Skip to content

Commit 63cc6d4

Browse files
OlivierDehaeneOlivierDehaene
authored
feat: support loading from .pth (#12)
1 parent ead3de3 commit 63cc6d4

File tree

10 files changed

+845
-291
lines changed

10 files changed

+845
-291
lines changed

‎.github/workflows/build_75.yaml

Lines changed: 158 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,158 @@
1+
name: Build and push Cuda Turing docker image to registry
2+
3+
on:
4+
workflow_dispatch:
5+
push:
6+
branches:
7+
- 'main'
8+
tags:
9+
- 'v*'
10+
pull_request:
11+
paths:
12+
- ".github/workflows/build_75.yaml"
13+
# - "integration-tests/**"
14+
- "backends/**"
15+
- "core/**"
16+
- "router/**"
17+
- "Cargo.lock"
18+
- "rust-toolchain.toml"
19+
- "Dockerfile"
20+
branches:
21+
- 'main'
22+
23+
jobs:
24+
start-runner:
25+
name: Start self-hosted EC2 runner
26+
runs-on: ubuntu-latest
27+
env:
28+
AWS_REGION: us-east-1
29+
EC2_AMI_ID: ami-03cfed9ea28f4b002
30+
EC2_INSTANCE_TYPE: r5.8xlarge
31+
EC2_SUBNET_ID: subnet-931b34f5,subnet-ecb993cd,subnet-943dc2d8,subnet-45371f1a,subnet-ee93e0df,subnet-fddc3dfc
32+
EC2_SECURITY_GROUP: sg-030175c435ac141d6
33+
outputs:
34+
label: ${{ steps.start-ec2-runner.outputs.label }}
35+
ec2-instance-id: ${{ steps.start-ec2-runner.outputs.ec2-instance-id }}
36+
steps:
37+
- name: Configure AWS credentials
38+
uses: aws-actions/configure-aws-credentials@v1
39+
with:
40+
aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
41+
aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
42+
aws-region: ${{ env.AWS_REGION }}
43+
- name: Start EC2 runner
44+
id: start-ec2-runner
45+
uses: philschmid/philschmid-ec2-github-runner@main
46+
with:
47+
mode: start
48+
github-token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
49+
ec2-image-id: ${{ env.EC2_AMI_ID }}
50+
ec2-instance-type: ${{ env.EC2_INSTANCE_TYPE }}
51+
subnet-id: ${{ env.EC2_SUBNET_ID }}
52+
security-group-id: ${{ env.EC2_SECURITY_GROUP }}
53+
aws-resource-tags: > # optional, requires additional permissions
54+
[
55+
{"Key": "Name", "Value": "ec2-tei-github-runner"},
56+
{"Key": "GitHubRepository", "Value": "${{ github.repository }}"}
57+
]
58+
59+
build-and-push-image:
60+
concurrency:
61+
group: ${{ github.workflow }}-${{ github.job }}-75-${{ github.head_ref || github.run_id }}
62+
cancel-in-progress: true
63+
needs: start-runner # required to start the main job when the runner is ready
64+
runs-on: ${{ needs.start-runner.outputs.label }} # run the job on the newly created runner
65+
permissions:
66+
contents: write
67+
packages: write
68+
# This is used to complete the identity challenge
69+
# with sigstore/fulcio when running outside of PRs.
70+
id-token: write
71+
security-events: write
72+
steps:
73+
- name: Checkout repository
74+
uses: actions/checkout@v3
75+
- name: Initialize Docker Buildx
76+
uses: docker/setup-buildx-action@v2.0.0
77+
with:
78+
install: true
79+
- name: Configure sccache
80+
uses: actions/github-script@v6
81+
with:
82+
script: |
83+
core.exportVariable('ACTIONS_CACHE_URL', process.env.ACTIONS_CACHE_URL || '');
84+
