v0.1.11

What's Changed

• feat: change the default pack builder to builder-jammy-base by @jjleng in #127
• chore: version bump up by @jjleng in #128

Full Changelog: v0.1.10...v0.1.11

What's Changed

• feat: change the default pack builder to builder-jammy-base by @jjleng in #127
• chore: version bump up by @jjleng in #128

Full Changelog: v0.1.10...v0.1.11

v0.1.10

What's Changed

• refactor: disable image pull always for function containers by @jjleng in #125
• chore: bump up version by @jjleng in #126

Full Changelog: v0.1.9...v0.1.10

v0.1.9

What's Changed

• feat: options to set env, volume mounts and probes for model group containers by @jjleng in #101
• feat: options to pass in env vars for functions by @jjleng in #102
• feat: be able to display public and private endpoints for MG by @jjleng in #103
• chore: upgrade prometheus chart by @jjleng in #104
• feat: default to openai compatible server for vllm by @jjleng in #105
• feat: run vLLM with a served model name by @jjleng in #106
• fix: make prometheus scrape the inferrence engine's metrics correctly by @jjleng in #107
• feat: set knative resource requests by @jjleng in #108
• feat: dedicated node groups for functions and job workers by @jjleng in #109
• feat: be able to pass resource requests when creating jobs by @jjleng in #110
• feat: detect the right ami types for different node types by @jjleng in #111
• feat: command to push a prebuilt image to the container registry by @jjleng in #112
• refactor: make the config fields naming consistent by @jjleng in #113
• feat: enable host ipc for multiple GPU inference by @jjleng in #114
• docs: rewrite README by @jjleng in #115
• docs: add sample model launching templates by @jjleng in #116
• docs: updates to the invoice_extraction example by @jjleng in #117
• docs: update the website_rag example by @jjleng in #118
• feat: gateway vpc endpoint for s3 by @jjleng in #119
• feat: default to have at least one funciton instance running by @jjleng in #120
• docs: add faq by @jjleng in #121
• docs: detailed descriptions of the cluster config yaml by @jjleng in #122
• docs: quick start doc by @jjleng in #123
• chore: version bump by @jjleng in #124

Full Changelog: v0.1.8...v0.1.9

v0.1.8

What's Changed

• feat: support inference with spot instances by @jjleng in #86
• feat: support mixed model groups by @jjleng in #87
• feat: flag to make model groups not exposed to the public by @jjleng in #88
• feat: heuristics for caculating the memory, cpu and gpu resource requests by @jjleng in #89
• fix: typing issues in e2e test code by @jjleng in #90
• feat: command to list function revisions by @jjleng in #91
• feat: command to split traffic among revisions by @jjleng in #92
• refactor: make typer command args more consistent by @jjleng in #93
• refactor: save kubeconfig as soon as cluster is provisioned by @jjleng in #94
• feat: options to pass resource requests and limits when creating fuctions by @jjleng in #95
• feat: utilize multiple GPUs for vllm inferrence by @jjleng in #96
• feat: utility to parse gguf metadata by @jjleng in #97
• fix: permissions to allow cluster autoscaler scale from 0 by @jjleng in #98
• chore: bump up version by @jjleng in #99
• chore: add homepage info to pyproject by @jjleng in #100

Full Changelog: v0.1.7...v0.1.8

v0.1.7

What's Changed

• refactor: windows compatibility by @jjleng in #68
• feat: install kubectl as a dependency for pulumi by @jjleng in #69
• docs: aws CLI is required by @jjleng in #70
• tests: parse the cluster config yamls of examples by @jjleng in #71
• refactor: harden the config code and files by @jjleng in #72
• feat: cleanup stabled model group resources by @jjleng in #73
• refactor: change opt-in to opt-out for kubeconfig update by @jjleng in #74
• feat: command to list clusters being managed by @jjleng in #75
• feat: command to switch current cluster by @jjleng in #76
• feat: remove local cluster state when a cluster is taken down by @jjleng in #77
• refactor: get rid of local cluster state by @jjleng in #78
• fix: make removing crd finalizers more stable by @jjleng in #79
• feat: turn on flash attention for llama.cpp by default by @jjleng in #80
• fix: fix misc issues by @jjleng in #81
• fix: reattempt to delete CRD finalizers in a separate thread by @jjleng in #82
• feat: support vLLM as inference runtime by @jjleng in #83
• feat: check for config version forward compatibility by @jjleng in #84
• chore: bump up version by @jjleng in #85

