Add files using upload-large-folder tool

saves_hf/qwen2.5_3B_it_rubikscube3_withthink_sas_rl/global_step_200/qwen2.5_3b_actor_hf/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.05,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.57.3"
+}

saves_hf/qwen2.5_3B_it_rubikscube3_withthink_sas_rl/global_step_200/qwen2.5_3b_actor_hf/model.safetensors.index.json

@@ -0,0 +1,443 @@
+{
+  "metadata": {
+    "total_parameters": 3397103616,
+    "total_size": 6794207232
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00002-of-00002.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.1.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.1.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.12.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.13.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.14.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.17.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.17.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.2.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.3.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.33.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.33.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.34.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.8.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.norm.weight": "model-00001-of-00002.safetensors"
+  }
+}

saves_hf/qwen2.5_3B_it_rubikscube3_withthink_sas_rl/global_step_200/qwen2.5_3b_actor_hf/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

saves_hf/qwen2.5_3B_it_rubikscube3_withthink_sas_rl/global_step_200/qwen2.5_3b_actor_hf/tokenizer_config.json

@@ -0,0 +1,208 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

saves_hf/qwen2.5_3B_it_sokoban1_withthink_sa_rl/global_step_200/qwen2.5_3b_actor_hf/chat_template.jinja

@@ -0,0 +1,54 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- messages[0]['content'] }}
+    {%- else %}
+        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}
+    {%- endif %}
+    {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
+    {%- else %}
+        {{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- for message in messages %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role }}
+        {%- if message.content %}
+            {{- '\n' + message.content }}
+        {%- endif %}
+        {%- for tool_call in message.tool_calls %}
+            {%- if tool_call.function is defined %}
+                {%- set tool_call = tool_call.function %}
+            {%- endif %}
+            {{- '\n<tool_call>\n{"name": "' }}
+            {{- tool_call.name }}
+            {{- '", "arguments": ' }}
+            {{- tool_call.arguments | tojson }}
+            {{- '}\n</tool_call>' }}
+        {%- endfor %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

saves_hf/qwen2.5_3B_it_sokoban1_withthink_sa_rl/global_step_200/qwen2.5_3b_actor_hf/config.json

@@ -0,0 +1,66 @@
+{
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 32768,
+  "max_window_layers": 70,
+  "model_type": "qwen2",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 36,
+  "num_key_value_heads": 2,
+  "pad_token_id": 151643,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "transformers_version": "4.57.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

saves_hf/qwen2.5_3B_it_sokoban1_withthink_sas_rl/global_step_200/qwen2.5_3b_actor_hf/model.safetensors.index.json

@@ -0,0 +1,443 @@
+{
+  "metadata": {
+    "total_parameters": 3397103616,
+    "total_size": 6794207232
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00001-of-00002.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.0.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.0.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.10.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.10.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.11.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.12.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.12.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.14.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.15.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.16.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.17.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.17.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.18.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.18.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.2.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.2.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.32.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.33.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.34.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.34.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.35.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.35.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.4.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.5.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.5.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.norm.weight": "model-00001-of-00002.safetensors"
+  }
+}

saves_hf/qwen2.5_3B_it_sokoban1_withthink_sas_rl/global_step_200/qwen2.5_3b_actor_hf/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

saves_hf/qwen2.5_3B_it_sokoban1_withthink_sas_rl/global_step_200/qwen2.5_3b_actor_hf/tokenizer_config.json

@@ -0,0 +1,208 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

saves_hf/qwen2.5_3B_it_sokoban2_withthink_fulltraj_sa_rl/global_step_200/qwen2.5_3b_actor_hf/added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

saves_hf/qwen2.5_3B_it_sokoban2_withthink_fulltraj_sa_rl/global_step_200/qwen2.5_3b_actor_hf/chat_template.jinja

@@ -0,0 +1,54 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- messages[0]['content'] }}
+    {%- else %}
+        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}
+    {%- endif %}
+    {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
+    {%- else %}
+        {{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- for message in messages %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role }}
+        {%- if message.content %}
+            {{- '\n' + message.content }}
+        {%- endif %}
+        {%- for tool_call in message.tool_calls %}
+            {%- if tool_call.function is defined %}
+                {%- set tool_call = tool_call.function %}
+            {%- endif %}
+            {{- '\n<tool_call>\n{"name": "' }}
+            {{- tool_call.name }}
+            {{- '", "arguments": ' }}
+            {{- tool_call.arguments | tojson }}
+            {{- '}\n</tool_call>' }}
+        {%- endfor %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

saves_hf/qwen2.5_3B_it_sokoban2_withthink_fulltraj_sa_rl/global_step_200/qwen2.5_3b_actor_hf/config.json

@@ -0,0 +1,66 @@
+{
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 32768,
+  "max_window_layers": 70,
+  "model_type": "qwen2",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 36,
+  "num_key_value_heads": 2,
+  "pad_token_id": 151643,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "transformers_version": "4.57.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

saves_hf/qwen2.5_3B_it_sokoban2_withthink_fulltraj_sa_rl/global_step_200/qwen2.5_3b_actor_hf/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.05,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.57.3"
+}

saves_hf/qwen2.5_3B_it_sokoban2_withthink_fulltraj_sa_rl/global_step_200/qwen2.5_3b_actor_hf/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

saves_hf/qwen2.5_3B_it_sokoban2_withthink_fulltraj_sa_rl/global_step_200/qwen2.5_3b_actor_hf/tokenizer_config.json

@@ -0,0 +1,208 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

saves_hf/qwen2.5_3B_it_sokoban2_withthink_sa_rl/global_step_200/qwen2.5_3b_actor_hf/chat_template.jinja

@@ -0,0 +1,54 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- messages[0]['content'] }}
+    {%- else %}
+        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}
+    {%- endif %}
+    {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
+    {%- else %}
+        {{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- for message in messages %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role }}
+        {%- if message.content %}
+            {{- '\n' + message.content }}
+        {%- endif %}
+        {%- for tool_call in message.tool_calls %}
+            {%- if tool_call.function is defined %}
+                {%- set tool_call = tool_call.function %}
+            {%- endif %}
+            {{- '\n<tool_call>\n{"name": "' }}
+            {{- tool_call.name }}
+            {{- '", "arguments": ' }}
+            {{- tool_call.arguments | tojson }}
+            {{- '}\n</tool_call>' }}
+        {%- endfor %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

scripts/eval_qwen_sokoban.sh

@@ -0,0 +1,69 @@
+#!/bin/bash
+# Evaluate Qwen model on Sokoban environment
+#
+# Usage:
+#   bash scripts/eval_qwen_sokoban.sh <model_version> [num_trajectories] [gpu_id]
+#
+# Examples:
+#   bash scripts/eval_qwen_sokoban.sh 3B 128 0
+#   bash scripts/eval_qwen_sokoban.sh 7B 256 1
+#   bash scripts/eval_qwen_sokoban.sh 14B 128 0,1
+
+set -e
+
+# Parse arguments
+MODEL_VERSION=${1:-"14B"}
+NUM_TRAJECTORIES=${2:-128}
+GPU_ID=${3:-"0"}
+
+# Calculate env_groups and group_size
+# Strategy: Use group_size=16 (common default), calculate groups accordingly
+GROUP_SIZE=16
+ENV_GROUPS=$((NUM_TRAJECTORIES / GROUP_SIZE))
+
+if [ $((NUM_TRAJECTORIES % GROUP_SIZE)) -ne 0 ]; then
+    echo "Warning: NUM_TRAJECTORIES ($NUM_TRAJECTORIES) is not divisible by GROUP_SIZE ($GROUP_SIZE)"
+    echo "Rounding up to $((ENV_GROUPS * GROUP_SIZE)) trajectories"
+    ENV_GROUPS=$(( (NUM_TRAJECTORIES + GROUP_SIZE - 1) / GROUP_SIZE ))
+fi
+
+ACTUAL_TRAJECTORIES=$((ENV_GROUPS * GROUP_SIZE))
+
+# Model configuration
+MODEL_PATH="Qwen/Qwen2.5-${MODEL_VERSION}-Instruct"
+OUTPUT_DIR="outputs"
+OUTPUT_FILE="${OUTPUT_DIR}/qwen-${MODEL_VERSION}-sokoban-${ACTUAL_TRAJECTORIES}.jsonl"
+
+# Create output directory
+mkdir -p ${OUTPUT_DIR}
+
+echo "=========================================="
+echo "RAGEN Evaluation Configuration"
+echo "=========================================="
+echo "Model:              ${MODEL_PATH}"
+echo "Environment:        Sokoban (CoordSokoban)"
+echo "Trajectories:       ${ACTUAL_TRAJECTORIES} (${ENV_GROUPS} groups × ${GROUP_SIZE} size)"
+echo "GPU:                ${GPU_ID}"
+echo "Output:             ${OUTPUT_FILE}"
+echo "=========================================="
+echo ""
+
+# Run evaluation
+python -m ragen.llm_agent.agent_proxy \
+    --config-name eval \
+    system.CUDA_VISIBLE_DEVICES="${GPU_ID}" \
+    model_path="${MODEL_PATH}" \
+    es_manager.val.env_groups=${ENV_GROUPS} \
+    es_manager.val.group_size=${GROUP_SIZE} \
+    es_manager.val.env_configs.tags=["CoordSokoban"] \
+    es_manager.val.env_configs.n_groups=[${ENV_GROUPS}] \
+    output.dir="${OUTPUT_DIR}" \
+    output.filename="qwen-${MODEL_VERSION}-sokoban-${ACTUAL_TRAJECTORIES}.jsonl" \
+    output.format=jsonl \
+    output.append_timestamp=false
+
+echo ""
+echo "=========================================="
+echo "Evaluation completed!"
+echo "Results saved to: ${OUTPUT_FILE}"
+echo "=========================================="

