zeekayandHanzo Dev dd4556ec1c gym: LoRA SFT and GRPO on hanzo-ml, end to end ...

model: one trainable Llama-family decoder (Qwen3 with q/k norm, Llama) built
from differentiable hanzo-nn primitives, LoRA on the projections the config
names, PEFT adapter save/load/merge; logits match hanzo-transformers to 1e-4.
data: chat_template / alpaca / completion with Axolotl's label semantics
(dummy-render turn spans, train_on_eos), jsonl/json/parquet, HF datasets and
split grammar, seeded shuffle, validation tail, GRPO prompts, collate.
train: chunked masked cross-entropy, warmup + cosine/linear/constant,
accumulation, global-norm clipping, AdamW, checkpoints, eval.
grpo: the hanzo-training core ported unchanged in its math, with LmPolicy
over CausalLm (batched masked scoring), exact/numeric/format/length/http
verifiers.

61 tests. gradient_checkpointing, QLoRA, full fine-tune, multi-GPU and the
DPO family are rejected at config load rather than approximated.

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

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

..

alpaca.rs

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

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

chat_template.rs

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

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

completion.rs

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

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

mod.rs

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

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

source.rs

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

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

text.rs

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

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