Full Changelog: v0.1.6...v0.1.7

v0.1.6

What's Changed

• refactor: restructured code base by @jjleng in #51
• fix: print messages after closing progress bar by @jjleng in #52
• refactor: get rid of hard coded model mount path in containers by @jjleng in #53
• feat: remove k8s resources related to a staled model group by @jjleng in #54
• fix: make llama.cpp save hf files to addressable local paths by @jjleng in #55
• fix: the hf model file path problem of the llama.cpp runtime by @jjleng in #56
• tests: e2e tests with KinD by @jjleng in #57
• ci: add e2e to ci by @jjleng in #58
• refactor: make pack be installed to a non-global location by @jjleng in #59
• docs: get rid of the pack installation info from readme files by @jjleng in #60
• feat: avoid mutating the model stored in model store by @jjleng in #61
• refactor: remove aws CLI as a dependency when pushing images to ECR by @jjleng in #62
• refactor: make the code removing k8s finalizers more robust when taki… by @jjleng in #63
• refactor: make docker login more secure by @jjleng in #64
• Isolate paka's pulumi from system pulumi by @jjleng in #65
• docs: doc updates by @jjleng in #66
• chore: version bump by @jjleng in #67

Full Changelog: v0.1.5...v0.1.6

v0.1.5

What's Changed

• feat: add a new model registry by @jjleng in #42
• test: fix test_registry by @jjleng in #44
• Add new workflow. by @erika-tsay in #45
• refactor: remove runtime and devices fields from model class by @jjleng in #46
• feat: llama.cpp runtime that supports the new model abstraction by @jjleng in #47
• refactor: make default llama.cpp params more conservative by @jjleng in #48
• fix: correct the wrong dependency in the invoice example by @jjleng in #49
• chore: bump version by @jjleng in #50

New Contributors

• @erika-tsay made their first contribution in #45

Full Changelog: v0.1.4...v0.1.5

v0.1.4

What's Changed

• chore: fix mypy errors by @jjleng in #37
• test: replace patch with patch.object for eaiser error discovery by @jjleng in #38
• chore: pin pulumi to a specific version that honors depends_on by @jjleng in #39
• feat: command to sync cluster state by @jjleng in #40
• chore: bump version by @jjleng in #41

Full Changelog: v0.1.3...v0.1.4

v0.1.3

What's Changed

• docs: instructions for installing the pack CLI by @jjleng in #22
• docs: add pulumi CLI as a dependency by @jjleng in #23
• Add HuggingFaceModel, HttpSouceModel by @SoftewareArtist in #24
• fix: fix failing hf model tests by @jjleng in #32
• Refactor the model and model store abstraction by @jjleng in #33
• chore: make paka compatible with py3.8 and above by @jjleng in #34
• chore: remove CHANGELOG.md by @jjleng in #35
• chore: bump version by @jjleng in #36

Full Changelog: v0.1.2...v0.1.3

v0.1.2

What's Changed

• docs: example for invoice extraction by @jjleng in #13
• Add GPU (CUDA) support by @jjleng in #15
• docs: update README with the GPU support message by @jjleng in #16
• docs + GPU inference for the invoice extraction example by @jjleng in #17
• feat: remove finalizers before tearing down a cluster by @jjleng in #18
• chore: bump version to 0.1.2 by @jjleng in #19

Full Changelog: v0.1.1...v0.1.2