brainwires-finetune 0.11.0

Cloud fine-tune APIs (OpenAI, Anthropic, Together, Fireworks, Anyscale, Bedrock, Vertex AI) + dataset pipelines for the Brainwires Agent Framework. Local LoRA/QLoRA/DoRA fine-tuning and training-from-scratch live in the sibling `rullama` workspace (`rullama-finetune`, `rullama-training`).
Documentation

brainwires-finetune

Crates.io Documentation License

Cloud fine-tune APIs and dataset pipelines for Brainwires agents.

Fine-tuning surface:

  • brainwires-finetune (this crate) — cloud fine-tune APIs (OpenAI / Anthropic / Together / Fireworks / Anyscale / Bedrock / Vertex AI) plus dataset pipelines.
  • rullama-finetune (sibling rullama workspace) — local PEFT (LoRA / QLoRA / DoRA), Burn-backed. (Lived in this workspace as a separate local-PEFT crate prior to v0.11; moved out alongside the rest of the wgpu inference engine.)
  • rullama-training (sibling rullama workspace) — placeholder for actual training-from-scratch.

What lives here

  • manager::TrainingManager — dispatches fine-tune jobs to whichever provider implements FineTuneProvider.
  • cloud::FineTuneProvider + FineTuneProviderFactory — provider-agnostic trait + factory.
  • cloud::providers (one module per cloud API) — concrete impls.
  • config — hyperparameter / adapter / alignment-method types shared with rullama-finetune.
  • datasets — JSONL / format conversion / tokenization / dedup (absorbed from the deprecated brainwires-datasets crate).
  • error::TrainingError, types::{TrainingJobId, TrainingJobStatus, ...} — shared infrastructure.

Features

Feature Default Notes
cloud yes reqwest-based cloud provider clients
bedrock no AWS Bedrock fine-tune (sigv4)
vertex no Google Vertex AI (gcp_auth)
datasets-hf-tokenizer no HuggingFace tokenizers
datasets-tiktoken no OpenAI tiktoken
datasets-dedup no sha2 + rand for content dedup
datasets-full no All three datasets sub-features
full no cloud + bedrock + vertex + datasets-full

Usage

[dependencies]
brainwires-finetune = "0.11"

use brainwires_finetune::{TrainingManager, CloudFineTuneConfig};

let manager = TrainingManager::new(/* ... */);
let job = manager.submit(CloudFineTuneConfig { /* ... */ }).await?;

See also

  • rullama-finetune (sibling rullama workspace) — local PEFT (LoRA / QLoRA / DoRA), Burn-backed. Reuses this crate's shared config / error / types modules.
  • brainwires-provider — LLM chat clients (separate crate).
  • brainwires — umbrella facade with training / training-cloud features (cloud only since v0.11).

License

MIT OR Apache-2.0