scripts/reward_diagnosis/plot_reward_matrix.py

@@ -0,0 +1,302 @@
+#!/usr/bin/env python3
+"""
+Plot the reward triangular matrix from multi-rollout inference results.
+
+Reads the JSON output from run_inference.py and generates a heatmap where:
+  - Each row = one prompt, rollout rewards sorted high-to-low within the row
+  - Rows sorted by mean reward descending (easy on top, hard on bottom)
+  - Color: reward value (0 = red, 1 = green)
+  - Region labels and RV annotations on the right margin
+  - Classification: Easy (mean >= threshold), Hard (mean <= threshold), Mixed (in between)
+
+The resulting upper-triangular shape shows:
+  - Top rows: easy prompts (all green) — too easy, no RL signal
+  - Middle rows: mixed prompts (left green, right red) — learnable
+  - Bottom rows: hard prompts (all red) — too hard or broken
+
+Usage:
+    python scripts/reward_diagnosis/plot_reward_matrix.py \
+        --input logs/inference_results.json \
+        --output logs/reward_matrix.png
+"""
+
+import argparse
+import json
+import sys
+from pathlib import Path
+
+import numpy as np
+
+try:
+    import matplotlib
+    matplotlib.use("Agg")
+    import matplotlib.pyplot as plt
+    import matplotlib.gridspec as gridspec
+    from matplotlib.colors import LinearSegmentedColormap
+    from matplotlib.patches import FancyBboxPatch
+except ImportError:
+    print("matplotlib is required: pip install matplotlib")
+    sys.exit(1)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Plot reward triangular matrix")
+    parser.add_argument("--input", required=True, help="Path to inference JSON from run_inference.py")
+    parser.add_argument("--output", default=None, help="Output image path (default: <input_stem>_matrix.png)")
+    parser.add_argument("--max_prompts", type=int, default=100, help="Max prompts to display")
+    parser.add_argument("--easy_threshold", type=float, default=0.8,
+                        help="Mean reward >= this is classified as Easy")
+    parser.add_argument("--hard_threshold", type=float, default=0.2,
+                        help="Mean reward <= this is classified as Hard")
+    parser.add_argument("--dpi", type=int, default=200)
+    args = parser.parse_args()
+
+    # Load data
+    with open(args.input) as f:
+        data = json.load(f)
+
+    config = data["config"]
+    prompts = data["prompts"]
+    summary = data["summary"]
+    n_rollouts = config["rollouts_per_prompt"]
+
+    # Build matrix: each row = sorted rewards (descending) for one prompt
+    reward_rows = []
+    rv_values = []
+
+    for p in prompts:
+        rewards = sorted(p["rewards"], reverse=True)
+        reward_rows.append(rewards)
+        rv_values.append(p["reward_variance"])
+
+    # Sort by mean reward descending (easy on top, hard on bottom → upper triangle)
+    mean_rewards = [np.mean(row) for row in reward_rows]
+    sort_idx = np.argsort(mean_rewards)[::-1]
+    reward_rows = [reward_rows[i] for i in sort_idx]
+    rv_values = [rv_values[i] for i in sort_idx]
+    mean_rewards = [mean_rewards[i] for i in sort_idx]
+
+    # Truncate for display
+    n_display = min(len(reward_rows), args.max_prompts)
+    reward_rows = reward_rows[:n_display]
+    rv_values = rv_values[:n_display]
+
+    matrix = np.array(reward_rows)
+
+    # Classify by mean reward thresholds
+    n_easy = sum(1 for m in mean_rewards if m >= args.easy_threshold)
+    n_hard = sum(1 for m in mean_rewards if m <= args.hard_threshold)
+    n_mixed = n_display - n_easy - n_hard
+    n_other = 0
+
+    # ── Figure layout ──
+    fig_w = 10
+    fig_h = max(5, n_display * 0.18 + 2.5)
+    fig = plt.figure(figsize=(fig_w, fig_h), facecolor="#fafafa")
+
+    gs = gridspec.GridSpec(
+        2, 2,
+        width_ratios=[1, 0.04],
+        height_ratios=[1, 0.08],
+        hspace=0.35, wspace=0.08,
+        left=0.08, right=0.85, top=0.88, bottom=0.08,
+    )
+    ax_main = fig.add_subplot(gs[0, 0])
+    ax_cbar = fig.add_subplot(gs[0, 1])
+    ax_summary = fig.add_subplot(gs[1, :])
+
+    # ── Colormap ──
+    cmap = LinearSegmentedColormap.from_list(
+        "reward",
+        [
+            (0.0, "#c62828"),   # deep red
+            (0.15, "#e53935"),  # red
+            (0.35, "#ff8f00"),  # amber
+            (0.50, "#fdd835"),  # yellow
+            (0.65, "#7cb342"),  # light green
+            (0.85, "#388e3c"),  # green
+            (1.0, "#1b5e20"),   # deep green
+        ],
+    )
+
+    # ── Main heatmap ──
+    im = ax_main.imshow(
+        matrix, aspect="auto", cmap=cmap, vmin=0, vmax=1,
+        interpolation="nearest",
+    )
+
+    # Cell value annotations (only if matrix is small enough)
+    if n_display <= 40 and n_rollouts <= 16:
+        for i in range(n_display):
+            for j in range(n_rollouts):
+                val = matrix[i, j]
+                color = "white" if val < 0.4 or val > 0.8 else "black"
+                ax_main.text(j, i, f"{val:.1f}", ha="center", va="center",
+                             fontsize=6, color=color, fontweight="bold")
+
+    # X axis
+    ax_main.set_xlabel("Rollouts (sorted high → low)", fontsize=10, labelpad=8)
+    ax_main.set_xticks(range(n_rollouts))
+    ax_main.set_xticklabels([str(i + 1) for i in range(n_rollouts)], fontsize=8)
+    ax_main.xaxis.set_ticks_position("bottom")
+
+    # Y axis
+    ax_main.set_ylabel("Prompts (sorted by mean reward ↓)", fontsize=10, labelpad=8)
+    if n_display <= 50:
+        ax_main.set_yticks(range(n_display))
+        ax_main.set_yticklabels(range(1, n_display + 1), fontsize=6)
+    else:
+        step = max(1, n_display // 25)
+        ticks = list(range(0, n_display, step))
+        ax_main.set_yticks(ticks)
+        ax_main.set_yticklabels([i + 1 for i in ticks], fontsize=7)
+
+    # ── Region brackets on the right ──
+    region_x = n_rollouts - 0.5 + 0.6  # just outside the matrix
+    bracket_style = dict(fontsize=8, va="center", ha="left", fontweight="bold")
+
+    # RV labels on the right of each row
+    for i in range(n_display):
+        rv_color = "#1565c0" if args.hard_threshold < mean_rewards[i] < args.easy_threshold else (
+            "#2e7d32" if mean_rewards[i] >= args.easy_threshold else "#c62828")
+        ax_main.text(
+            n_rollouts - 0.5 + 0.3, i, f"{rv_values[i]:.3f}",
+            fontsize=5, va="center", ha="left", color=rv_color,
+            clip_on=False,
+        )
+
+    # Region label header
+    ax_main.text(
+        n_rollouts - 0.5 + 0.3, -0.8, "RV",
+        fontsize=6, va="center", ha="left", color="#424242",
+        fontweight="bold", clip_on=False,
+    )
+
+    # Divider lines between regions
+    # Find boundaries: mixed (0.2 < mean < 0.8), easy (mean >= 0.8), hard (mean <= 0.2)
+    # Since sorted by RV desc, regions may not be contiguous, so draw lines at transitions
+    for i in range(n_display - 1):
+        cat_i = "mixed" if args.hard_threshold < mean_rewards[i] < args.easy_threshold else ("easy" if mean_rewards[i] >= args.easy_threshold else "hard")
+        cat_next = "mixed" if args.hard_threshold < mean_rewards[i+1] < args.easy_threshold else ("easy" if mean_rewards[i+1] >= args.easy_threshold else "hard")
+        if cat_i != cat_next:
+            ax_main.axhline(y=i + 0.5, color="#455a64", linewidth=1.0, linestyle="--", alpha=0.5)
+
+    # Grid lines
+    ax_main.set_xticks([x - 0.5 for x in range(1, n_rollouts)], minor=True)
+    ax_main.set_yticks([y - 0.5 for y in range(1, n_display)], minor=True)
+    ax_main.grid(which="minor", color="#e0e0e0", linewidth=0.3)
+    ax_main.tick_params(which="minor", length=0)
+
+    # ── Colorbar ──
+    cbar = fig.colorbar(im, cax=ax_cbar)
+    cbar.set_label("Reward", fontsize=9, labelpad=8)
+    cbar.ax.tick_params(labelsize=8)
+
+    # ── Title ──
+    model_short = config["model"].split("/")[-1]
+    fig.suptitle(
+        f"Reward Matrix — {model_short}",
+        fontsize=14, fontweight="bold", y=0.96,
+    )
+    ax_main.set_title(
+        f"{n_display} prompts × {n_rollouts} rollouts  |  temp = {config['temperature']}",
+        fontsize=10, color="#616161", pad=10,
+    )
+
+    # ── Summary bar at bottom ──
+    ax_summary.set_xlim(0, 1)
+    ax_summary.set_ylim(0, 1)
+    ax_summary.axis("off")
+
+    # Stacked bar showing proportions
+    bar_y, bar_h = 0.55, 0.35
+    segments = []
+    if n_mixed > 0:
+        segments.append((n_mixed / n_display, "#1565c0", f"Mixed: {n_mixed} ({n_mixed/n_display*100:.0f}%)"))
+    if n_other > 0:
+        segments.append((n_other / n_display, "#78909c", f"Other: {n_other}"))
+    if n_easy > 0:
+        segments.append((n_easy / n_display, "#43a047", f"Easy: {n_easy} ({n_easy/n_display*100:.0f}%)"))
+    if n_hard > 0:
+        segments.append((n_hard / n_display, "#e53935", f"Hard: {n_hard} ({n_hard/n_display*100:.0f}%)"))
+
+    x_pos = 0.05
+    bar_total_w = 0.6
+    for frac, color, label in segments:
+        w = frac * bar_total_w
+        rect = FancyBboxPatch(
+            (x_pos, bar_y), w, bar_h,
+            boxstyle="round,pad=0.01", facecolor=color, edgecolor="white", linewidth=1.5,
+        )
+        ax_summary.add_patch(rect)
+        if w > 0.05:
+            ax_summary.text(x_pos + w / 2, bar_y + bar_h / 2, label,
+                            ha="center", va="center", fontsize=7, color="white", fontweight="bold")
+        x_pos += w
+
+    # Diagnosis text
+    mean_rv_mixed = np.mean([rv for rv in rv_values if rv > 0]) if n_mixed > 0 else 0
+    diag_x = 0.72
+    ax_summary.text(diag_x, 0.85, f"Mean reward: {summary['mean_reward']:.3f}",
+                    fontsize=8, color="#424242", transform=ax_summary.transAxes)
+    ax_summary.text(diag_x, 0.55, f"Mean RV (mixed): {mean_rv_mixed:.4f}",
+                    fontsize=8, color="#424242", transform=ax_summary.transAxes)
+
+    mixed_pct = n_mixed / max(n_display, 1) * 100
+    if mixed_pct >= 20:
+        verdict = "✓ Good RL signal"
+        verdict_color = "#2e7d32"
+    elif mixed_pct >= 10:
+        verdict = "~ Weak RL signal"
+        verdict_color = "#f57f17"
+    else:
+        verdict = "✗ Poor RL signal"
+        verdict_color = "#c62828"
+    ax_summary.text(diag_x, 0.2, verdict,
+                    fontsize=9, color=verdict_color, fontweight="bold", transform=ax_summary.transAxes)
+
+    # ── Save ──
+    if args.output is None:
+        output_path = Path(args.input).with_name(Path(args.input).stem + "_matrix.png")
+    else:
+        output_path = Path(args.output)
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+
+    fig.savefig(output_path, dpi=args.dpi, bbox_inches="tight", facecolor=fig.get_facecolor())
+    plt.close(fig)
+    print(f"Saved reward matrix to {output_path}")
+
+    # Text summary
+    print(f"\n{'=' * 60}")
+    print(f"REWARD MATRIX SUMMARY")
+    print(f"{'=' * 60}")
+    print(f"Model:           {config['model']}")
+    print(f"Prompts:         {n_display}")
+    print(f"Rollouts/prompt: {n_rollouts}")
+    print(f"Temperature:     {config['temperature']}")
+    print()
+    print(f"Mixed (RV > 0):  {n_mixed:4d}  ({n_mixed/n_display*100:5.1f}%)  <- RL can learn from these")
+    print(f"All correct:     {n_easy:4d}  ({n_easy/n_display*100:5.1f}%)  <- too easy, no signal")
+    print(f"All wrong:       {n_hard:4d}  ({n_hard/n_display*100:5.1f}%)  <- too hard or broken")
+    print()
+    print(f"Mean RV (mixed only): {mean_rv_mixed:.4f}")
+    print(f"Overall mean reward:  {summary['mean_reward']:.4f}")
+    print()
+
+    if n_hard / max(n_display, 1) > 0.5:
+        print("DIAGNOSIS: >50% prompts are all-wrong.")
+        print("  -> Check: Is the environment set up correctly?")
+        print("  -> Check: Does the model understand the expected action format?")
+    elif n_easy / max(n_display, 1) > 0.5:
+        print("DIAGNOSIS: >50% prompts are all-correct.")
+        print("  -> Task may be too easy. Consider harder subset or lower temperature.")
+    elif n_mixed / max(n_display, 1) < 0.2:
+        print("DIAGNOSIS: <20% prompts have mixed rewards. Weak RL signal.")
+        print("  -> Adjust temperature, check environment setup, or use different data.")
+    else:
+        print(f"DIAGNOSIS: Good RL signal. {n_mixed/n_display*100:.0f}% prompts are learnable.")
+        print("  -> Proceed to training.")
+
+
+if __name__ == "__main__":
+    main()

scripts/runs/README_webshop_small_combos.md

@@ -0,0 +1,71 @@
+# Webshop Small Combos — Running Guide
+
+## Combos
+
+The script defines 4 model×algorithm combinations:
+
+| Combo | Model | Algorithm |
+|-------|-------|-----------|
+| 1 | Qwen2.5-3B-Instruct | PPO |
+| 2 | Qwen2.5-3B-Instruct | GRPO |
+| 3 | Qwen2.5-7B-Instruct | PPO |
+| 4 | Llama-3.2-3B-Instruct | PPO |
+
+Use `--combos` to select which ones to run (1-indexed, comma-separated). Default: all.
+
+## Filter Modes
+
+- `--filters filter` → top_p=0.9 (rollout filtering enabled)
+- `--filters nofilter` → top_p=1.0 (no filtering)
+- `--filters all` → runs both (default)
+
+## Multi-GPU for One Experiment
+
+Use `--gpus-per-exp N` to assign multiple GPUs to a single experiment. The GPUs listed in `--gpus` are split into groups of N.
+
+Example: 2 GPUs per experiment on GPUs 0-3:
+```bash
+bash scripts/runs/run_webshop_small_combos.sh --gpus 0,1,2,3 --gpus-per-exp 2 --combos 3
+# Creates 2 slots: [0,1] and [2,3]
+```
+
+## Example: Running All Experiments (8 GPUs)
+
+Run these in separate terminals (disjoint GPUs, safe to run in parallel):
+
+```bash
+# Terminal 1 — 7B filter (GPUs 0,1)
+bash scripts/runs/run_webshop_small_combos.sh --steps 200 --gpus 0,1 --gpus-per-exp 2 --filters filter --combos 3
+
+# Terminal 2 — 7B nofilter (GPUs 2,3)
+bash scripts/runs/run_webshop_small_combos.sh --steps 200 --gpus 2,3 --gpus-per-exp 2 --filters nofilter --combos 3
+
+# Terminal 3 — 3B Qwen PPO nofilter (GPU 4)
+bash scripts/runs/run_webshop_small_combos.sh --steps 200 --gpus 4 --filters nofilter --combos 1
+
+# Terminal 4 — 3B Qwen GRPO nofilter (GPU 5)
+bash scripts/runs/run_webshop_small_combos.sh --steps 200 --gpus 5 --filters nofilter --combos 2
+
+# Terminal 5 — 3B Llama PPO nofilter (GPU 6)
+bash scripts/runs/run_webshop_small_combos.sh --steps 200 --gpus 6 --filters nofilter --combos 4
+```
+
+## Key Parameters
+
+| Flag | Default | Description |
+|------|---------|-------------|
+| `--steps N` | 100 | Training steps |
+| `--gpus LIST` | auto-detect | Comma-separated GPU IDs |
+| `--gpus-per-exp N` | 1 | GPUs per experiment |
+| `--combos LIST` | all | Which combos to run (1-4) |
+| `--filters LIST` | all | filter, nofilter, or all |
+| `--save-freq N` | 100 | Checkpoint save frequency |
+| `--cooldown N` | 30 | Seconds between runs on same GPU slot |
+| `--gpu-memory-utilization V` | 0.3 | vLLM rollout GPU memory fraction |
+
+## Output
+
+- Logs: `logs/webshop_small_combos/<name>.log`
+- Results: `logs/webshop_small_combos/<name>.result`
+- Checkpoints: `model_saving/webshop_small_combos/<model>/<algo>/<filter>/<name>/`
+- Wandb project: `main_webshop`

