scriptory/magister.example.toml at master · unattended-backpack/scriptory · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
# Example configuration for Magister.
# Copy this file to magister.toml and adjust as needed.

# REQUIRED: Publicly accessible address where this Magister can be reached.
# Used by Hierophant to send drop requests via HTTP.
# Format: http://[host] (do not include port, trailing slash, or /drop path)
# Port is specified separately via http_port configuration.
this_magister_addr = "http://magister"

# REQUIRED: IP address or hostname where Contemplants can reach Hierophant.
# Passed to Contemplants when they are created so they know where to connect.
hierophant_ip = "hierophant"

# REQUIRED: HTTP port where Hierophant is listening.
# Contemplants will connect via WebSocket at ws://[hierophant_ip]:[hierophant_http_port]/ws
hierophant_http_port = 9010

# REQUIRED: Vast.ai API key for managing instances.
# Obtain from https://vast.ai/ under Account > API Keys
vast_api_key = "your-vast-api-key-here"

# REQUIRED: Vast.ai template hash to use for creating instances.
# Template should contain a Contemplant image configured to run on startup.
# Find template hash in the Vast.ai web console under your saved templates.
template_hash = "819fdf2e42fc8ceb32f295465b5bb21e"

# REQUIRED: Number of instances to maintain.
# Magister will continuously monitor and ensure this many instances are running.
number_instances = 1

# OPTIONAL: HTTP server port (default: 8555).
# http_port = 8555

# OPTIONAL: Seconds to wait between Vast.ai API calls (default: 10).
# Helps avoid rate limiting from Vast.ai.
# vast_api_call_backoff_secs = 10

# OPTIONAL: Seconds between instance polling checks (default: 30).
# How often Magister checks if instances need to be created or cleaned up.
# task_polling_interval_secs = 30

# OPTIONAL: Seconds to wait for Contemplant verification (default: 180).
# How long to wait after creating an instance for the Contemplant to call /verify
# before considering the instance failed and dropping it.
# contemplant_verification_timeout_secs = 300

# OPTIONAL: List of Vast.ai host IDs to avoid.
# Instances will not be created on these hosts.
# bad_hosts = [213498, 74292, 113132]

# OPTIONAL: List of Vast.ai machine IDs to avoid.
# Instances will not be created on these machines.
# bad_machines = [12217, 19571]

# OPTIONAL: List of preferred Vast.ai host IDs.
# These hosts will be prioritized when creating instances.
# good_hosts = [207289, 1276]

# OPTIONAL: List of preferred Vast.ai machine IDs.
# These machines will be prioritized when creating instances.
# good_machines = [13428, 8218]

# Vast query configuration controls which machines are eligible for instance creation.
# These settings are converted into a Vast.ai search query.
[vast_query]

# REQUIRED: Allocated storage in GB for the instance.
allocated_storage = 16

# REQUIRED: GPU model name to search for.
# Common options: "RTX 4090", "RTX 3090", "A100", etc.
gpu_name = "RTX 4090"

# REQUIRED: Minimum host reliability score (0.0 to 1.0).
# Higher values mean more reliable hosts but fewer options.
reliability = 0.99

# REQUIRED: Minimum CUDA version required.
min_cuda_version = 12.8

# REQUIRED: Minimum GPU RAM in GB.
gpu_ram = 21

# REQUIRED: Minimum disk space in GB.
disk_space = 16

# REQUIRED: Minimum rental duration in hours.
# Higher values may give better availability.
duration = 192679

# REQUIRED: Maximum cost per hour in USD.
# Instances costing more than this will be filtered out.
cost_per_hour = 0.60

# Contemplant configuration controls settings for Contemplants spawned by this Magister.
# These settings are passed as environment variables to Contemplants on Vast.ai.
# Note: MAGISTER_DROP_ENDPOINT and HIEROPHANT_WS_ADDRESS are managed automatically.
[contemplant]

# OPTIONAL: Prover type - "cpu" or "cuda" (default: "cpu").
# CPU proving is slower but doesn't require GPU-specific setup.
# prover_type = "cpu"

# OPTIONAL: Human-readable name for Contemplants (default: generated from names.txt).
# contemplant_name = "paulos"

# OPTIONAL: Port for HTTP health check server (default: 9011).
# http_port = 9011

# OPTIONAL: Moongate CUDA prover endpoint (default: none).
# Only used when prover_type is "cuda". If not set, Contemplant will spin up
# a moongate-server Docker container automatically.
# moongate_endpoint = "http://localhost:3000/twirp/"

# OPTIONAL: Heartbeat interval in seconds (default: 30).
# How often Contemplants tell Hierophant they are still alive.
# heartbeat_interval_seconds = 30

# OPTIONAL: Maximum number of finished proofs stored in memory (default: 2).
# max_proofs_stored = 2

# OPTIONAL: Path to log file for progress tracking (default: "./moongate.log").
# moongate_log_path = "./moongate.log"

# OPTIONAL: Log polling interval in milliseconds (default: 2000).
# How frequently to check the moongate log file for progress updates.
# watcher_polling_interval_ms = 2000

# OPTIONAL: SSH public keys for debugging access (default: none).
# Allows SSH access to Contemplant instances on port 2222 for debugging.
# Format: newline-separated SSH public keys.
# ssh_authorized_keys = """
# ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIAbc123... user@host
# ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQC... another@host
# """