ai4-papi - v1.3.0

1.3.0 (2025-08-13)

Features

• adapt to new node type metadata (9d77872)
• add ai4llm proxy endpoint (#109) (a91ddb1)
• add async support for OSCAR (#72) (387a21f)
• add code carbon variable to fedserver (#97) (ead0838)
• add current PAPI branch+commit in the docs' description (c5bca42)
• add custom username to llm tool (#105) (46a2a0e)
• add email notification when slow deployment times (#60) (22b066c)
• add FedAvgOpt strategy (#127) (975fce5)
• add item id to the metadata dict (d535086)
• add LLM catalog (#103) (53618f0)
• add LLM tool (#89) (f39d2c4)
• add metric privacy variable to federated server (#95) (d06d0a6)
• add MinIO info to service detail (#125) (77ef5d4)
• add support for CVAT tool (#73) (8cc62ab)
• add support for OSCAR services (#25) (5954569)
• add UI to Nomad inference endpoints (#76) (429c726)
• AI4OS Development Environment as a tool (#111) (7a0497d)
• allow also URLs for dataset downloads (#65) (17563f5)
• allow listing/deleting try-me (#68) (6fa977a)
• allow to deploy as Nomad job (#101) (3501be5)
• be conservative with tryme resource cap (4026f7a)
• catalog: point FL server to new repo (d5075cc)
• config thinking for deepseek (#98) (3bbbbac)
• disable some code sections if we are doing local development (a8b81a9)
• disallow usage of GPUs to CPU-only jobs (#96) (39383ab)
• enforce docker image to belong to either deephdc or ai4oshub DockerHub orgs (or our Harbor) (78fe123)
• enforce minimum resources for user's metadata (#110) (9158100)
• improve imports (45475de)
• increase metadata cache expiration (88fc55d)
• increase RAM for fedserver (028b28e)
• increase snapshot quota to 15GB (aedc058)
• inject MLflow credentials in Nomad jobs (a759d4d)
• metadata: support MLDCAT-AP profile (#108) (e0fbfa8)
• migrate to keycloak (#117) (38e5f98)
• migrate to new metadata (#63) (14146a5)
• overwrite some metadata with Github info (6964cbe)
• remove custom domain field from configuration forms (#67) (9a23f87)
• remove deprecated tags method (e9f87aa)
• remove trailing slash in endpoints (#64) (1808741)
• start using prevent_reschedule_on_lost (0fb6b5a)
• start using custom UI for ai4life loader (081c04a)
• start using CVAT images from ai4oshub Dockerhub (4ee60d1)
• stats: keep track of reallocations (982767d)
• support AI4Life models as a tool (#86) (9502b50)
• support AI4Life project (vo.ai4life.eu) (#57) (f183423)
• support deleting CVAT snapshots from storage (#77) (695e7a3)
• support docker tags and custom port in ai4life tool (#90) (0aee8f3)
• support making snapshots from deployments (#62) (b49d07d)
• support nomad batch jobs (#114) (70c368b)
• support Nomad info queries from the provenance pipeline (#75) (a3cf0aa)
• support periodic CVAT backups (#115) (8b0adda)
• support refreshing metadata from CI/CD (#70) (233b38f)
• support refreshing only the catalog index (83983e3)
• support try-me endpoints in Nomad (#59) (de9b373)
• update PAPI version in OpenAPI display (382af38)
• update parameters for Fedserver (Differential Privacy) (#66) (beec1cb)

Bug Fixes

• adapt MLFLOW_URI depending on VO (ee39a1b)
• adapt tryme check to new nomad nomenclature (230e281)
• add more statuses when allowing purging (5b4c67f)
• allow purging stuck Nomad jobs (status=queued) (21be7a2)
• allow uppercase module names (f2938c4)
• always use latest metadata version to validate (ed2e5bd)
• avoid checking for vo membership in get_user_info (9c0243f)
• better catch errors in storage/ls (649f284)
• better catch missing tryme resources (dda2ff3)
• change found status code (a99746e)
• check vllm is ready before starting LLM tool (#99) (56565dc)
• correct some references to old federated server (61204dc)
• deployments: allow purging jobs with main task dead (765af52)
• do not enforce Harbor password when developing (8060576)
• fix breaking change in harborapi (d9c981a)
• fix CI/CD link for tools (d6cacfb)
• fix Dockerhub link (e950cb9)
• fix error retrieving status for queued jobs (c2a3fc6)
• fix incorrect task names (13024bd)
• fix MinIO info (02646c4)
• fix refresh (484fff3)
• fix retrieval of Github license (d8a3161)
• fix storage rm function (0a524f1)
• handle case where user does not belong to VO supported by the project (d4a2dbc)
• handle case where user does not belong to VO supported by the project (4f67727)
• improve get_metadata() logic (772dbcd)
• inference UI should not be killed after 10 mins (ec41a7e)
• job reschedule on lost (#123) (e6ac106)
• metadata: update to new Jenkins domain (16950af)
• minor tweaks in minimum resources (2158aa4)
• more accurate deployment status (5626029)
• obscure rclone password on behalf of users (3612ae6)
• only parse ready nodes (6e3fa32)
• oscar: fix deepaas version retrieval (ba6e0fc)
• oscar: remove typo (1e03571)
• properly cache cluster stats endpoint (b7f2fb8)
• recover zero-reschedule policy (b11b0c7)
• remove "hub" from Docker image metadata value (cb9b07c)
• remove references to old deep-oc-generic-dev (9129dbd)
• rename proxies routes (#128) (f78ad89)
• restore OSCAR service sorting (4db6659)
• return 404 if directory does not exist (#100) (455084b)
• secrets for each VO should be stored in different paths (#71) (111f20f)
• set email task to force pull docker image (#74) (2c6f2f8)
• snapshot error codes (#88) (a867b92)
• stats compute should run on the background (e9ac53d)
• stats: account for federated cluster migration (ab1d208)
• stats: allow initializing cluster stats when PAPI is used as package (66898f9)
• stats: fix node status (0accdb5)
• stats: return reserved disk (7b2ed94)
• support OSCAR in ai4life (f2d026e)
• update refresh endpoint (47a6bb5)
• upgrade CVAT from v2.7.3 to v2.25.0 (#85) (7e930e9)
• upgrade to CVAT v2.28.0 (#87) (b8a3582)