scripts/runs/run_entropy_sweep.sh

@@ -0,0 +1,463 @@
+#!/bin/bash
+# Entropy coefficient sweep for sokoban + GAE, keeping KL loss fixed and top_p filtering.
+set -euo pipefail
+
+# Defaults
+STEPS=400
+MODEL_NAME="Qwen2.5-3B"
+MODEL_PATH="Qwen/${MODEL_NAME}"
+PROJECT_NAME="ragen_release_entropy_sweep"
+CONFIG_NAME="_2_sokoban"
+SAVE_FREQ=-1
+ENTROPY_VALUES="0,0.001,0.003,0.01,0.03,0.1"
+ROLL_FILTER_INCLUDE_ZERO="True"
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=0
+GPU_MEMORY_UTILIZATION=0.5
+RAY_NUM_CPUS=16
+
+usage() {
+    cat <<'EOF'
+Usage: $0 [options]
+Options:
+  --steps N                     Training steps (default: 400)
+  --entropy-values LIST         Comma-separated entropy_coeff values (default: 0,0.001,0.003,0.01,0.03,0.1)
+  --rollout_filter_include_zero BOOL  Whether rollout_filter_include_zero (default: True)
+  --gpus LIST                   Comma-separated GPU IDs
+  --gpus-per-exp N              GPUs per experiment (default: 1)
+  --ray-num-cpus N              Max CPUs per task for ray.init (default: 16)
+  --gpu-memory-utilization V    GPU memory utilization for rollouts (default: 0.5)
+  --save-freq N                 Checkpoint save frequency (default: -1)
+  -h, --help                    Show this help
+EOF
+    exit 0
+}
+
+parse_bool() {
+    local val="${1,,}"
+    case "$val" in
+        true|1|yes|y) echo "True" ;;
+        false|0|no|n) echo "False" ;;
+        *)
+            echo "Error: expected boolean, got '$1'" >&2
+            exit 1
+            ;;
+    esac
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --entropy-values) ENTROPY_VALUES="$2"; shift 2 ;;
+        --entropy-values=*) ENTROPY_VALUES="${1#*=}"; shift ;;
+        --rollout_filter_include_zero) ROLL_FILTER_INCLUDE_ZERO=$(parse_bool "$2"); shift 2 ;;
+        --rollout_filter_include_zero=*) ROLL_FILTER_INCLUDE_ZERO=$(parse_bool "${1#*=}"); shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --ray-num-cpus) RAY_NUM_CPUS="$2"; shift 2 ;;
+        --ray-num-cpus=*) RAY_NUM_CPUS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+if ! [[ "$RAY_NUM_CPUS" =~ ^[0-9]+$ ]] || [ "$RAY_NUM_CPUS" -lt 1 ]; then
+    echo "Error: --ray-num-cpus must be a positive integer"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+NUM_SLOTS=${#GPU_GROUPS[@]}
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+declare -A GPU_LABELS
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_model_label() {
+    local models=()
+    local id label model
+    for id in "${GPUS[@]}"; do
+        label=$(get_gpu_label "$id")
+        model="${label#1x}"
+        models+=("$model")
+    done
+    local unique_models=()
+    local m found
+    for m in "${models[@]}"; do
+        found=false
+        for u in "${unique_models[@]}"; do
+            if [ "$u" = "$m" ]; then
+                found=true
+                break
+            fi
+        done
+        if [ "$found" = false ]; then
+            unique_models+=("$m")
+        fi
+    done
+    if [ ${#unique_models[@]} -eq 1 ]; then
+        echo "${unique_models[0]}"
+    else
+        echo "mixed"
+    fi
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+GPU_MODEL_LABEL=$(get_gpu_model_label)
+GPU_LOG_LABEL="${GPUS_PER_EXP}x${GPU_MODEL_LABEL}"
+LOG_FILE="logs/entropy_sweep_${MODEL_NAME}.log"
+RESULT_ROOT="logs/entropy_sweep_${MODEL_NAME}"
+CHECKPOINT_ROOT="model_saving/entropy_sweep_${MODEL_NAME}"
+
+mkdir -p logs
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== Entropy Sweep Runner (${MODEL_NAME}): $(date) ===" | tee "$LOG_FILE"
+echo "Values: ${ENTROPY_VALUES} | Steps: ${STEPS} | GPUs per exp: ${GPU_LOG_LABEL}" | tee -a "$LOG_FILE"
+echo "Groups: ${GPU_GROUPS[*]} | GPU memory util: ${GPU_MEMORY_UTILIZATION} | ray_num_cpus: ${RAY_NUM_CPUS} | save_freq: ${SAVE_FREQ}" | tee -a "$LOG_FILE"
+
+parse_values() {
+    IFS=',' read -r -a raw <<< "$1"
+    VALUES=()
+    for token in "${raw[@]}"; do
+        token="${token// /}"
+        if [ -n "$token" ]; then
+            VALUES+=("$token")
+        fi
+    done
+    if [ ${#VALUES[@]} -eq 0 ]; then
+        echo "Error: no entries provided for value list" >&2
+        exit 1
+    fi
+}
+
+parse_values "$ENTROPY_VALUES"
+EXPERIMENTS=("${VALUES[@]}")
+
+run_experiment() {
+    local value="$1"
+    local gpu_list="$2"
+    local safe_label="${value//./}"
+    safe_label="${safe_label,,}"
+    local filter_tag="nofilter"
+    if [ "${ROLL_FILTER_INCLUDE_ZERO}" = "False" ]; then
+        filter_tag="filter_zero"
+    fi
+
+    local name="sokoban_entropy_sweep_${filter_tag}_${safe_label}-${MODEL_NAME}"
+    local task_dir="${RESULT_ROOT}/${filter_tag}/${safe_label}"
+    local log_path="${task_dir}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${safe_label}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    mkdir -p "$task_dir"
+    mkdir -p "$checkpoint_dir"
+
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$CONFIG_NAME" \
+        model_path="${MODEL_PATH}" \
+        trainer.project_name="${PROJECT_NAME}" \
+        trainer.experiment_name="${name}" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        ray_kwargs.ray_init.num_cpus="${RAY_NUM_CPUS}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        algorithm.adv_estimator=gae \
+        actor_rollout_ref.actor.use_kl_loss=False \
+        actor_rollout_ref.actor.kl_loss_coef=0 \
+        actor_rollout_ref.actor.entropy_coeff="${value}" \
+        actor_rollout_ref.actor.entropy_from_logits_with_chunking=True \
+        actor_rollout_ref.actor.filter_loss_scaling=none \
+        actor_rollout_ref.actor.ppo_mini_batch_size=32 \
+        actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
+        actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=8 \
+        critic.ppo_mini_batch_size=32 \
+        critic.ppo_micro_batch_size_per_gpu=4 \
+        ppo_mini_batch_size=32 \
+        actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=8 \
+        actor_rollout_ref.rollout.rollout_filter_strategy=top_p \
+        actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=softmax \
+        actor_rollout_ref.rollout.rollout_filter_value=1 \
+        actor_rollout_ref.rollout.rollout_filter_include_zero=${ROLL_FILTER_INCLUDE_ZERO} \
+        actor_rollout_ref.rollout.rollout_filter_type=largest \
+        actor_rollout_ref.rollout.gpu_memory_utilization="${GPU_MEMORY_UTILIZATION}" \
+        actor_rollout_ref.rollout.rollout_filter_metric=reward_variance \
+        es_manager.train.env_groups=8 \
+        es_manager.train.group_size=16 \
+        es_manager.train.env_configs.n_groups='[8]' \
+        es_manager.val.env_groups=512 \
+        es_manager.val.group_size=1 \
+        es_manager.val.env_configs.n_groups='[512]' \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+
+    END=$(date +%s)
+    TOTAL_TIME=$((END - START))
+
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\s]+([\d.]+)",
+    r"timing_s/eval_total[:\s]+([\d.]+)",
+    r"timing_s/total[:\s]+([\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="entropy=${value} | filter=${filter_tag} | include_zero=${ROLL_FILTER_INCLUDE_ZERO} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${task_dir}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+}
+
+EXPERIMENT_COUNT=${#EXPERIMENTS[@]}
+if [ $EXPERIMENT_COUNT -eq 0 ]; then
+    echo "No experiments to run" >&2
+    exit 1
+fi
+
+QUEUE_FILE=$(mktemp -t ragen_entropy_queue.XXXXXX)
+echo 0 > "$QUEUE_FILE"
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+USE_FLOCK=false
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$MAIN_PID" != "$$" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list="$1"
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        local value="${EXPERIMENTS[$idx]}"
+        run_experiment "$value" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== Entropy Sweep Summary ==="
+    echo "Project: ${PROJECT_NAME} | Steps: ${STEPS} | GPU per exp: ${GPU_LOG_LABEL}"
+    for value in "${EXPERIMENTS[@]}"; do
+        safe_label="${value//./}"
+        safe_label="${safe_label,,}"
+        filter_tag="nofilter"
+        if [ "${ROLL_FILTER_INCLUDE_ZERO}" = "False" ]; then
+            filter_tag="filter_zero"
+        fi
+        name="sokoban_entropy_sweep_${filter_tag}_${safe_label}-${MODEL_NAME}"
+        task_dir="${RESULT_ROOT}/${filter_tag}/${safe_label}"
+        if [ -f "${task_dir}/${name}.result" ]; then
+            cat "${task_dir}/${name}.result"
+        else
+            echo "entropy=${value} | status=missing"
+        fi
+    done
+} | tee -a "$LOG_FILE"

scripts/runs/run_filtering_final.sh

@@ -0,0 +1,144 @@
+#!/bin/bash
+set -euo pipefail
+
+# usage: bash scripts/runs/run_filtering_final.sh [gpus_per_exp]
+GPUS_PER_EXP="${1:-2}" 
+
+# -----------------------
+# GPU AUTO-DETECTION
+# -----------------------
+detect_gpus() {
+  if command -v nvidia-smi >/dev/null 2>&1; then
+    nvidia-smi -L 2>/dev/null | grep -c '^GPU ' || true
+  else
+    echo 0
+  fi
+}
+
+TOTAL_GPUS=$(detect_gpus)
+if [ "$TOTAL_GPUS" -eq 0 ]; then
+    echo "ERROR: No GPUs detected via nvidia-smi." >&2
+    exit 1
+fi
+echo "INFO: Detected $TOTAL_GPUS GPUs."
+
+# -----------------------
+# Parallel GPU Management
+# -----------------------
+GPU_POOL_FIFO="/tmp/gpu_pool_$$"
+mkfifo "$GPU_POOL_FIFO"
+exec 3<>"$GPU_POOL_FIFO"
+rm "$GPU_POOL_FIFO"
+
+for ((i=0; i<TOTAL_GPUS; i++)); do
+    echo "$i" >&3
+done
+
+# -----------------------
+# Experiment List
+# -----------------------
+# Format: "ALGO METRIC STRATEGY VALUE TYPE INCLUDE_ZERO EXP_SUFFIX"
+EXPS=(
+    # Reward Variance - topP
+    "ppo reward_variance top_p 1.0 largest False rv_tp1.0"
+    "ppo reward_variance top_p 0.95 largest False rv_tp0.95"
+    "ppo reward_variance top_p 0.9 largest False rv_tp0.9"
+    "ppo reward_variance top_p 0.5 largest False rv_tp0.5"
+    
+    # Reward Variance - minP
+    "ppo reward_variance min_p 0.1 largest False rv_mp0.1"
+    "ppo reward_variance min_p 0.05 largest False rv_mp0.05"
+    "ppo reward_variance min_p 0.2 largest False rv_mp0.2"
+    
+    # Reward Variance - topK (Fractional)
+    "ppo reward_variance top_k 0.25 smallest False rv_tk0.25_rev"
+    "ppo reward_variance top_k 0.5 smallest False rv_tk0.5_rev"
+    "ppo reward_variance top_k 0.25 largest False rv_tk0.25"
+    "ppo reward_variance top_k 0.5 largest False rv_tk0.5"
+    
+    # Entropy
+    "ppo entropy top_p 0.9 largest False ent_tp0.9"
+    
+    # Entropy Variance
+    "ppo entropy_variance top_p 0.9 largest False entv_tp0.9"
+    
+    # Length
+    "ppo length top_p 0.9 largest False len_tp0.9"
+    "ppo length top_p 0.9 smallest False len_tp0.9_rev"
+)
+
+# -----------------------
+# Setup
+# -----------------------
+ENV="_2_sokoban"
+OUTPUT_DIR="/mnt/permanent/xjin/20260126_filters_final"
+mkdir -p "$OUTPUT_DIR"
+DONE_LIST="filter_final_donelist.txt"
+touch "$DONE_LIST"
+
+COMMON_FLAGS_BASE="trainer.total_training_steps=400 micro_batch_size_per_gpu=4 ppo_mini_batch_size=32 trainer.save_freq=-1 \
+    trainer.project_name=filtering_final \
+    algorithm.kl_ctrl.kl_coef=0.001 actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    es_manager.train.env_groups=8 es_manager.train.group_size=16 es_manager.train.env_configs.n_groups=[8]"
+
+echo "========================================"
+echo "Starting Final Filtering Experiments"
+echo "Pool: $TOTAL_GPUS GPUs | Per-Exp: $GPUS_PER_EXP"
+echo "Output: $OUTPUT_DIR"
+echo "========================================"
+
+# -----------------------
+# Execution Loop
+# -----------------------
+for exp_str in "${EXPS[@]}"; do
+    read -r alg metric strategy value ftype inc_zero suffix <<< "$exp_str"
+    
+    exp_name="soko_3b_${alg}_${suffix}"
+    
+    if grep -q "^${exp_name}$" "$DONE_LIST"; then
+        echo "Skipping ${exp_name} (Already Done)"
+        continue
+    fi
+
+    # Acquire GPUs
+    allocated_gpus=()
+    for ((i=0; i<GPUS_PER_EXP; i++)); do
+        read -u 3 gid
+        allocated_gpus+=("$gid")
+    done
+    gpu_csv=$(IFS=,; echo "${allocated_gpus[*]}")
+
+    (
+        echo "Running: $exp_name on GPUs $gpu_csv"
+        
+        alg_flag="algorithm.adv_estimator=grpo"
+        [ "$alg" == "ppo" ] && alg_flag="algorithm.adv_estimator=gae"
+
+        if CUDA_VISIBLE_DEVICES="$gpu_csv" python train.py --config-name "$ENV" \
+            trainer.experiment_name="${exp_name}" \
+            actor_rollout_ref.rollout.rollout_filter_metric="${metric}" \
+            actor_rollout_ref.rollout.rollout_filter_strategy="${strategy}" \
+            actor_rollout_ref.rollout.rollout_filter_value="${value}" \
+            actor_rollout_ref.rollout.rollout_filter_type="${ftype}" \
+            actor_rollout_ref.rollout.rollout_filter_include_zero="${inc_zero}" \
+            trainer.n_gpus_per_node="${GPUS_PER_EXP}" \
+            system.CUDA_VISIBLE_DEVICES="\"${gpu_csv}\"" \
+            $alg_flag \
+            $COMMON_FLAGS_BASE \
+            trainer.default_local_dir="${OUTPUT_DIR}/${exp_name}"; then
+            
+            echo "$exp_name" >> "$DONE_LIST"
+        else
+            echo "ERROR: $exp_name failed on GPUs $gpu_csv." >&2
+        fi
+
+        # Release GPUs
+        for gid in "${allocated_gpus[@]}"; do
+            echo "$gid" >&3
+        done
+    ) &
+done
+
+wait
+exec 3>&-
+echo "All requested experiments completed."

