Anyscale Fine-Tuning
AnyscaleAnyscale Fine-Tuning is a managed, Ray-based platform for post-training open-weight language models. It runs LLaMA-Factory, SkyRL, and Ray Train across Ray clusters for continued pretraining, supervised fine-tuning, preference optimization (DPO, KTO, PPO, ORPO), and RLVR or agentic tuning, then serves the checkpoints with multi-LoRA support.
Anyscale's own fine-tuning library, LLMForge, no longer appears in its documentation, which now describes the platform through the open frameworks it runs.
Openness
1 high confidence- license
- proprietary(Anyscale-platform-only)
- source
- closed(managed SaaS platform, Terms of Service deliver Platform Services in object code format only and no Anyscale-authored fine-tuning implementation is published)
- built-on
- Ray+DeepSpeed+HF-Accelerate(OSS deps) but LLMForge itself not redistributable
- methods
- LoRA+full-FT
- status
- deprecated
Anyscale's fine-tuning platform is proprietary SaaS, delivered in object-code form only under its terms of service, with no Anyscale-authored fine-tuning implementation published. LLMForge, its own library, has been retired in favor of orchestrating open frameworks such as LLaMA-Factory, SkyRL and Ray Train, but the platform wrapping them stays closed either way.
- https://docs.anyscale.com/llm/fine-tuning recorded 2026-09-23
Describes Anyscale's current LLM post-training capability as delivered through the managed console orchestrating open frameworks on Ray clusters: the page states "Anyscale supports multiple frameworks for LLM post-training, including LLaMA-Factory, SkyRL, and Ray Train", names no Anyscale-authored library, and points at the Anyscale console for every workflow. A full-text search of the fetched page for "LLMForge" finds zero occurrences, where the score's prior evidence cited that library by name. No self-hostable Anyscale fine-tuning implementation is offered.
- https://www.anyscale.com/terms recorded 2026-09-23
Terms of Service defines the delivered product, "Platform Services", as Anyscale's own proprietary software-as-a-service platform, made available to customers in object code format only, with its associated SDKs and APIs likewise object-code-only.
Adoption
not assessedAnyscale publishes no usage figure for its fine-tuning feature, and the open frameworks it runs publish none that is specific to the platform, so the map leaves this product unscored.
- https://docs.anyscale.com/llm/fine-tuning recorded 2026-09-28
Searched the body for downloads, users and customers: none found, and the only "usage" is VRAM usage; the page describes methods, frameworks and integrations but publishes no usage figure for the fine-tuning capability, and states "Anyscale supports multiple frameworks for LLM post-training".
Capability
3 medium confidenceAnyscale runs full fine-tuning, LoRA and QLoRA, preference methods such as DPO, KTO and ORPO, and GRPO reinforcement learning by running open frameworks, LLaMA-Factory, SkyRL and Ray Train, on Ray clusters. That matches most of Axolotl's method coverage but not its quantization-aware training or sequence parallelism, which places it one step below Axolotl.
- https://docs.anyscale.com/llm/fine-tuning recorded 2026-09-28
The page lists post-training capabilities: "algorithms such as PPO, DPO, KTO, and ORPO, or RLVR with GRPO and DAPO" and "full fine-tuning and parameter-efficient methods (LoRA, QLoRA, freeze-tuning)", with scaling "compatible with FSDP, DeepSpeed, and Megatron".
- https://docs.anyscale.com/llm/fine-tuning recorded 2026-09-28
The page states "Anyscale supports multiple frameworks for LLM post-training, including LLaMA-Factory, SkyRL, and Ray Train" and cautions "Feature availability and exact configurations can vary by model family and framework."
Verified 2026-09-23