core.exportVariable('ACTIONS_RUNTIME_TOKEN', process.env.ACTIONS_RUNTIME_TOKEN || '');
85+
- name: Inject slug/short variables
86+
uses: rlespinasse/github-slug-action@v4.4.1
87+
- name: Tailscale
88+
uses: tailscale/github-action@7bd8039bf25c23c4ab1b8d6e2cc2da2280601966
89+
with:
90+
authkey: ${{ secrets.TAILSCALE_AUTHKEY }}
91+
- name: Login to GitHub Container Registry
92+
if: github.event_name != 'pull_request'
93+
uses: docker/login-action@v2
94+
with:
95+
registry: ghcr.io
96+
username: ${{ github.actor }}
97+
password: ${{ secrets.GITHUB_TOKEN }}
98+
- name: Login to internal Container Registry
99+
uses: docker/login-action@v2.1.0
100+
with:
101+
username: ${{ secrets.TAILSCALE_DOCKER_USERNAME }}
102+
password: ${{ secrets.TAILSCALE_DOCKER_PASSWORD }}
103+
registry: registry.internal.huggingface.tech
104+
- name: Extract metadata (tags, labels) for Docker
105+
id: meta-75
106+
uses: docker/metadata-action@v4.3.0
107+
with:
108+
images: |
109+
registry.internal.huggingface.tech/api-inference/text-embeddings-inference
110+
ghcr.io/huggingface/text-embeddings-inference
111+
tags: |
112+
type=semver,pattern=turing-{{version}}
113+
type=semver,pattern=turing-{{major}}.{{minor}}
114+
type=raw,value=turing-latest,enable=${{ github.ref == format('refs/heads/{0}', github.event.repository.default_branch) }}
115+
type=raw,value=turing-sha-${{ env.GITHUB_SHA_SHORT }}
116+
- name: Build and push Docker image
117+
id: build-and-push-75
118+
uses: docker/build-push-action@v4
119+
with:
120+
context: .
121+
file: Dockerfile-cuda
122+
push: ${{ github.event_name != 'pull_request' }}
123+
platforms: 'linux/amd64'
124+
build-args: |
125+
SCCACHE_GHA_ENABLED=on
126+
ACTIONS_CACHE_URL=${{ env.ACTIONS_CACHE_URL }}
127+
ACTIONS_RUNTIME_TOKEN=${{ env.ACTIONS_RUNTIME_TOKEN }}
128+
CUDA_COMPUTE_CAP=75
129+
GIT_SHA=${{ env.GITHUB_SHA }}
130+
DOCKER_LABEL=sha-${{ env.GITHUB_SHA_SHORT }}
131+
tags: ${{ steps.meta-75.outputs.tags }}
132+
labels: ${{ steps.meta-75.outputs.labels }}
133+
cache-from: type=registry,ref=registry.internal.huggingface.tech/api-inference/text-embeddings-inference:cache-75,mode=max
134+
cache-to: type=registry,ref=registry.internal.huggingface.tech/api-inference/text-embeddings-inference:cache-75,mode=max
135+
136+
stop-runner:
137+
name: Stop self-hosted EC2 runner
138+
needs:
139+
- start-runner
140+
- build-and-push-image
141+
runs-on: ubuntu-latest
142+
env:
143+
AWS_REGION: us-east-1
144+
if: ${{ always() }} # required to stop the runner even if the error happened in the previous jobs
145+
steps:
146+
- name: Configure AWS credentials
147+
uses: aws-actions/configure-aws-credentials@v1
148+
with:
149+
aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
150+
aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
151+
aws-region: ${{ env.AWS_REGION }}
152+
- name: Stop EC2 runner
153+
uses: philschmid/philschmid-ec2-github-runner@main
154+
with:
155+
mode: stop
156+
github-token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
157+
label: ${{ needs.start-runner.outputs.label }}
158+
ec2-instance-id: ${{ needs.start-runner.outputs.ec2-instance-id }}