scripts/runs/run_frozen_lake_slipper_rate_sweep.sh

@@ -0,0 +1,527 @@
+#!/bin/bash
+# FrozenLake slipper_rate sweep with fixed KL/Entropy and filter vs no-filter baselines.
+set -euo pipefail
+
+# Defaults
+STEPS=400
+MODEL_NAME="Qwen2.5-3B"
+MODEL_PATH="Qwen/${MODEL_NAME}"
+PROJECT_NAME="ragen_release_frozenlake_slipper_rate_sweep"
+CONFIG_NAME="_3_frozen_lake"
+SAVE_FREQ=-1
+SLIPPER_RATES="100,50,20,10,5,2,0"
+FILTER_MODES="filter,nofilter"
+FILTER_TOP_P="0.9"
+NOFILTER_TOP_P="1.0"
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=30
+GPU_MEMORY_UTILIZATION=0.5
+RAY_NUM_CPUS=16
+declare -A GPU_LABELS
+
+usage() {
+    cat <<'EOF'
+Usage: $0 [options]
+Options:
+  --steps N                     Training steps (default: 400)
+  --slipper-rate LIST           Comma-separated slipper rates. Supports:
+                                percentages (100,50,20,10,5,2,0),
+                                ratios (1.0,0.5,0.2...), and '%' suffix.
+  --filter-modes LIST           Comma-separated modes: filter,nofilter (default: both)
+  --filter-top-p V              top-p for filter mode (default: 0.9)
+  --nofilter-top-p V            top-p for nofilter mode (default: 1.0)
+  --gpus LIST                   Comma-separated GPU IDs (auto-detected if omitted)
+  --gpus-per-exp N              GPUs per experiment (default: 1)
+  --ray-num-cpus N              Max CPUs per task for ray.init (default: 16)
+  --cooldown SECONDS            Cooldown between runs on same GPU group (default: 30)
+  --gpu-memory-utilization V    Rollout gpu_memory_utilization (default: 0.5)
+  --save-freq N                 Checkpoint save frequency (default: -1)
+  -h, --help                    Show this help message
+EOF
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --slipper-rate) SLIPPER_RATES="$2"; shift 2 ;;
+        --slipper-rate=*) SLIPPER_RATES="${1#*=}"; shift ;;
+        --filter-modes) FILTER_MODES="$2"; shift 2 ;;
+        --filter-modes=*) FILTER_MODES="${1#*=}"; shift ;;
+        --filter-top-p) FILTER_TOP_P="$2"; shift 2 ;;
+        --filter-top-p=*) FILTER_TOP_P="${1#*=}"; shift ;;
+        --nofilter-top-p) NOFILTER_TOP_P="$2"; shift 2 ;;
+        --nofilter-top-p=*) NOFILTER_TOP_P="${1#*=}"; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --ray-num-cpus) RAY_NUM_CPUS="$2"; shift 2 ;;
+        --ray-num-cpus=*) RAY_NUM_CPUS="${1#*=}"; shift ;;
+        --cooldown) COOLDOWN_SECONDS="$2"; shift 2 ;;
+        --cooldown=*) COOLDOWN_SECONDS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+if ! [[ "$RAY_NUM_CPUS" =~ ^[0-9]+$ ]] || [ "$RAY_NUM_CPUS" -lt 1 ]; then
+    echo "Error: --ray-num-cpus must be a positive integer"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+normalize_slipper_rate() {
+    local raw="$1"
+    local val="${raw// /}"
+    val="${val%%%}"
+    python - "$val" <<'PY'
+import sys
+
+token = sys.argv[1].strip()
+if not token:
+    print("ERR:empty")
+    raise SystemExit(0)
+try:
+    x = float(token)
+except Exception:
+    print("ERR:not_numeric")
+    raise SystemExit(0)
+
+# <=1: already ratio; >1: treat as percentage
+if x > 1.0:
+    x = x / 100.0
+
+if x < 0.0 or x > 1.0:
+    print("ERR:out_of_range")
+    raise SystemExit(0)
+
+print(f"{x:.6f}")
+PY
+}
+
+parse_slipper_rates() {
+    IFS=',' read -r -a raw <<< "$SLIPPER_RATES"
+    SLIPPER_VALUES=()
+    for token in "${raw[@]}"; do
+        token="${token// /}"
+        if [ -z "$token" ]; then
+            continue
+        fi
+        local normalized
+        normalized=$(normalize_slipper_rate "$token")
+        if [[ "$normalized" == ERR:* ]]; then
+            echo "Error: invalid slipper rate '${token}' (${normalized#ERR:})" >&2
+            exit 1
+        fi
+        SLIPPER_VALUES+=("$normalized")
+    done
+    if [ ${#SLIPPER_VALUES[@]} -eq 0 ]; then
+        echo "Error: no valid slipper-rate entries provided" >&2
+        exit 1
+    fi
+}
+
+parse_filter_modes() {
+    IFS=',' read -r -a raw <<< "$FILTER_MODES"
+    MODES=()
+    for token in "${raw[@]}"; do
+        token="${token// /}"
+        token="${token,,}"
+        if [ -z "$token" ]; then
+            continue
+        fi
+        case "$token" in
+            filter|nofilter) MODES+=("$token") ;;
+            *)
+                echo "Error: invalid filter mode '${token}'. Use filter or nofilter." >&2
+                exit 1
+                ;;
+        esac
+    done
+    if [ ${#MODES[@]} -eq 0 ]; then
+        echo "Error: no valid filter modes provided" >&2
+        exit 1
+    fi
+}
+
+parse_slipper_rates
+parse_filter_modes
+
+format_rate_label() {
+    local raw="$1"
+    python - "$raw" <<'PY'
+import sys
+x = float(sys.argv[1])
+s = f"{x:.6f}".rstrip("0").rstrip(".")
+print(s.replace(".", "p"))
+PY
+}
+
+EXPERIMENTS=()
+for mode in "${MODES[@]}"; do
+    if [ "$mode" = "filter" ]; then
+        top_p="$FILTER_TOP_P"
+        include_zero="False"
+    else
+        top_p="$NOFILTER_TOP_P"
+        include_zero="True"
+    fi
+    for slipper_rate in "${SLIPPER_VALUES[@]}"; do
+        success_rate=$(python - "$slipper_rate" <<'PY'
+import sys
+s = float(sys.argv[1])
+print(f"{1.0 - s:.6f}")
+PY
+)
+        EXPERIMENTS+=("${mode}|${top_p}|${include_zero}|${slipper_rate}|${success_rate}")
+    done
+done
+
+if [ ${#EXPERIMENTS[@]} -eq 0 ]; then
+    echo "Error: no experiments generated" >&2
+    exit 1
+fi
+
+LOG_FILE="logs/frozenlake_slipper_rate_sweep_${MODEL_NAME}.log"
+RESULT_ROOT="logs/frozenlake_slipper_rate_sweep_${MODEL_NAME}"
+CHECKPOINT_ROOT="model_saving/frozenlake_slipper_rate_sweep_${MODEL_NAME}"
+
+mkdir -p logs
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== FrozenLake slipper_rate sweep (${MODEL_NAME}): $(date) ===" | tee "$LOG_FILE"
+echo "Slipper rates: ${SLIPPER_RATES} | Modes: ${FILTER_MODES} | Steps: ${STEPS}" | tee -a "$LOG_FILE"
+echo "Filter top-p: ${FILTER_TOP_P} | NoFilter top-p: ${NOFILTER_TOP_P}" | tee -a "$LOG_FILE"
+echo "GPU groups: ${GPU_GROUPS[*]} | cooldown=${COOLDOWN_SECONDS}s | ray_num_cpus=${RAY_NUM_CPUS}" | tee -a "$LOG_FILE"
+
+run_experiment() {
+    local mode="$1"
+    local top_p="$2"
+    local include_zero="$3"
+    local slipper_rate="$4"
+    local success_rate="$5"
+    local gpu_list="$6"
+
+    local slip_pct
+    slip_pct=$(python - "$slipper_rate" <<'PY'
+import sys
+s = float(sys.argv[1])
+print(f"{s * 100:.1f}")
+PY
+)
+    local slip_label
+    slip_label=$(format_rate_label "$slipper_rate")
+    local safe_label="slip${slip_label}"
+    safe_label="${safe_label,,}"
+
+    local name="frozenlake_${mode}_${safe_label}-${MODEL_NAME}"
+    local task_dir="${RESULT_ROOT}/${mode}/${safe_label}"
+    local log_path="${task_dir}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${mode}/${safe_label}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    mkdir -p "$task_dir"
+    mkdir -p "$checkpoint_dir"
+
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$CONFIG_NAME" \
+        model_path="${MODEL_PATH}" \
+        trainer.project_name="${PROJECT_NAME}" \
+        trainer.experiment_name="${name}" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        ray_kwargs.ray_init.num_cpus="${RAY_NUM_CPUS}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        algorithm.adv_estimator=gae \
+        actor_rollout_ref.actor.loss_agg_mode=token-mean \
+        actor_rollout_ref.actor.use_kl_loss=False \
+        actor_rollout_ref.actor.kl_loss_type=low-var-kl \
+        actor_rollout_ref.actor.kl_loss_coef=0 \
+        actor_rollout_ref.actor.entropy_coeff=0 \
+        actor_rollout_ref.actor.entropy_from_logits_with_chunking=True \
+        actor_rollout_ref.actor.filter_loss_scaling=none \
+        actor_rollout_ref.actor.ppo_mini_batch_size=32 \
+        actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
+        actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=8 \
+        critic.ppo_mini_batch_size=32 \
+        critic.ppo_micro_batch_size_per_gpu=4 \
+        ppo_mini_batch_size=32 \
+        actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=8 \
+        actor_rollout_ref.rollout.gpu_memory_utilization="${GPU_MEMORY_UTILIZATION}" \
+        actor_rollout_ref.rollout.rollout_filter_strategy=top_p \
+        actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=softmax \
+        actor_rollout_ref.rollout.rollout_filter_value="${top_p}" \
+        actor_rollout_ref.rollout.rollout_filter_include_zero="${include_zero}" \
+        actor_rollout_ref.rollout.rollout_filter_type=largest \
+        actor_rollout_ref.rollout.rollout_filter_metric=reward_variance \
+        custom_envs.CoordFrozenLake.env_config.success_rate="${success_rate}" \
+        actor_rollout_ref.actor.checkpoint.save_contents=[model] \
+        critic.checkpoint.save_contents=[model] \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+    END=$(date +%s)
+    TOTAL_TIME=$((END - START))
+
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\\s]+([\\d.]+)",
+    r"timing_s/eval_total[:\\s]+([\\d.]+)",
+    r"timing_s/total[:\\s]+([\\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="mode=${mode} | slipper_rate=${slipper_rate} (${slip_pct}%) | success_rate=${success_rate} | top_p=${top_p} | include_zero=${include_zero} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${task_dir}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+}
+
+QUEUE_FILE=$(mktemp -t ragen_frozenlake_slipper_queue.XXXXXX)
+echo 0 > "$QUEUE_FILE"
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+USE_FLOCK=false
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$MAIN_PID" != "$$" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list="$1"
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        IFS='|' read -r mode top_p include_zero slipper_rate success_rate <<< "${EXPERIMENTS[$idx]}"
+        run_experiment "$mode" "$top_p" "$include_zero" "$slipper_rate" "$success_rate" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== FrozenLake slipper_rate Sweep Summary ==="
+    echo "Project: ${PROJECT_NAME} | Steps: ${STEPS} | Modes: ${FILTER_MODES}"
+    for exp in "${EXPERIMENTS[@]}"; do
+        IFS='|' read -r mode top_p include_zero slipper_rate success_rate <<< "$exp"
+        slip_label=$(format_rate_label "$slipper_rate")
+        safe_label="slip${slip_label}"
+        safe_label="${safe_label,,}"
+        name="frozenlake_${mode}_${safe_label}-${MODEL_NAME}"
+        task_dir="${RESULT_ROOT}/${mode}/${safe_label}"
+        if [ -f "${task_dir}/${name}.result" ]; then
+            cat "${task_dir}/${name}.result"
+        else
+            echo "mode=${mode} | slipper_rate=${slipper_rate} | top_p=${top_p} | status=missing"
+        fi
+    done
+} | tee -a "$LOG_FILE"

