AI-Examples/environment-config-example.yaml at main · awesome-release/AI-Examples · 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
---
auto_deploy: false
context: release-ry6clz
domain: ai-playground.releaseapp.io
repo_name: Draiken/AI-Examples
hostnames:
  - nemo-inference: nemo-${env_id}.${domain}
resources:
  cpu:
    limits: 24
    requests: 12
  memory:
    limits: 64Gi
    requests: 16Gi
  replicas: 1
shared_volumes:
  - name: tmp
    size: 20Gi
    type: persistent
jobs:
  - name: peft-tuning
    command:
      - "/bin/peft_tuning"
    from_services: nemo-training
    completed_timeout: 5400
    node_selector:
      - key: nvidia.com/gpu
        value: "true"
    volumes:
      - name: shmem
        type: shmem
        size: 16Gi
        mount_path: "/dev/shm"
      - type: s3
        bucket: release-ry6clz-static-builds
        mount_path: "/bucket"
services:
  - name: nemo-training
    image: draiken/ai-examples/nemo-training
    build:
      context: "."
    node_selector:
      - key: nvidia.com/gpu
        value: "true"
    volumes:
      - name: shmem
        type: shmem
        size: 16Gi
        mount_path: "/dev/shm"
      - claim: tmp
        mount_path: "/tmp"
      - type: s3
        bucket: release-ry6clz-static-builds
        mount_path: "/bucket"
    cpu:
      limits: 24
      requests: 12
    memory:
      limits: 64Gi
      requests: 16Gi
  - name: nemo-inference
    image: nvcr.io/ea-bignlp/ga-participants/nemofw-inference:23.10
    has_repo: false
    command:
      - python
      - "/opt/NeMo/scripts/deploy/deploy_triton.py"
      - "--nemo_checkpoint"
      - "/bucket/ai-models-tmp/llama-2-7b-hf.nemo"
      - "--model_type"
      - llama
      - "--triton_model_name"
      - release-model
    node_selector:
      - key: nvidia.com/gpu
        value: "true"
    volumes:
      - name: shmem
        type: shmem
        size: 16Gi
        mount_path: "/dev/shm"
      - claim: tmp
        mount_path: "/tmp"
      - type: s3
        bucket: release-ry6clz-static-builds
        mount_path: "/bucket"
    cpu:
      limits: 24
      requests: 8
    memory:
      limits: 64Gi
      requests: 16Gi
s3_volumes:
  - bucket: release-ry6clz-static-builds
    region: us-west-2
workflows:
  - name: setup
    parallelize:
      - step: fine-tuning
        tasks:
          - jobs.peft-tuning
        halt_on_error: true
        wait_for_finish: true
      - step: inference
        tasks:
          - services.nemo-inference
  - name: patch
    parallelize:
      - step: inference
        tasks:
          - services.nemo-inference
  - name: teardown
    parallelize:
      - step: remove-environment
        tasks:
          - release.remove_environment
tracking_branch: peft