zeekayandHanzo Dev df832f2104 gym 0.1.1: cached decoding, gym synth, gym trainjob ...

Rollouts go through a KV cache (prompt prefilled once per group, one decode
step per token); cached logits match the full forward to 1e-4. GRPO runs end
to end on Metal — a mixed group at step 5 gives reward 0.75, loss −0.032,
gradient norm 1.19 — after the clipping reduction stopped going through f64,
which Metal does not have.

gym synth ports Data Designer's core: sampler, expression and LLM columns
(text, structured, judge) generated per row in dependency order against any
OpenAI-compatible endpoint, validators that drop or regenerate, jsonl or
parquet out.

gym trainjob is the Kubeflow TrainJob entry for the hosted fine-tuning API:
the broker's env contract becomes a Config; METHOD=lora is served, qlora and
full are refused by name. Dockerfile.cuda and image.yml build
ghcr.io/hanzoai/gym on the forge runners.

84 tests.

Co-authored-by: Hanzo Dev <[email protected]>

2026-09-01 10:49:00 -07:00

..

data

gym: LoRA SFT and GRPO on hanzo-ml, end to end

2026-09-01 02:28:36 -07:00

grpo

gym 0.1.1: cached decoding, gym synth, gym trainjob

2026-09-01 10:49:00 -07:00

model

gym 0.1.1: cached decoding, gym synth, gym trainjob

2026-09-01 10:49:00 -07:00

synth

gym 0.1.1: cached decoding, gym synth, gym trainjob

2026-09-01 10:49:00 -07:00

train

gym 0.1.1: cached decoding, gym synth, gym trainjob

2026-09-01 10:49:00 -07:00

config.rs

gym: LoRA SFT and GRPO on hanzo-ml, end to end

2026-09-01 02:28:36 -07:00

hub.rs

gym: LoRA SFT and GRPO on hanzo-ml, end to end

2026-09-01 02:28:36 -07:00

lib.rs

gym 0.1.1: cached decoding, gym synth, gym trainjob

2026-09-01 10:49:00 -07:00

main.rs

gym 0.1.1: cached decoding, gym synth, gym trainjob

2026-09-01 10:49:00 -07:00

trainjob.rs

gym 0.1.1: cached decoding, gym synth, gym trainjob

2026-09-01 10:49:00 -07:00