scripts/runs/run_kl_sweep.sh

@@ -0,0 +1,467 @@
+#!/bin/bash
+# KL coefficient sweep for sokoban + GAE, fixing entropy at 0 and top_p filtering.
+set -euo pipefail
+
+# Defaults
+STEPS=400
+MODEL_NAME="Qwen2.5-3B"
+MODEL_PATH="Qwen/${MODEL_NAME}"
+PROJECT_NAME="ragen_release_kl_sweep"
+CONFIG_NAME="_2_sokoban"
+SAVE_FREQ=-1
+KL_VALUES="0,0.001,0.003,0.01,0.03,0.1"
+ROLL_FILTER_INCLUDE_ZERO="True"
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=0
+GPU_MEMORY_UTILIZATION=0.5
+RAY_NUM_CPUS=16
+
+usage() {
+    cat <<'EOF'
+Usage: $0 [options]
+Options:
+  --steps N                     Training steps (default: 400)
+  --kl-values LIST              Comma-separated kl_loss_coef values (default: 0,0.001,0.003,0.01,0.03,0.1)
+  --rollout_filter_include_zero BOOL  Whether rollout_filter_include_zero (default: True)
+  --gpus LIST                   Comma-separated GPU IDs
+  --gpus-per-exp N              GPUs per experiment (default: 1)
+  --ray-num-cpus N              Max CPUs per task for ray.init (default: 16)
+  --gpu-memory-utilization V    GPU memory utilization for rollouts (default: 0.5)
+  --save-freq N                 Checkpoint save frequency (default: -1)
+  -h, --help                    Show this help
+EOF
+    exit 0
+}
+
+parse_bool() {
+    local val="${1,,}"
+    case "$val" in
+        true|1|yes|y) echo "True" ;;
+        false|0|no|n) echo "False" ;;
+        *)
+            echo "Error: expected boolean, got '$1'" >&2
+            exit 1
+            ;;
+    esac
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --kl-values) KL_VALUES="$2"; shift 2 ;;
+        --kl-values=*) KL_VALUES="${1#*=}"; shift ;;
+        --rollout_filter_include_zero) ROLL_FILTER_INCLUDE_ZERO=$(parse_bool "$2"); shift 2 ;;
+        --rollout_filter_include_zero=*) ROLL_FILTER_INCLUDE_ZERO=$(parse_bool "${1#*=}"); shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --ray-num-cpus) RAY_NUM_CPUS="$2"; shift 2 ;;
+        --ray-num-cpus=*) RAY_NUM_CPUS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+if ! [[ "$RAY_NUM_CPUS" =~ ^[0-9]+$ ]] || [ "$RAY_NUM_CPUS" -lt 1 ]; then
+    echo "Error: --ray-num-cpus must be a positive integer"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+NUM_SLOTS=${#GPU_GROUPS[@]}
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+declare -A GPU_LABELS
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_model_label() {
+    local models=()
+    local id label model
+    for id in "${GPUS[@]}"; do
+        label=$(get_gpu_label "$id")
+        model="${label#1x}"
+        models+=("$model")
+    done
+    local unique_models=()
+    local m found
+    for m in "${models[@]}"; do
+        found=false
+        for u in "${unique_models[@]}"; do
+            if [ "$u" = "$m" ]; then
+                found=true
+                break
+            fi
+        done
+        if [ "$found" = false ]; then
+            unique_models+=("$m")
+        fi
+    done
+    if [ ${#unique_models[@]} -eq 1 ]; then
+        echo "${unique_models[0]}"
+    else
+        echo "mixed"
+    fi
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+GPU_MODEL_LABEL=$(get_gpu_model_label)
+GPU_LOG_LABEL="${GPUS_PER_EXP}x${GPU_MODEL_LABEL}"
+LOG_FILE="logs/kl_sweep_${MODEL_NAME}.log"
+RESULT_ROOT="logs/kl_sweep_${MODEL_NAME}"
+CHECKPOINT_ROOT="model_saving/kl_sweep_${MODEL_NAME}"
+
+mkdir -p logs
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== KL Sweep Runner (${MODEL_NAME}): $(date) ===" | tee "$LOG_FILE"
+echo "Values: ${KL_VALUES} | Steps: ${STEPS} | GPUs per exp: ${GPU_LOG_LABEL}" | tee -a "$LOG_FILE"
+echo "Groups: ${GPU_GROUPS[*]} | GPU memory util: ${GPU_MEMORY_UTILIZATION} | ray_num_cpus: ${RAY_NUM_CPUS} | save_freq: ${SAVE_FREQ}" | tee -a "$LOG_FILE"
+
+parse_values() {
+    IFS=',' read -r -a raw <<< "$1"
+    VALUES=()
+    for token in "${raw[@]}"; do
+        token="${token// /}"
+        if [ -n "$token" ]; then
+            VALUES+=("$token")
+        fi
+    done
+    if [ ${#VALUES[@]} -eq 0 ]; then
+        echo "Error: no entries provided for value list" >&2
+        exit 1
+    fi
+}
+
+parse_values "$KL_VALUES"
+EXPERIMENTS=("${VALUES[@]}")
+
+run_experiment() {
+    local value="$1"
+    local gpu_list="$2"
+    local safe_label="${value//./}"
+    local use_kl_loss="False"
+    if [ "$value" != "0" ]; then
+        use_kl_loss="True"
+    fi
+    safe_label="${safe_label,,}"
+    local filter_tag="nofilter"
+    if [ "${ROLL_FILTER_INCLUDE_ZERO}" = "False" ]; then
+        filter_tag="filter_zero"
+    fi
+
+    local name="sokoban_kl_sweep_${filter_tag}_${safe_label}-${MODEL_NAME}"
+    local task_dir="${RESULT_ROOT}/${filter_tag}/${safe_label}"
+    local log_path="${task_dir}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${safe_label}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    mkdir -p "$task_dir"
+    mkdir -p "$checkpoint_dir"
+
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$CONFIG_NAME" \
+        model_path="${MODEL_PATH}" \
+        trainer.project_name="${PROJECT_NAME}" \
+        trainer.experiment_name="${name}" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        ray_kwargs.ray_init.num_cpus="${RAY_NUM_CPUS}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        algorithm.adv_estimator=gae \
+        actor_rollout_ref.actor.use_kl_loss=${use_kl_loss} \
+        actor_rollout_ref.actor.kl_loss_coef="${value}" \
+        actor_rollout_ref.actor.entropy_coeff=0.0 \
+        actor_rollout_ref.actor.entropy_from_logits_with_chunking=True \
+        actor_rollout_ref.actor.filter_loss_scaling=none \
+        actor_rollout_ref.actor.ppo_mini_batch_size=32 \
+        actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
+        actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=8 \
+        critic.ppo_mini_batch_size=32 \
+        critic.ppo_micro_batch_size_per_gpu=4 \
+        ppo_mini_batch_size=32 \
+        actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=8 \
+        actor_rollout_ref.rollout.rollout_filter_strategy=top_p \
+        actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=softmax \
+        actor_rollout_ref.rollout.rollout_filter_value=1 \
+        actor_rollout_ref.rollout.rollout_filter_include_zero=${ROLL_FILTER_INCLUDE_ZERO} \
+        actor_rollout_ref.rollout.rollout_filter_type=largest \
+        actor_rollout_ref.rollout.gpu_memory_utilization="${GPU_MEMORY_UTILIZATION}" \
+        actor_rollout_ref.rollout.rollout_filter_metric=reward_variance \
+        es_manager.train.env_groups=8 \
+        es_manager.train.group_size=16 \
+        es_manager.train.env_configs.n_groups='[8]' \
+        es_manager.val.env_groups=512 \
+        es_manager.val.group_size=1 \
+        es_manager.val.env_configs.n_groups='[512]' \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+
+    END=$(date +%s)
+    TOTAL_TIME=$((END - START))
+
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\s]+([\d.]+)",
+    r"timing_s/eval_total[:\s]+([\d.]+)",
+    r"timing_s/total[:\s]+([\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="kl=${value} | filter=${filter_tag} | include_zero=${ROLL_FILTER_INCLUDE_ZERO} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${task_dir}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+}
+
+EXPERIMENT_COUNT=${#EXPERIMENTS[@]}
+if [ $EXPERIMENT_COUNT -eq 0 ]; then
+    echo "No experiments to run" >&2
+    exit 1
+fi
+
+QUEUE_FILE=$(mktemp -t ragen_kl_queue.XXXXXX)
+echo 0 > "$QUEUE_FILE"
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+USE_FLOCK=false
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$MAIN_PID" != "$$" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list="$1"
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        local value="${EXPERIMENTS[$idx]}"
+        run_experiment "$value" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== KL Sweep Summary ==="
+    echo "Project: ${PROJECT_NAME} | Steps: ${STEPS} | GPU per exp: ${GPU_LOG_LABEL}"
+    for value in "${EXPERIMENTS[@]}"; do
+        safe_label="${value//./}"
+        safe_label="${safe_label,,}"
+        filter_tag="nofilter"
+        if [ "${ROLL_FILTER_INCLUDE_ZERO}" = "False" ]; then
+            filter_tag="filter_zero"
+        fi
+        name="sokoban_kl_sweep_${filter_tag}_${safe_label}-${MODEL_NAME}"
+        task_dir="${RESULT_ROOT}/${filter_tag}/${safe_label}"
+        if [ -f "${task_dir}/${name}.result" ]; then
+            cat "${task_dir}/${name}.result"
+        else
+            echo "kl=${value} | status=missing"
+        fi
+    done
+} | tee -a "$LOG_FILE"

scripts/runs/run_main_table_diff_algo.sh

@@ -0,0 +1,546 @@
+#!/bin/bash
+# Main Table: Different algorithms (PPO/DAPO/GRPO/DrGRPO) × tasks × filter/no-filter
+# Model: Qwen2.5-3B only
+# Filtering rule: filter => top_p=0.9, nofilter => top_p=1.0
+
+set -euo pipefail
+
+# Defaults
+STEPS=400
+MODEL_NAME="Qwen2.5-3B"
+TASKS=("sokoban" "frozenlake" "webshop" "metamathqa" "countdown")
+ALGORITHMS=("PPO" "DAPO" "GRPO" "DrGRPO")
+MODEL_PATH=""
+SAVE_FREQ=-1
+FILTER_MODES=("filter" "nofilter")
+FILTERS_OPTION="all"
+SELECTED_FILTERS=("${FILTER_MODES[@]}")
+
+# GPU settings
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=30
+GPU_MEMORY_UTILIZATION=0.3
+RAY_NUM_CPUS=16
+declare -A GPU_LABELS
+
+usage() {
+    echo "Usage: $0 [options]"
+    echo "Options:"
+    echo "  --steps N             Training steps (default: 400)"
+    echo "  --tasks LIST          Comma-separated tasks (default: sokoban,frozenlake,webshop,metamathqa,countdown)"
+    echo "  --algos LIST          Comma-separated algorithms (default: PPO,DAPO,GRPO,DrGRPO)"
+    echo "  --gpus LIST           Comma-separated GPU IDs (default: auto-detect)"
+    echo "  --gpus-per-exp N      GPUs per experiment (default: 1)"
+    echo "  --cooldown SECONDS    Cooldown between runs on the same GPU group (default: 30)"
+    echo "  --gpu-memory-utilization V  Rollout gpu_memory_utilization (default: 0.3)"
+    echo "  --ray-num-cpus N      Max CPUs per task for ray.init (default: 16)"
+    echo "  --save-freq N         Checkpoint save frequency (default: -1 to disable saving)"
+    echo "  --filters LIST        Comma-separated filter modes (filter,nofilter,all). Default: all"
+    echo "  -h, --help            Show this help"
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --tasks) IFS=',' read -r -a TASKS <<< "$2"; shift 2 ;;
+        --tasks=*) IFS=',' read -r -a TASKS <<< "${1#*=}"; shift ;;
+        --algos) IFS=',' read -r -a ALGORITHMS <<< "$2"; shift 2 ;;
+        --algos=*) IFS=',' read -r -a ALGORITHMS <<< "${1#*=}"; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --cooldown) COOLDOWN_SECONDS="$2"; shift 2 ;;
+        --cooldown=*) COOLDOWN_SECONDS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --ray-num-cpus) RAY_NUM_CPUS="$2"; shift 2 ;;
+        --ray-num-cpus=*) RAY_NUM_CPUS="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        --filters) FILTERS_OPTION="$2"; shift 2 ;;
+        --filters=*) FILTERS_OPTION="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+MODEL_PATH="Qwen/${MODEL_NAME}"
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+if ! [[ "$RAY_NUM_CPUS" =~ ^[0-9]+$ ]] || [ "$RAY_NUM_CPUS" -lt 1 ]; then
+    echo "Error: --ray-num-cpus must be a positive integer"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+NUM_SLOTS=${#GPU_GROUPS[@]}
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_model_label() {
+    local models=()
+    local id label model
+    for id in "${GPUS[@]}"; do
+        label=$(get_gpu_label "$id")
+        model="${label#1x}"
+        models+=("$model")
+    done
+    local unique_models=()
+    local m found
+    for m in "${models[@]}"; do
+        found=false
+        for u in "${unique_models[@]}"; do
+            if [ "$u" = "$m" ]; then
+                found=true
+                break
+            fi
+        done
+        if [ "$found" = false ]; then
+            unique_models+=("$m")
+        fi
+    done
+    if [ ${#unique_models[@]} -eq 1 ]; then
+        echo "${unique_models[0]}"
+    else
+        echo "mixed"
+    fi
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+GPU_MODEL_LABEL=$(get_gpu_model_label)
+GPU_LOG_LABEL="${GPUS_PER_EXP}x${GPU_MODEL_LABEL}"
+LOG_FILE="logs/diff_algo_${MODEL_NAME}.log"
+RESULT_ROOT="logs"
+CHECKPOINT_ROOT="model_saving/diff_algo_${MODEL_NAME}"
+
+mkdir -p logs
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== Perf Table Runner for ${MODEL_NAME}: $(date) ===" | tee "$LOG_FILE"
+echo "Tasks: ${TASKS[*]} | Steps: ${STEPS} | GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL}" | tee -a "$LOG_FILE"
+echo "GPUS: ${GPUS[*]} | groups: ${GPU_GROUPS[*]} | cooldown=${COOLDOWN_SECONDS}s" | tee -a "$LOG_FILE"
+
+get_config_for_task() {
+    case "$1" in
+        countdown) echo "_4_countdown" ;;
+        sokoban) echo "_2_sokoban" ;;
+        frozenlake) echo "_3_frozen_lake" ;;
+        webshop) echo "_6_webshop" ;;
+        metamathqa) echo "_5_metamathqa" ;;
+        *) echo "" ;;
+    esac
+}
+
+get_algo_overrides() {
+    case "$1" in
+        PPO)
+            echo "algorithm.adv_estimator=gae actor_rollout_ref.actor.loss_agg_mode=token-mean"
+            ;;
+        DAPO)
+            # DAPO here = PPO + higher clip + no KL + token-level loss.
+            echo "algorithm.adv_estimator=gae actor_rollout_ref.actor.loss_agg_mode=token-mean actor_rollout_ref.actor.clip_ratio_low=0.2 actor_rollout_ref.actor.clip_ratio_high=0.28 actor_rollout_ref.actor.use_kl_loss=False actor_rollout_ref.actor.kl_loss_coef=0.0 algorithm.use_kl_in_reward=False algorithm.kl_ctrl.kl_coef=0.0"
+            ;;
+        GRPO)
+            echo "algorithm.adv_estimator=grpo algorithm.norm_adv_by_std_in_grpo=True actor_rollout_ref.actor.loss_agg_mode=seq-mean-token-mean"
+            ;;
+        DrGRPO)
+            echo "algorithm.adv_estimator=grpo algorithm.norm_adv_by_std_in_grpo=False actor_rollout_ref.actor.loss_agg_mode=seq-mean-token-sum"
+            ;;
+        *)
+            echo ""
+            ;;
+    esac
+}
+
+run_experiment() {
+    local task=$1
+    local algo=$2
+    local filter=$3
+    local config=$4
+    local gpu_list=$5
+
+    local filter_value
+    if [ "$filter" = "filter" ]; then
+        filter_value=0.9
+    else
+        filter_value=1.0
+    fi
+    local filter_strategy="top_p"
+
+    local common_overrides=(
+        "actor_rollout_ref.actor.use_kl_loss=False"
+        "actor_rollout_ref.actor.kl_loss_type=low-var-kl"
+        "actor_rollout_ref.actor.kl_loss_coef=0.001"
+        "actor_rollout_ref.actor.entropy_coeff=0.001"
+        "actor_rollout_ref.actor.entropy_from_logits_with_chunking=True"
+        "actor_rollout_ref.actor.filter_loss_scaling=none"
+        "actor_rollout_ref.rollout.gpu_memory_utilization=${GPU_MEMORY_UTILIZATION}"
+        "actor_rollout_ref.rollout.rollout_filter_strategy=${filter_strategy}"
+        "actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=softmax"
+        "actor_rollout_ref.rollout.rollout_filter_type=largest"
+        "actor_rollout_ref.rollout.rollout_filter_metric=reward_variance"
+        "actor_rollout_ref.rollout.rollout_filter_include_zero=True"
+    )
+
+    local env_overrides=()
+    if [ "$task" = "frozenlake" ]; then
+        env_overrides+=("custom_envs.CoordFrozenLake.env_config.success_rate=1.0")
+    fi
+
+    local checkpoint_overrides=(
+        "actor_rollout_ref.actor.checkpoint.save_contents=[model]"
+        "critic.checkpoint.save_contents=[model]"
+    )
+
+    local algo_overrides
+    algo_overrides=$(get_algo_overrides "$algo")
+    read -r -a algo_args <<< "$algo_overrides"
+
+    local name="${task}-${algo}-${filter}-${MODEL_NAME}"
+    local task_dir="${RESULT_ROOT}/diff_algo_${task}_${MODEL_NAME}"
+    local log_path="${task_dir}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${task}/${algo}/${filter}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    mkdir -p "$task_dir"
+    mkdir -p "${checkpoint_dir}"
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$config" \
+        model_path="${MODEL_PATH}" \
+        trainer.project_name="ragen_main_table_diff_algo" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.experiment_name="${name}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        ray_kwargs.ray_init.num_cpus="${RAY_NUM_CPUS}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        actor_rollout_ref.rollout.rollout_filter_value="${filter_value}" \
+        "${common_overrides[@]}" \
+        "${env_overrides[@]}" \
+        "${checkpoint_overrides[@]}" \
+        "${algo_args[@]}" \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+    END=$(date +%s)
+
+    TOTAL_TIME=$((END - START))
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\s]+([\d.]+)",
+    r"timing_s/eval_total[:\s]+([\d.]+)",
+    r"timing_s/total[:\s]+([\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="task=${task} | algo=${algo} | filter=${filter} | model=${MODEL_NAME} | steps=${STEPS} | filter=${filter_strategy}:${filter_value} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${task_dir}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+    return 0
+}
+
+EXPERIMENTS=()
+GROUP_LABELS=()
+CURRENT_GROUP=""
+
+set_group() {
+    CURRENT_GROUP="$1"
+    GROUP_LABELS+=("$1")
+}
+
+add_experiment() {
+    local task=$1
+    local algo=$2
+    local filter=$3
+    local config=$4
+    EXPERIMENTS+=("${CURRENT_GROUP}|${task}|${algo}|${filter}|${config}")
+}
+
+resolve_filter_selection() {
+    local raw="$1"
+    if [ -z "$raw" ] || [ "$raw" = "all" ]; then
+        SELECTED_FILTERS=("${FILTER_MODES[@]}")
+        return
+    fi
+    IFS=',' read -r -a candidates <<< "$raw"
+    SELECTED_FILTERS=()
+    for candidate in "${candidates[@]}"; do
+        candidate="${candidate// /}"
+        case "$candidate" in
+            filter|nofilter)
+                SELECTED_FILTERS+=("$candidate")
+                ;;
+            "")
+                continue
+                ;;
+            *)
+                echo "Unknown filter mode: $candidate" >&2
+                exit 1
+                ;;
+        esac
+    done
+    if [ ${#SELECTED_FILTERS[@]} -eq 0 ]; then
+        echo "No valid filters selected via --filters" >&2
+        exit 1
+    fi
+}
+
+resolve_filter_selection "$FILTERS_OPTION"
+
+for algo in "${ALGORITHMS[@]}"; do
+    set_group "Algorithm: ${algo}"
+    for task in "${TASKS[@]}"; do
+        config=$(get_config_for_task "$task")
+        if [ -z "$config" ]; then
+            echo "Unknown task: $task" >&2
+            exit 1
+        fi
+        for filter in "${SELECTED_FILTERS[@]}"; do
+            add_experiment "$task" "$algo" "$filter" "$config"
+        done
+    done
+done
+
+QUEUE_FILE=$(mktemp -t ragen_main_table_queue.XXXXXX)
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+echo 0 > "$QUEUE_FILE"
+USE_FLOCK=false
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$$" -ne "$MAIN_PID" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list=$1
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        local exp="${EXPERIMENTS[$idx]}"
+        IFS='|' read -r exp_group task algo filter config <<< "$exp"
+        run_experiment "$task" "$algo" "$filter" "$config" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== Grouped Summary ==="
+echo "GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL} | Model: ${MODEL_NAME} | Steps: ${STEPS}"
+    for group_label in "${GROUP_LABELS[@]}"; do
+        echo "=== ${group_label} ==="
+        for exp in "${EXPERIMENTS[@]}"; do
+            IFS='|' read -r exp_group task algo filter config <<< "$exp"
+            if [ "$exp_group" != "$group_label" ]; then
+                continue
+            fi
+            name="${task}-${algo}-${filter}-${MODEL_NAME}"
+            task_dir="${RESULT_ROOT}/diff_algo_${task}_${MODEL_NAME}"
+            if [ -f "${task_dir}/${name}.result" ]; then
+                cat "${task_dir}/${name}.result"
+            else
+                echo "task=${task} | algo=${algo} | filter=${filter} | model=${MODEL_NAME} | status=missing"
+            fi
+        done
+    done
+} | tee -a "$LOG_FILE"

