togethercomputer/together-go v0.13.0

v0.13.0

Repository: togethercomputer/together-go

Tag: v0.13.0

Published: 2026-10-22T19:17:48Z

Prerelease: yes

Release notes:

0.13.0 (2026-10-21)

⚠ BREAKING CHANGES

• remove code interpreter APIs
• rl: collapse forward into forward-backward

Features

• add canonical NVIDIA version selector (1ca46d2)
• add deployment capacity type (a6f2589)
• add DPPO loss to RL OpenAPI (09487c5)
• add rl checkpoint list endpoint (115bc68)
• add RL model resource queue position (6013cbb)
• Add Rollouts SDK code for endpoint deployments management (39214e5)
• clusters: Promote GPU clusters off beta! (0924a58)
• containers: expose deployment model mounts field (e4e64ea)
• deployments: add deployment compliance policy for hipaa requirements (6d34197)
• deployments: Add inactiveTimeout to deployments (#644) (3b49f97)
• deployments: expose autoscaling policies and reset semantics (MLE-7727) (151b622)
• document shaping calibration files (ba9cb5f)
• endpoint: Add adapter validation status fields (9d86afa)
• endpoints: add supported model adapter mode field (18c22f2)
• endpoints: expose deployment concurrency limit response (3e8937f)
• endpoints: List enum values for endpoint scaling metric values (ff0435e)
• expose RL checkpoint registry artifacts (04957a3)
• expose RL checkpoint registry ids in OpenAPI (21d2485)
• expose RL GPU configurations in OpenAPI (c347bc7)
• expose RL session policy state (b4a94ee)
• fine-tuning: sync tokenized dataset download OpenAPI (1868d94)
• fix: sync RL checkpoint OpenAPI schemas; feat: document shaping calibration files; feat: expose RL checkpoint registry artifacts; internal: Duplicate RL spec to non-beta to prep migration; Internal: Remove beta duplicate config for RL; feat: add RL model resource queue position; fix: sync adapter attachment ids in OpenAPI; fix(docs): remove substrate wording from GPU cluster descriptions (db62f30)
• fix: sync RL checkpoint OpenAPI schemas; feat: document shaping calibration files; feat: expose RL checkpoint registry artifacts; internal: Duplicate RL spec to non-beta to prep migration; Internal: Remove beta duplicate config for RL; feat: add RL model resource queue position; fix: sync adapter attachment ids in OpenAPI; fix(docs): remove substrate wording from GPU cluster descriptions; fix: stop offering H100-80GB for RL model resources (30ed437)
• fix: sync RL checkpoint OpenAPI schemas; feat: document shaping calibration files; feat: expose RL checkpoint registry artifacts; internal: Duplicate RL spec to non-beta to prep migration; Internal: Remove beta duplicate config for RL; feat: add RL model resource queue position; fix: sync adapter attachment ids in OpenAPI; fix(docs): remove substrate wording from GPU cluster descriptions; fix: stop offering H100-80GB for RL model resources; MOSH-5642: rename prepare-for-fp4-inference adapter_* fields to model_* (e21a18e)
• jig: add deployment revision and rollback APIs (afa0a0a)
• jig: add S3 volume origins (f5a740e)
• openapi: add B300 GPU cluster type (1bd70a4)
• openapi: add RL LoRA train_unembed (cf8dac5)
• openapi: add RL prompt cache hit tokens (ad60dc8)
• openapi: add RL routing replay fields (b6dfcc0)
• openapi: add RL weights sync operation...