mirror of
https://github.com/hanzoai/gym.git
synced 2026-09-30 22:37:59 +00:00
zeekayandHanzo Dev
df832f2104
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]>
63 lines
1.7 KiB
TOML
63 lines
1.7 KiB
TOML
[package]
|
|
name = "hanzo-gym"
|
|
version = "0.1.1"
|
|
edition = "2021"
|
|
rust-version = "1.85"
|
|
license = "Apache-2.0"
|
|
description = "Hanzo Gym — LoRA fine-tuning and GRPO for language models, in Rust on hanzo-ml"
|
|
repository = "https://github.com/hanzoai/gym"
|
|
keywords = ["llm", "fine-tuning", "lora", "grpo", "training"]
|
|
categories = ["science"]
|
|
|
|
[lib]
|
|
name = "gym"
|
|
path = "src/lib.rs"
|
|
|
|
[[bin]]
|
|
name = "gym"
|
|
path = "src/main.rs"
|
|
|
|
[dependencies]
|
|
hanzo-ml = "0.11.95"
|
|
hanzo-nn = "0.11.39"
|
|
tokenizers = { version = "0.21", default-features = false, features = ["onig"] }
|
|
minijinja = { version = "2", features = ["builtins", "json", "loader"] }
|
|
hf-hub = { version = "0.4", default-features = false, features = ["ureq", "rustls-tls"] }
|
|
safetensors = "0.7"
|
|
serde = { version = "1", features = ["derive"] }
|
|
serde_json = "1"
|
|
serde_yaml = "0.9"
|
|
clap = { version = "4", features = ["derive"] }
|
|
anyhow = "1"
|
|
thiserror = "2"
|
|
rand = "0.9"
|
|
rand_distr = "0.5"
|
|
regex = "1"
|
|
indicatif = "0.18"
|
|
tracing = "0.1"
|
|
tracing-subscriber = { version = "0.3", features = ["env-filter"] }
|
|
ureq = { version = "3", features = ["json"] }
|
|
memmap2 = "0.9"
|
|
half = "2"
|
|
parquet = { version = "58", default-features = false, features = ["json", "snap", "zstd", "lz4", "flate2-zlib-rs", "brotli"] }
|
|
minijinja-contrib = { version = "2", features = ["pycompat"] }
|
|
chrono = "0.4"
|
|
jsonschema = { version = "0.49", default-features = false }
|
|
|
|
[features]
|
|
default = []
|
|
cuda = ["hanzo-ml/cuda", "hanzo-nn/cuda"]
|
|
cudnn = ["cuda", "hanzo-ml/cudnn"]
|
|
metal = ["hanzo-ml/metal", "hanzo-nn/metal"]
|
|
accelerate = ["hanzo-ml/accelerate", "hanzo-nn/accelerate"]
|
|
mkl = ["hanzo-ml/mkl", "hanzo-nn/mkl"]
|
|
|
|
[profile.release]
|
|
lto = "thin"
|
|
codegen-units = 1
|
|
|
|
[workspace]
|
|
|
|
[dev-dependencies]
|
|
hanzo-transformers = "0.11.9"
|