scripts/runs/run_main_table_diff_model.sh

@@ -0,0 +1,541 @@
+#!/bin/bash
+# Model Type: Different model types (Instruct/Reasoning) × tasks × filter/no-filter
+# Algorithm: PPO only
+# Filtering rule: filter => top_p=0.9, nofilter => top_p=1.0
+
+set -euo pipefail
+
+# Defaults
+STEPS=400
+MODEL_NAMES=("Qwen2.5-3B-Instruct")
+TASKS=("sokoban" "frozenlake" "metamathqa" "countdown")
+SAVE_FREQ=-1
+FILTER_MODES=("filter" "nofilter")
+FILTERS_OPTION="all"
+SELECTED_FILTERS=("${FILTER_MODES[@]}")
+
+# GPU settings
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=30
+GPU_MEMORY_UTILIZATION=0.3
+RAY_NUM_CPUS=16
+declare -A GPU_LABELS
+
+usage() {
+    echo "Usage: $0 [options]"
+    echo "Options:"
+    echo "  --steps N             Training steps (default: 400)"
+    echo "  --tasks LIST          Comma-separated tasks (default: sokoban,frozenlake,webshop,metamathqa,countdown)"
+    echo "  --models LIST         Comma-separated model names or HF paths (default: Qwen2.5-3B-Instruct)"
+    echo "  --gpus LIST           Comma-separated GPU IDs (default: auto-detect)"
+    echo "  --gpus-per-exp N      GPUs per experiment (default: 1)"
+    echo "  --cooldown SECONDS    Cooldown between runs on the same GPU group (default: 30)"
+    echo "  --gpu-memory-utilization V  Rollout gpu_memory_utilization (default: 0.3)"
+    echo "  --ray-num-cpus N      Max CPUs per task for ray.init (default: 16)"
+    echo "  --save-freq N         Checkpoint save frequency (default: -1 to disable saving)"
+    echo "  --filters LIST        Comma-separated filter modes (filter,nofilter,all). Default: all"
+    echo "  -h, --help            Show this help"
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --tasks) IFS=',' read -r -a TASKS <<< "$2"; shift 2 ;;
+        --tasks=*) IFS=',' read -r -a TASKS <<< "${1#*=}"; shift ;;
+        --models) IFS=',' read -r -a MODEL_NAMES <<< "$2"; shift 2 ;;
+        --models=*) IFS=',' read -r -a MODEL_NAMES <<< "${1#*=}"; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --cooldown) COOLDOWN_SECONDS="$2"; shift 2 ;;
+        --cooldown=*) COOLDOWN_SECONDS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --ray-num-cpus) RAY_NUM_CPUS="$2"; shift 2 ;;
+        --ray-num-cpus=*) RAY_NUM_CPUS="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        --filters) FILTERS_OPTION="$2"; shift 2 ;;
+        --filters=*) FILTERS_OPTION="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+# Map model names to HuggingFace paths
+get_model_path() {
+    if [[ "$1" == *"/"* ]]; then
+        echo "$1"
+        return
+    fi
+    case "$1" in
+        Qwen2.5-3B-Instruct) echo "Qwen/Qwen2.5-3B-Instruct" ;;
+        QwQ-32B) echo "Qwen/QwQ-32B" ;;
+        Llama-3.2-3B-Instruct) echo "meta-llama/Llama-3.2-3B-Instruct" ;;
+        *) echo "Qwen/$1" ;;
+    esac
+}
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+if ! [[ "$RAY_NUM_CPUS" =~ ^[0-9]+$ ]] || [ "$RAY_NUM_CPUS" -lt 1 ]; then
+    echo "Error: --ray-num-cpus must be a positive integer"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+NUM_SLOTS=${#GPU_GROUPS[@]}
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_model_label() {
+    local models=()
+    local id label model
+    for id in "${GPUS[@]}"; do
+        label=$(get_gpu_label "$id")
+        model="${label#1x}"
+        models+=("$model")
+    done
+    local unique_models=()
+    local m found
+    for m in "${models[@]}"; do
+        found=false
+        for u in "${unique_models[@]}"; do
+            if [ "$u" = "$m" ]; then
+                found=true
+                break
+            fi
+        done
+        if [ "$found" = false ]; then
+            unique_models+=("$m")
+        fi
+    done
+    if [ ${#unique_models[@]} -eq 1 ]; then
+        echo "${unique_models[0]}"
+    else
+        echo "mixed"
+    fi
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+GPU_MODEL_LABEL=$(get_gpu_model_label)
+GPU_LOG_LABEL="${GPUS_PER_EXP}x${GPU_MODEL_LABEL}"
+LOG_FILE="logs/diff_model_PPO.log"
+RESULT_ROOT="logs"
+CHECKPOINT_ROOT="model_saving/diff_model"
+
+mkdir -p logs
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== Model Type Runner (PPO): $(date) ===" | tee "$LOG_FILE"
+echo "Models: ${MODEL_NAMES[*]} | Tasks: ${TASKS[*]} | Steps: ${STEPS} | GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL}" | tee -a "$LOG_FILE"
+echo "GPUS: ${GPUS[*]} | groups: ${GPU_GROUPS[*]} | cooldown=${COOLDOWN_SECONDS}s" | tee -a "$LOG_FILE"
+
+get_config_for_task() {
+    case "$1" in
+        countdown) echo "_4_countdown" ;;
+        sokoban) echo "_2_sokoban" ;;
+        frozenlake) echo "_3_frozen_lake" ;;
+        webshop) echo "_6_webshop" ;;
+        metamathqa) echo "_5_metamathqa" ;;
+        *) echo "" ;;
+    esac
+}
+
+run_experiment() {
+    local task=$1
+    local model_name=$2
+    local filter=$3
+    local config=$4
+    local gpu_list=$5
+
+    local model_path
+    model_path=$(get_model_path "$model_name")
+    local algo="PPO"
+    local safe_model_name="${model_name//\//__}"
+
+    local filter_value
+    if [ "$filter" = "filter" ]; then
+        filter_value=0.9
+    else
+        filter_value=1.0
+    fi
+    local filter_strategy="top_p"
+
+    local common_overrides=(
+        "actor_rollout_ref.actor.use_kl_loss=False"
+        "actor_rollout_ref.actor.kl_loss_type=low-var-kl"
+        "actor_rollout_ref.actor.kl_loss_coef=0.001"
+        "actor_rollout_ref.actor.entropy_coeff=0.001"
+        "actor_rollout_ref.actor.entropy_from_logits_with_chunking=True"
+        "actor_rollout_ref.actor.filter_loss_scaling=none"
+        "actor_rollout_ref.rollout.gpu_memory_utilization=${GPU_MEMORY_UTILIZATION}"
+        "actor_rollout_ref.rollout.rollout_filter_strategy=${filter_strategy}"
+        "actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=softmax"
+        "actor_rollout_ref.rollout.rollout_filter_type=largest"
+        "actor_rollout_ref.rollout.rollout_filter_metric=reward_variance"
+        "actor_rollout_ref.rollout.rollout_filter_include_zero=True"
+    )
+
+    local env_overrides=()
+    if [ "$task" = "frozenlake" ]; then
+        env_overrides+=("custom_envs.CoordFrozenLake.env_config.success_rate=1.0")
+    fi
+
+    local checkpoint_overrides=(
+        "actor_rollout_ref.actor.checkpoint.save_contents=[model]"
+        "critic.checkpoint.save_contents=[model]"
+    )
+
+    # PPO algorithm overrides
+    local algo_overrides="algorithm.adv_estimator=gae actor_rollout_ref.actor.loss_agg_mode=token-mean"
+    read -r -a algo_args <<< "$algo_overrides"
+
+    local name="${task}-${algo}-${filter}-${safe_model_name}"
+    local task_dir="${RESULT_ROOT}/diff_model_${task}"
+    local log_path="${task_dir}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${task}/${safe_model_name}/${filter}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    mkdir -p "$task_dir"
+    mkdir -p "${checkpoint_dir}"
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$config" \
+        model_path="${model_path}" \
+        trainer.project_name="ragen_main_table_diff_model" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.experiment_name="${name}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        ray_kwargs.ray_init.num_cpus="${RAY_NUM_CPUS}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        actor_rollout_ref.rollout.rollout_filter_value="${filter_value}" \
+        "${common_overrides[@]}" \
+        "${env_overrides[@]}" \
+        "${checkpoint_overrides[@]}" \
+        "${algo_args[@]}" \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+    END=$(date +%s)
+
+    TOTAL_TIME=$((END - START))
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\s]+([\d.]+)",
+    r"timing_s/eval_total[:\s]+([\d.]+)",
+    r"timing_s/total[:\s]+([\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="task=${task} | algo=${algo} | filter=${filter} | model=${model_name} | steps=${STEPS} | filter=${filter_strategy}:${filter_value} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${task_dir}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+    return 0
+}
+
+EXPERIMENTS=()
+GROUP_LABELS=()
+CURRENT_GROUP=""
+
+resolve_filter_selection() {
+    local raw="$1"
+    if [ -z "$raw" ] || [ "$raw" = "all" ]; then
+        SELECTED_FILTERS=("${FILTER_MODES[@]}")
+        return
+    fi
+    IFS=',' read -r -a candidates <<< "$raw"
+    SELECTED_FILTERS=()
+    for candidate in "${candidates[@]}"; do
+        candidate="${candidate// /}"
+        case "$candidate" in
+            filter|nofilter)
+                SELECTED_FILTERS+=("$candidate")
+                ;;
+            "")
+                continue
+                ;;
+            *)
+                echo "Unknown filter mode: $candidate" >&2
+                exit 1
+                ;;
+        esac
+    done
+    if [ ${#SELECTED_FILTERS[@]} -eq 0 ]; then
+        echo "No valid filters selected via --filters" >&2
+        exit 1
+    fi
+}
+
+set_group() {
+    CURRENT_GROUP="$1"
+    GROUP_LABELS+=("$1")
+}
+
+add_experiment() {
+    local task=$1
+    local model_name=$2
+    local filter=$3
+    local config=$4
+    EXPERIMENTS+=("${CURRENT_GROUP}|${task}|${model_name}|${filter}|${config}")
+}
+
+resolve_filter_selection "$FILTERS_OPTION"
+
+for model_name in "${MODEL_NAMES[@]}"; do
+    set_group "Model: ${model_name}"
+    for task in "${TASKS[@]}"; do
+        config=$(get_config_for_task "$task")
+        if [ -z "$config" ]; then
+            echo "Unknown task: $task" >&2
+            exit 1
+        fi
+        for filter in "${SELECTED_FILTERS[@]}"; do
+            add_experiment "$task" "$model_name" "$filter" "$config"
+        done
+    done
+done
+
+QUEUE_FILE=$(mktemp -t ragen_model_type_queue.XXXXXX)
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+echo 0 > "$QUEUE_FILE"
+USE_FLOCK=false
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$$" -ne "$MAIN_PID" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list=$1
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        local exp="${EXPERIMENTS[$idx]}"
+        IFS='|' read -r exp_group task model_name filter config <<< "$exp"
+        run_experiment "$task" "$model_name" "$filter" "$config" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== Grouped Summary ==="
+    echo "GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL} | Algorithm: PPO | Steps: ${STEPS}"
+    for group_label in "${GROUP_LABELS[@]}"; do
+        echo "=== ${group_label} ==="
+        for exp in "${EXPERIMENTS[@]}"; do
+            IFS='|' read -r exp_group task model_name filter config <<< "$exp"
+            if [ "$exp_group" != "$group_label" ]; then
+                continue
+            fi
+            local safe_model_name="${model_name//\//__}"
+            name="${task}-PPO-${filter}-${safe_model_name}"
+            task_dir="${RESULT_ROOT}/diff_model_${task}"
+            if [ -f "${task_dir}/${name}.result" ]; then
+                cat "${task_dir}/${name}.result"
+            else
+                echo "task=${task} | algo=PPO | filter=${filter} | model=${model_name} | status=missing"
+            fi
+        done
+    done
+} | tee -a "$LOG_FILE"