Performance Improvements

• use requests session (119d28d)

Documentation

• add oidc agent configuration with keycloak (c793a1b)
• change route titles in the Swagger UI (72fb3d0)
• change route titles in the Swagger UI (44afe9f)
• update README (94aa9ab)
• update readme (#122) (c754a5a)

- Python
Published by github-actions[bot] 7 months ago

ai4-papi - v1.2.0

1.2.0 (2024-08-05)

Features

• add also vscode for old dev-env (backward compatibility) (4ec9c31)
• add support for downloading datasets (#53) (55f6b77)
• better tag sorting for dev env (feb2883)
• module migration to ai4os-hub (#51) (87473e0)
• move to federated cluster (#56) (6355cd6)
• stats: account for ineligible nodes (6560a10)
• stats: properly aggregate cluster resources (fd00d14)
• update conf for deep-oc-federated-server (#55) (36082a7)

Bug Fixes

• better catch exception (4133723)
• reenable dataset checks (#54) (af9e8eb)
• stats: account for failing GPU nodes (93a1608)
• stats: move stat in to if loop (c056894)
• temporarily disable DOI checks (d6a1599)
• zenodo: properly handle params=None (af090c1)

Documentation

• update README (30133ba)

- Python
Published by IgnacioHeredia over 1 year ago

ai4-papi - v1.1.0

1.1.0 (2024-05-15)

Features

• add support for retrieving stats (bc19170)
• add CORS for new endpoints (9f7ce1f)
• add support for Vault secrets (#44) (11116ec)

Bug Fixes

• force pulling of Docker images (c811bba)
• only retrieve GPU models from eligible nodes (3733159)
• properly monkey-patch Catalog class using MethodType (ce8156b)
• set license year/owner (ecbcde7)
• set max RAM memory (39a1384)

- Python
Published by IgnacioHeredia almost 2 years ago

ai4-papi - v1.0.0

1.0.0 (2024-01-30)

⚠ BREAKING CHANGES

• change main endpoint
• create separate routes for tools

Features

• add cpu_MHz (ce1d74a)
• add active endpoints (35f53f0)
• add checks for JWT scopes (c71e918)
• add datacenter to deployment info (7773a02)
• add federated token as env variable (1646bcc)
• add monitor port to fedserver tool to enable ttyd (c849fe4)
• add name and email to Nomad jobs (8743e31)
• add release-please support (16f17c3)
• add storage task (0efa70f)
• allow SSL in deployments (6a3857d)
• auto-discover available gpu models (d31e3f5)
• create separate routes for tools (6fd0fc5)
• deployment creation uses string.Template (ed79f27)
• disable custom domain, leave custom host (1771a31)
• implement total GPUs quota per user (55baab8)
• remove proxy + let's encrypt (d7c0ef0)
• replace deepaas with api in Nomad jobs (da764eb)
• return requested resources for queued jobs (7037e4e)
• update to new Nextcloud instance (e2c11e7)

Bug Fixes

• gpu_num for multi-gpu deployments (4b39af1)
• module_name parsing (be3e502)
• add back /ui to api endpoint (205daa7)
• allow requests from ai4os-proxy (f09a07d)
• avoid restarting jobs when network is temporarily lost (49bc1e6)
• better error catching (9894fbc)
• better hardware limit (8d9a30a)
• change lowercase (de8630d)
• check in gpu_model affinity is empty (2fabe18)
• comment cpu_MHz (d58f7d7)
• disable Nomad storage tasks if credentials not provided (de3783d)
• docker tags pagination (929325c)
• endpoints retrieval (7aeea2b)
• GPU modelnames (4e37c55)
• ignore user disk (c2ba8aa)
• increase shared memory limit in Docker (c5949cd)
• parse modules from .gitmodules file instead of YAML (b38f11b)
• pyyaml version (e69422c)
• remove gpu_model constraint if model is not specified (3d11fdb)
• set cpu_num to cores, not to MHz (da9186b)
• update Docker compose files (487be86)
• use constraint instead of affinity for GPU models (897ac78)

Performance Improvements

• improve check_domain (643aed6)
• improve job retrieval with Nomad filters (a21272d)

Code Refactoring

• change main endpoint (2558b4b)

- Python
Published by IgnacioHeredia about 2 years ago