name = "humanize-p5050-qwen35-2b-base-full400-env0315-r1"
model = "Qwen/Qwen3.5-2B"
max_steps = 400
batch_size = 64
rollouts_per_example = 8
max_inflight_rollouts = 32
learning_rate = 8e-5
lora_alpha = 32

[sampling]
max_tokens = 1024
temperature = 0.7
enable_thinking = false

[[env]]
id = "jayshah5696/humanize-rl-env"
name = "train_mix_v2_p5050_qwen35_2b_base_full400_env0315_r1"
version = "0.3.15"
args = { split = "train", task_set = "mix_v2_p5050", reward_mode = "p50_50_no_penalty" }

[eval]
interval = 50
num_examples = 64
rollouts_per_example = 1
eval_base_model = true

[eval.sampling]
max_tokens = 1024
temperature = 0.2
enable_thinking = false

[[eval.env]]
id = "jayshah5696/humanize-rl-env"
name = "eval_mix_v2_p5050_env0315"
version = "0.3.15"
args = { split = "validation", task_set = "mix_v2_p5050", reward_mode = "p50_50_no_penalty" }

[[eval.env]]
id = "jayshah5696/humanize-rl-env"
name = "eval_v02_strict_env0315"
version = "0.3.15"
args = { split = "validation", task_set = "v02_smoke", reward_mode = "strict" }

[[eval.env]]
id = "jayshah5696/humanize-rl-env"
name = "eval_v03_strict_env0315"
version = "0.3.15"
args = { split = "validation", task_set = "v03", reward_mode = "strict" }

[val]
num_examples = 64
rollouts_per_example = 1
interval = 20

[checkpoints]
interval = 50
keep_cloud = 8

[adapters]
interval = 50
keep_last = 8

[infrastructure]
compute_size = "L"