scripts/runs/run_main_table_diff_size.sh

@@ -0,0 +1,524 @@
+#!/bin/bash
+# Model Size: Different model sizes (0.5B/1.5B/3B/7B) × tasks × filter/no-filter
+# Algorithm: PPO only
+# Filtering rule: filter => top_p=0.9, nofilter => top_p=1.0
+
+set -euo pipefail
+
+# Defaults
+STEPS=400
+MODEL_NAMES=("Qwen2.5-0.5B" "Qwen2.5-1.5B" "Qwen2.5-3B" "Qwen2.5-7B")
+TASKS=("sokoban" "frozenlake" "webshop" "metamathqa" "countdown")
+SAVE_FREQ=-1
+FILTER_MODES=("filter" "nofilter")
+FILTERS_OPTION="all"
+SELECTED_FILTERS=("${FILTER_MODES[@]}")
+
+# GPU settings
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=30
+GPU_MEMORY_UTILIZATION=0.3
+RAY_NUM_CPUS=16
+declare -A GPU_LABELS
+
+usage() {
+    echo "Usage: $0 [options]"
+    echo "Options:"
+    echo "  --steps N             Training steps (default: 400)"
+    echo "  --tasks LIST          Comma-separated tasks (default: sokoban,frozenlake,webshop,metamathqa,countdown)"
+    echo "  --models LIST         Comma-separated model names (default: Qwen2.5-0.5B,Qwen2.5-1.5B,Qwen2.5-3B,Qwen2.5-7B)"
+    echo "  --gpus LIST           Comma-separated GPU IDs (default: auto-detect)"
+    echo "  --gpus-per-exp N      GPUs per experiment (default: 1)"
+    echo "  --cooldown SECONDS    Cooldown between runs on the same GPU group (default: 30)"
+    echo "  --gpu-memory-utilization V  Rollout gpu_memory_utilization (default: 0.3)"
+    echo "  --ray-num-cpus N      Max CPUs per task for ray.init (default: 16)"
+    echo "  --save-freq N         Checkpoint save frequency (default: -1 to disable saving)"
+    echo "  --filters LIST        Comma-separated filter modes (filter,nofilter,all). Default: all"
+    echo "  -h, --help            Show this help"
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --tasks) IFS=',' read -r -a TASKS <<< "$2"; shift 2 ;;
+        --tasks=*) IFS=',' read -r -a TASKS <<< "${1#*=}"; shift ;;
+        --models) IFS=',' read -r -a MODEL_NAMES <<< "$2"; shift 2 ;;
+        --models=*) IFS=',' read -r -a MODEL_NAMES <<< "${1#*=}"; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --cooldown) COOLDOWN_SECONDS="$2"; shift 2 ;;
+        --cooldown=*) COOLDOWN_SECONDS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --ray-num-cpus) RAY_NUM_CPUS="$2"; shift 2 ;;
+        --ray-num-cpus=*) RAY_NUM_CPUS="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        --filters) FILTERS_OPTION="$2"; shift 2 ;;
+        --filters=*) FILTERS_OPTION="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+if ! [[ "$RAY_NUM_CPUS" =~ ^[0-9]+$ ]] || [ "$RAY_NUM_CPUS" -lt 1 ]; then
+    echo "Error: --ray-num-cpus must be a positive integer"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+NUM_SLOTS=${#GPU_GROUPS[@]}
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_model_label() {
+    local models=()
+    local id label model
+    for id in "${GPUS[@]}"; do
+        label=$(get_gpu_label "$id")
+        model="${label#1x}"
+        models+=("$model")
+    done
+    local unique_models=()
+    local m found
+    for m in "${models[@]}"; do
+        found=false
+        for u in "${unique_models[@]}"; do
+            if [ "$u" = "$m" ]; then
+                found=true
+                break
+            fi
+        done
+        if [ "$found" = false ]; then
+            unique_models+=("$m")
+        fi
+    done
+    if [ ${#unique_models[@]} -eq 1 ]; then
+        echo "${unique_models[0]}"
+    else
+        echo "mixed"
+    fi
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+GPU_MODEL_LABEL=$(get_gpu_model_label)
+GPU_LOG_LABEL="${GPUS_PER_EXP}x${GPU_MODEL_LABEL}"
+LOG_FILE="logs/diff_size_PPO.log"
+RESULT_ROOT="logs"
+CHECKPOINT_ROOT="model_saving/diff_size"
+
+mkdir -p logs
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== Model Size Runner (PPO): $(date) ===" | tee "$LOG_FILE"
+echo "Models: ${MODEL_NAMES[*]} | Tasks: ${TASKS[*]} | Steps: ${STEPS} | GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL}" | tee -a "$LOG_FILE"
+echo "GPUS: ${GPUS[*]} | groups: ${GPU_GROUPS[*]} | cooldown=${COOLDOWN_SECONDS}s" | tee -a "$LOG_FILE"
+
+get_config_for_task() {
+    case "$1" in
+        countdown) echo "_4_countdown" ;;
+        sokoban) echo "_2_sokoban" ;;
+        frozenlake) echo "_3_frozen_lake" ;;
+        webshop) echo "_6_webshop" ;;
+        metamathqa) echo "_5_metamathqa" ;;
+        *) echo "" ;;
+    esac
+}
+
+run_experiment() {
+    local task=$1
+    local model_name=$2
+    local filter=$3
+    local config=$4
+    local gpu_list=$5
+
+    local model_path="Qwen/${model_name}"
+    local algo="PPO"
+
+    local filter_value
+    if [ "$filter" = "filter" ]; then
+        filter_value=0.9
+    else
+        filter_value=1.0
+    fi
+    local filter_strategy="top_p"
+
+    local common_overrides=(
+        "actor_rollout_ref.actor.use_kl_loss=False"
+        "actor_rollout_ref.actor.kl_loss_type=low-var-kl"
+        "actor_rollout_ref.actor.kl_loss_coef=0.001"
+        "actor_rollout_ref.actor.entropy_coeff=0.001"
+        "actor_rollout_ref.actor.entropy_from_logits_with_chunking=True"
+        "actor_rollout_ref.actor.filter_loss_scaling=none"
+        "actor_rollout_ref.rollout.gpu_memory_utilization=${GPU_MEMORY_UTILIZATION}"
+        "actor_rollout_ref.rollout.rollout_filter_strategy=${filter_strategy}"
+        "actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=softmax"
+        "actor_rollout_ref.rollout.rollout_filter_type=largest"
+        "actor_rollout_ref.rollout.rollout_filter_metric=reward_variance"
+        "actor_rollout_ref.rollout.rollout_filter_include_zero=True"
+    )
+
+    local env_overrides=()
+    if [ "$task" = "frozenlake" ]; then
+        env_overrides+=("custom_envs.CoordFrozenLake.env_config.success_rate=1.0")
+    fi
+
+    local checkpoint_overrides=(
+        "actor_rollout_ref.actor.checkpoint.save_contents=[model]"
+        "critic.checkpoint.save_contents=[model]"
+    )
+
+    # PPO algorithm overrides
+    local algo_overrides="algorithm.adv_estimator=gae actor_rollout_ref.actor.loss_agg_mode=token-mean"
+    read -r -a algo_args <<< "$algo_overrides"
+
+    local name="${task}-${algo}-${filter}-${model_name}"
+    local task_dir="${RESULT_ROOT}/diff_size_${task}"
+    local log_path="${task_dir}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${task}/${model_name}/${filter}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    mkdir -p "$task_dir"
+    mkdir -p "${checkpoint_dir}"
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$config" \
+        model_path="${model_path}" \
+        trainer.project_name="ragen_main_table_diff_size" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.experiment_name="${name}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        ray_kwargs.ray_init.num_cpus="${RAY_NUM_CPUS}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        actor_rollout_ref.rollout.rollout_filter_value="${filter_value}" \
+        "${common_overrides[@]}" \
+        "${env_overrides[@]}" \
+        "${checkpoint_overrides[@]}" \
+        "${algo_args[@]}" \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+    END=$(date +%s)
+
+    TOTAL_TIME=$((END - START))
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\s]+([\d.]+)",
+    r"timing_s/eval_total[:\s]+([\d.]+)",
+    r"timing_s/total[:\s]+([\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="task=${task} | algo=${algo} | filter=${filter} | model=${model_name} | steps=${STEPS} | filter=${filter_strategy}:${filter_value} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${task_dir}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+    return 0
+}
+
+EXPERIMENTS=()
+GROUP_LABELS=()
+CURRENT_GROUP=""
+
+resolve_filter_selection() {
+    local raw="$1"
+    if [ -z "$raw" ] || [ "$raw" = "all" ]; then
+        SELECTED_FILTERS=("${FILTER_MODES[@]}")
+        return
+    fi
+    IFS=',' read -r -a candidates <<< "$raw"
+    SELECTED_FILTERS=()
+    for candidate in "${candidates[@]}"; do
+        candidate="${candidate// /}"
+        case "$candidate" in
+            filter|nofilter)
+                SELECTED_FILTERS+=("$candidate")
+                ;;
+            "")
+                continue
+                ;;
+            *)
+                echo "Unknown filter mode: $candidate" >&2
+                exit 1
+                ;;
+        esac
+    done
+    if [ ${#SELECTED_FILTERS[@]} -eq 0 ]; then
+        echo "No valid filters selected via --filters" >&2
+        exit 1
+    fi
+}
+
+set_group() {
+    CURRENT_GROUP="$1"
+    GROUP_LABELS+=("$1")
+}
+
+add_experiment() {
+    local task=$1
+    local model_name=$2
+    local filter=$3
+    local config=$4
+    EXPERIMENTS+=("${CURRENT_GROUP}|${task}|${model_name}|${filter}|${config}")
+}
+
+resolve_filter_selection "$FILTERS_OPTION"
+
+for model_name in "${MODEL_NAMES[@]}"; do
+    set_group "Model: ${model_name}"
+    for task in "${TASKS[@]}"; do
+        config=$(get_config_for_task "$task")
+        if [ -z "$config" ]; then
+            echo "Unknown task: $task" >&2
+            exit 1
+        fi
+        for filter in "${SELECTED_FILTERS[@]}"; do
+            add_experiment "$task" "$model_name" "$filter" "$config"
+        done
+    done
+done
+
+QUEUE_FILE=$(mktemp -t ragen_model_size_queue.XXXXXX)
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+echo 0 > "$QUEUE_FILE"
+USE_FLOCK=false
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$$" -ne "$MAIN_PID" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list=$1
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        local exp="${EXPERIMENTS[$idx]}"
+        IFS='|' read -r exp_group task model_name filter config <<< "$exp"
+        run_experiment "$task" "$model_name" "$filter" "$config" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== Grouped Summary ==="
+    echo "GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL} | Algorithm: PPO | Steps: ${STEPS}"
+    for group_label in "${GROUP_LABELS[@]}"; do
+        echo "=== ${group_label} ==="
+        for exp in "${EXPERIMENTS[@]}"; do
+            IFS='|' read -r exp_group task model_name filter config <<< "$exp"
+            if [ "$exp_group" != "$group_label" ]; then
+                continue
+            fi
+            name="${task}-PPO-${filter}-${model_name}"
+            task_dir="${RESULT_ROOT}/diff_size_${task}"
+            if [ -f "${task_dir}/${name}.result" ]; then
+                cat "${task_dir}/${name}.result"
+            else
+                echo "task=${task} | algo=PPO | filter=${filter} | model=${model_name} | status=missing"
+            fi
+        done
+    done
+} | tee -a "$LOG_FILE"

