rnow 0.2.4__py3-none-any.whl

rnow/templates/tutorial-tool/config.yml

@@ -0,0 +1,27 @@
+project_id: ""
+project_name: "BrowseComp - Web Research Agent"
+dataset_id: ""
+dataset_name: "browsecomp"
+dataset_type: rl
+organization_id: ""
+data:
+  train_file: train.jsonl
+  batch_size: 32
+  group_size: 16
+model:
+  path: Qwen/Qwen3-8B
+  qlora_rank: 32
+  name: "BrowseComp Agent"
+  description: "Multi-turn RL model trained on BrowseComp web research tasks"
+algorithm:
+  loss_fn: ppo
+  adv_estimator: grpo
+  kl_penalty_coef: 0.01
+rollout:
+  max_turns: 5
+  max_tokens: 16384
+  termination_policy: last_tool
+trainer:
+  num_epochs: 4
+  learning_rate: 0.0001
+  save_step: 421

rnow/templates/tutorial-tool/env.py

@@ -0,0 +1,7 @@
+from rnow.core.tool import tool
+
+
+@tool
+def internet_search(query: str) -> dict:
+    """Search the web and return up to 5 results (title, link, snippet)."""
+    pass

rnow/templates/tutorial-tool/requirements.txt

@@ -0,0 +1,3 @@
+requests>=2.28.0
+beautifulsoup4>=4.12.0
+jellyfish>=1.0.0

rnow/templates/tutorial-tool/rewards.py

@@ -0,0 +1,7 @@
+from rnow.core import RewardArgs, reward
+
+
+@reward
+def accuracy(args: RewardArgs, messages: list) -> float:
+    """Check if the boxed answer matches the expected answer."""
+    pass