‎.github/workflows/build_80.yaml

Lines changed: 158 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,158 @@
1+
name: Build and push Cuda Ampere docker image to registry
2+
3+
on:
4+
workflow_dispatch:
5+
push:
6+
branches:
7+
- 'main'
8+
tags:
9+
- 'v*'
10+
pull_request:
11+
paths:
12+
- ".github/workflows/build.yaml"
13+
# - "integration-tests/**"
14+
- "backends/**"
15+
- "core/**"
16+
- "router/**"
17+
- "Cargo.lock"
18+
- "rust-toolchain.toml"
19+
- "Dockerfile"
20+
branches:
21+
- 'main'
22+
23+
jobs:
24+
start-runner:
25+
name: Start self-hosted EC2 runner
26+
runs-on: ubuntu-latest
27+
env:
28+
AWS_REGION: us-east-1
29+
EC2_AMI_ID: ami-03cfed9ea28f4b002
30+
EC2_INSTANCE_TYPE: r5.8xlarge
31+
EC2_SUBNET_ID: subnet-931b34f5,subnet-ecb993cd,subnet-943dc2d8,subnet-45371f1a,subnet-ee93e0df,subnet-fddc3dfc
32+
EC2_SECURITY_GROUP: sg-030175c435ac141d6
33+
outputs:
34+
label: ${{ steps.start-ec2-runner.outputs.label }}
35+
ec2-instance-id: ${{ steps.start-ec2-runner.outputs.ec2-instance-id }}
36+
steps:
37+
- name: Configure AWS credentials
38+
uses: aws-actions/configure-aws-credentials@v1
39+
with:
40+
aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
41+
aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
42+
aws-region: ${{ env.AWS_REGION }}
43+
- name: Start EC2 runner
44+
id: start-ec2-runner
45+
uses: philschmid/philschmid-ec2-github-runner@main
46+
with:
47+
mode: start
48+
github-token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
49+
ec2-image-id: ${{ env.EC2_AMI_ID }}
50+
ec2-instance-type: ${{ env.EC2_INSTANCE_TYPE }}
51+
subnet-id: ${{ env.EC2_SUBNET_ID }}
52+
security-group-id: ${{ env.EC2_SECURITY_GROUP }}
53+
aws-resource-tags: > # optional, requires additional permissions
54+
[
55+
{"Key": "Name", "Value": "ec2-tei-github-runner"},
56+
{"Key": "GitHubRepository", "Value": "${{ github.repository }}"}
57+
]
58+
59+
build-and-push-image:
60+
concurrency:
61+
group: ${{ github.workflow }}-${{ github.job }}-80-${{ github.head_ref || github.run_id }}
62+
cancel-in-progress: true
63+
needs: start-runner # required to start the main job when the runner is ready
64+
runs-on: ${{ needs.start-runner.outputs.label }} # run the job on the newly created runner
65+
permissions:
66+
contents: write
67+
packages: write
68+
# This is used to complete the identity challenge
69+
# with sigstore/fulcio when running outside of PRs.
70+
id-token: write
71+
security-events: write
72+
steps:
73+
- name: Checkout repository
74+
uses: actions/checkout@v3
75+
- name: Initialize Docker Buildx
76+
uses: docker/setup-buildx-action@v2.0.0
77+
with:
78+
install: true
79+
- name: Configure sccache
80+
uses: actions/github-script@v6
81+
with:
82+
script: |
83+
core.exportVariable('ACTIONS_CACHE_URL', process.env.ACTIONS_CACHE_URL || '');
84+
core.exportVariable('ACTIONS_RUNTIME_TOKEN', process.env.ACTIONS_RUNTIME_TOKEN || '');
85+
- name: Inject slug/short variables
86+
uses: rlespinasse/github-slug-action@v4.4.1
87+
- name: Tailscale
88+
uses: tailscale/github-action@7bd8039bf25c23c4ab1b8d6e2cc2da2280601966
89+
with:
90+
authkey: ${{ secrets.TAILSCALE_AUTHKEY }}
91+
- name: Login to GitHub Container Registry
92+
if: github.event_name != 'pull_request'
93+
uses: docker/login-action@v2
94+
with:
95+
registry: ghcr.io
96+
username: ${{ github.actor }}
97+
password: ${{ secrets.GITHUB_TOKEN }}
98+
- name: Login to internal Container Registry
99+
uses: docker/login-action@v2.1.0
100+
with:
101+
username: ${{ secrets.TAILSCALE_DOCKER_USERNAME }}
102+
password: ${{ secrets.TAILSCALE_DOCKER_PASSWORD }}
103+
registry: registry.internal.huggingface.tech
104+
- name: Extract metadata (tags, labels) for Docker
105+
id: meta-80
106+
uses: docker/metadata-action@v4.3.0
107+
with:
108+
images: |
109+
registry.internal.huggingface.tech/api-inference/text-embeddings-inference
110+
ghcr.io/huggingface/text-embeddings-inference
111+
tags: |
112+
type=semver,pattern={{version}}
113+
type=semver,pattern={{major}}.{{minor}}
114+
type=raw,value=latest,enable=${{ github.ref == format('refs/heads/{0}', github.event.repository.default_branch) }}
115+
type=raw,value=sha-${{ env.GITHUB_SHA_SHORT }}
116+
- name: Build and push Docker image
117+
id: build-and-push-80
118+
uses: docker/build-push-action@v4
119+
with:
120+
context: .
121+
file: Dockerfile-cuda
122+
push: ${{ github.event_name != 'pull_request' }}
123+
platforms: 'linux/amd64'
124+
build-args: |
125+
SCCACHE_GHA_ENABLED=on
126+
ACTIONS_CACHE_URL=${{ env.ACTIONS_CACHE_URL }}
127+
ACTIONS_RUNTIME_TOKEN=${{ env.ACTIONS_RUNTIME_TOKEN }}
128+
CUDA_COMPUTE_CAP=80
129+
GIT_SHA=${{ env.GITHUB_SHA }}
130+
DOCKER_LABEL=sha-${{ env.GITHUB_SHA_SHORT }}
131+
tags: ${{ steps.meta-80.outputs.tags }}
132+
labels: ${{ steps.meta-80.outputs.labels }}
133+
cache-from: type=registry,ref=registry.internal.huggingface.tech/api-inference/text-embeddings-inference:cache-80,mode=max
134+
cache-to: type=registry,ref=registry.internal.huggingface.tech/api-inference/text-embeddings-inference:cache-80,mode=max
135+
136+
stop-runner:
137+
name: Stop self-hosted EC2 runner
138+
needs:
139+
- start-runner
140+
- build-and-push-image
141+
runs-on: ubuntu-latest
142+
env:
143+
AWS_REGION: us-east-1
144+
if: ${{ always() }} # required to stop the runner even if the error happened in the previous jobs
145+
steps:
146+
- name: Configure AWS credentials
147+
uses: aws-actions/configure-aws-credentials@v1
148+
with:
149+
aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
150+
aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
151+
aws-region: ${{ env.AWS_REGION }}
152+
- name: Stop EC2 runner
153+
uses: philschmid/philschmid-ec2-github-runner@main
154+
with:
155+
mode: stop
156+
github-token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
157+
label: ${{ needs.start-runner.outputs.label }}
158+
ec2-instance-id: ${{ needs.start-runner.outputs.ec2-instance-id }}

0 commit comments

Comments
 (0)