus 0,2
gctl set-gpus all

# Group concurrency limit
gctl set-limit <job_or_group_id> 2

# Reservations (block out GPUs for a user/time window)
gctl reserve create --user alice --gpus 2 --start '2026-01-28 14:00' --duration 2h
gctl reserve list --active
gctl reserve list --timeline --range 48h
gctl reserve cancel <reservation_id>

Time Format (--time)

• HH:MM:SS (e.g. 2:30:00)
• MM:SS (e.g. 5:30)
• MM minutes (e.g. 30)

Note: a single number is minutes. Use 0:30 for 30 seconds.

States

Code	State
PD	Queued
H	Hold
R	Running
CD	Finished
F	Failed
CA	Cancelled
TO	Timeout

Paths

~/.config/gflow/gflow.toml
~/.local/share/gflow/state.msgpack  (or state.json for legacy)
~/.local/share/gflow/logs/<job_id>.log

See Also

• Job Submission
• Job Dependencies
• Time Limits
• GPU Management