scripts/runs/run_search_benchmark.sh

@@ -0,0 +1,542 @@
+#!/bin/bash
+# Search Benchmark: models × algorithms × filter/no-filter on SearchQA (HotpotQA + Dense Retrieval)
+# Models: Qwen2.5-3B-Instruct, Qwen2.5-7B-Instruct, Llama-3.2-3B-Instruct
+# Algorithms: PPO, GRPO
+# Filtering rule: filter => top_p=0.9, nofilter => top_p=1.0
+
+set -euo pipefail
+
+# Activate ragen conda environment
+eval "$(conda shell.bash hook 2>/dev/null || true)"
+conda activate ragen 2>/dev/null || true
+
+# Use user-writable datasets cache to avoid permission conflicts with root-owned lock files
+export HF_DATASETS_CACHE="${HF_DATASETS_CACHE:-${HOME}/.hf_cache/datasets}"
+
+# Defaults
+STEPS=200
+MODEL_NAMES=("Qwen2.5-3B-Instruct")
+ALGORITHMS=("PPO")
+SAVE_FREQ=-1
+FILTER_STRATEGY="top_p"
+FILTER_VALUE=""
+
+# GPU settings
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=30
+GPU_MEMORY_UTILIZATION=0.6
+TENSOR_PARALLEL_SIZE=1
+MICRO_BATCH_SIZE=""
+MINI_BATCH_SIZE=""
+COLLAPSE_FREQ=""
+RETRIEVAL_PORT=""
+declare -A GPU_LABELS
+
+usage() {
+    echo "Usage: $0 [options]"
+    echo "Options:"
+    echo "  --steps N             Training steps (default: 400)"
+    echo "  --models LIST         Comma-separated model names (default: Qwen2.5-3B-Instruct)"
+    echo "  --algos LIST          Comma-separated algorithms: PPO,GRPO (default: PPO)"
+    echo "  --gpus LIST           Comma-separated GPU IDs (default: auto-detect)"
+    echo "  --gpus-per-exp N      GPUs per experiment (default: 1)"
+    echo "  --cooldown SECONDS    Cooldown between runs on the same GPU group (default: 30)"
+    echo "  --gpu-memory-utilization V  Rollout gpu_memory_utilization (default: 0.6)"
+    echo "  --tp N                Tensor parallel size for vLLM rollout (default: 1)"
+    echo "  --micro-batch N       micro_batch_size_per_gpu override"
+    echo "  --mini-batch N        ppo_mini_batch_size override"
+    echo "  --collapse-freq N     collapse_detection.compute_freq override"
+    echo "  --save-freq N         Checkpoint save frequency (default: -1 to disable saving)"
+    echo "  --retrieval-port N    Retrieval server port (overrides default 8000 in config)"
+    echo "  --filter-strategy S   Rollout filter strategy: top_p, top_k, etc. (default: top_p)"
+    echo "  --filter-value V     Rollout filter value (default: 1.0 for no filtering)"
+    echo "  -h, --help            Show this help"
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --models) IFS=',' read -r -a MODEL_NAMES <<< "$2"; shift 2 ;;
+        --models=*) IFS=',' read -r -a MODEL_NAMES <<< "${1#*=}"; shift ;;
+        --algos) IFS=',' read -r -a ALGORITHMS <<< "$2"; shift 2 ;;
+        --algos=*) IFS=',' read -r -a ALGORITHMS <<< "${1#*=}"; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --cooldown) COOLDOWN_SECONDS="$2"; shift 2 ;;
+        --cooldown=*) COOLDOWN_SECONDS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        --tp) TENSOR_PARALLEL_SIZE="$2"; shift 2 ;;
+        --tp=*) TENSOR_PARALLEL_SIZE="${1#*=}"; shift ;;
+        --micro-batch) MICRO_BATCH_SIZE="$2"; shift 2 ;;
+        --micro-batch=*) MICRO_BATCH_SIZE="${1#*=}"; shift ;;
+        --mini-batch) MINI_BATCH_SIZE="$2"; shift 2 ;;
+        --mini-batch=*) MINI_BATCH_SIZE="${1#*=}"; shift ;;
+        --collapse-freq) COLLAPSE_FREQ="$2"; shift 2 ;;
+        --collapse-freq=*) COLLAPSE_FREQ="${1#*=}"; shift ;;
+        --retrieval-port) RETRIEVAL_PORT="$2"; shift 2 ;;
+        --retrieval-port=*) RETRIEVAL_PORT="${1#*=}"; shift ;;
+        --filter-strategy) FILTER_STRATEGY="$2"; shift 2 ;;
+        --filter-strategy=*) FILTER_STRATEGY="${1#*=}"; shift ;;
+        --filter-value) FILTER_VALUE="$2"; shift 2 ;;
+        --filter-value=*) FILTER_VALUE="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+# Default filter value: 1.0 (no filtering) if not specified
+if [ -z "$FILTER_VALUE" ]; then
+    FILTER_VALUE=1.0
+fi
+
+# Derive a short filter label for experiment naming / logs
+if [ "$FILTER_VALUE" = "1.0" ] || [ "$FILTER_VALUE" = "1" ]; then
+    FILTER_LABEL="nofilter"
+else
+    FILTER_LABEL="${FILTER_STRATEGY}${FILTER_VALUE}"
+fi
+
+# Fixed: search task config
+CONFIG="_9_search"
+
+# Map model names to HuggingFace paths
+get_model_path() {
+    if [[ "$1" == *"/"* ]]; then
+        echo "$1"
+        return
+    fi
+    case "$1" in
+        Qwen2.5-3B-Instruct) echo "Qwen/Qwen2.5-3B-Instruct" ;;
+        Qwen2.5-7B-Instruct) echo "Qwen/Qwen2.5-7B-Instruct" ;;
+        Llama-3.2-3B-Instruct) echo "meta-llama/Llama-3.2-3B-Instruct" ;;
+        *) echo "Qwen/$1" ;;
+    esac
+}
+
+# Algorithm-specific overrides
+get_algo_overrides() {
+    case "$1" in
+        PPO)
+            echo "algorithm.adv_estimator=gae actor_rollout_ref.actor.loss_agg_mode=token-mean"
+            ;;
+        GRPO)
+            echo "algorithm.adv_estimator=grpo algorithm.norm_adv_by_std_in_grpo=True actor_rollout_ref.actor.loss_agg_mode=seq-mean-token-mean"
+            ;;
+        *)
+            echo ""
+            ;;
+    esac
+}
+
+# Auto-detect GPUs
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+NUM_SLOTS=${#GPU_GROUPS[@]}
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_model_label() {
+    local models=()
+    local id label model
+    for id in "${GPUS[@]}"; do
+        label=$(get_gpu_label "$id")
+        model="${label#1x}"
+        models+=("$model")
+    done
+    local unique_models=()
+    local m found
+    for m in "${models[@]}"; do
+        found=false
+        for u in "${unique_models[@]}"; do
+            if [ "$u" = "$m" ]; then
+                found=true
+                break
+            fi
+        done
+        if [ "$found" = false ]; then
+            unique_models+=("$m")
+        fi
+    done
+    if [ ${#unique_models[@]} -eq 1 ]; then
+        echo "${unique_models[0]}"
+    else
+        echo "mixed"
+    fi
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+GPU_MODEL_LABEL=$(get_gpu_model_label)
+GPU_LOG_LABEL="${GPUS_PER_EXP}x${GPU_MODEL_LABEL}"
+LOG_DIR="logs/search_benchmark"
+LOG_FILE="logs/search_benchmark.log"
+RESULT_ROOT="logs/search_benchmark"
+CHECKPOINT_ROOT="model_saving/search_benchmark"
+
+mkdir -p "$LOG_DIR"
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== Search Benchmark Runner: $(date) ===" | tee "$LOG_FILE"
+echo "Models: ${MODEL_NAMES[*]} | Algos: ${ALGORITHMS[*]} | Filter: ${FILTER_STRATEGY}:${FILTER_VALUE} | Steps: ${STEPS} | GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL}" | tee -a "$LOG_FILE"
+echo "GPUS: ${GPUS[*]} | groups: ${GPU_GROUPS[*]} | cooldown=${COOLDOWN_SECONDS}s" | tee -a "$LOG_FILE"
+
+run_experiment() {
+    local model_name=$1
+    local algo=$2
+    local gpu_list=$3
+
+    local model_path
+    model_path=$(get_model_path "$model_name")
+
+    local filter_strategy="$FILTER_STRATEGY"
+    local filter_value="$FILTER_VALUE"
+
+    local common_overrides=(
+        "actor_rollout_ref.actor.use_kl_loss=False"
+        "actor_rollout_ref.actor.kl_loss_type=low-var-kl"
+        "actor_rollout_ref.actor.kl_loss_coef=0.001"
+        "actor_rollout_ref.actor.entropy_coeff=0.001"
+        "actor_rollout_ref.actor.entropy_from_logits_with_chunking=True"
+        "actor_rollout_ref.actor.filter_loss_scaling=none"
+        "actor_rollout_ref.rollout.gpu_memory_utilization=${GPU_MEMORY_UTILIZATION}"
+        "actor_rollout_ref.rollout.tensor_model_parallel_size=${TENSOR_PARALLEL_SIZE}"
+    )
+
+    if [ -n "$MICRO_BATCH_SIZE" ]; then
+        common_overrides+=("micro_batch_size_per_gpu=${MICRO_BATCH_SIZE}")
+    fi
+    if [ -n "$MINI_BATCH_SIZE" ]; then
+        common_overrides+=("ppo_mini_batch_size=${MINI_BATCH_SIZE}")
+    fi
+    if [ -n "$COLLAPSE_FREQ" ]; then
+        common_overrides+=("collapse_detection.compute_freq=${COLLAPSE_FREQ}")
+    fi
+
+    local checkpoint_overrides=(
+        "actor_rollout_ref.actor.checkpoint.save_contents=[model]"
+        "critic.checkpoint.save_contents=[model]"
+    )
+
+    local retrieval_overrides=()
+    if [ -n "$RETRIEVAL_PORT" ]; then
+        retrieval_overrides+=("+custom_envs.SearchQA.env_config.retrieval_server_url=http://127.0.0.1:${RETRIEVAL_PORT}")
+    fi
+
+    local algo_overrides
+    algo_overrides=$(get_algo_overrides "$algo")
+    read -r -a algo_args <<< "$algo_overrides"
+
+    local name="search-${algo}-${FILTER_LABEL}-${model_name}"
+    local log_path="${LOG_DIR}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${model_name}/${algo}/${FILTER_LABEL}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    # Limit Ray CPU workers to avoid spawning hundreds of idle workers (default: 8 per GPU)
+    local ray_num_cpus=$((gpus_per_exp * 8))
+    common_overrides+=("ray_kwargs.ray_init.num_cpus=${ray_num_cpus}")
+
+    mkdir -p "${checkpoint_dir}"
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$CONFIG" \
+        model_path="${model_path}" \
+        trainer.project_name="ragen_search_benchmark" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.experiment_name="${name}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        actor_rollout_ref.rollout.rollout_filter_strategy="${filter_strategy}" \
+        actor_rollout_ref.rollout.rollout_filter_value="${filter_value}" \
+        "${common_overrides[@]}" \
+        "${checkpoint_overrides[@]}" \
+        "${retrieval_overrides[@]}" \
+        "${algo_args[@]}" \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+    END=$(date +%s)
+
+    TOTAL_TIME=$((END - START))
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\s]+([\d.]+)",
+    r"timing_s/eval_total[:\s]+([\d.]+)",
+    r"timing_s/total[:\s]+([\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="task=search | algo=${algo} | filter=${FILTER_LABEL} | model=${model_name} | steps=${STEPS} | strategy=${filter_strategy}:${filter_value} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${LOG_DIR}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+    return 0
+}
+
+EXPERIMENTS=()
+GROUP_LABELS=()
+CURRENT_GROUP=""
+
+set_group() {
+    CURRENT_GROUP="$1"
+    GROUP_LABELS+=("$1")
+}
+
+add_experiment() {
+    local model_name=$1
+    local algo=$2
+    EXPERIMENTS+=("${CURRENT_GROUP}|${model_name}|${algo}")
+}
+
+# Build experiment list: models × algos (filter is a global setting)
+for model_name in "${MODEL_NAMES[@]}"; do
+    for algo in "${ALGORITHMS[@]}"; do
+        set_group "${model_name} / ${algo}"
+        add_experiment "$model_name" "$algo"
+    done
+done
+
+QUEUE_FILE=$(mktemp -t ragen_search_bench_queue.XXXXXX)
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+echo 0 > "$QUEUE_FILE"
+USE_FLOCK=false
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$$" -ne "$MAIN_PID" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list=$1
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        local exp="${EXPERIMENTS[$idx]}"
+        IFS='|' read -r exp_group model_name algo <<< "$exp"
+        run_experiment "$model_name" "$algo" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== Grouped Summary ==="
+    echo "GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL} | Steps: ${STEPS}"
+    for group_label in "${GROUP_LABELS[@]}"; do
+        echo "=== ${group_label} ==="
+        for exp in "${EXPERIMENTS[@]}"; do
+            IFS='|' read -r exp_group model_name algo <<< "$exp"
+            if [ "$exp_group" != "$group_label" ]; then
+                continue
+            fi
+            name="search-${algo}-${FILTER_LABEL}-${model_name}"
+            if [ -f "${LOG_DIR}/${name}.result" ]; then
+                cat "${LOG_DIR}/${name}.result"
+            else
+                echo "task=search | algo=${algo} | filter=${FILTER_LABEL} | model=${model_name} | status=missing"
+            fi
+        done
+    done
+} | tee -a "$LOG_FILE"

scripts/runs/run_sokoban_ppo_filter_grad_analysis.sh

@@ -0,0 +1,205 @@
+#!/bin/bash
+# Runner for Sokoban PPO/GRPO with top-p=0.9 filtering plus periodic
+# gradient-analysis passes during training.
+#
+# This intentionally stays close to scripts/runs/run_main_table_diff_algo.sh:
+# - task: sokoban
+# - algo: PPO or GRPO
+# - filter: top_p=0.9
+# - model: Qwen2.5-3B
+# - train batch: 8 groups x 16 samples (same as the main-table setup)
+# - gradient-analysis batch: 128 groups x 16 samples
+#
+# Validation policy:
+# - one validation before training
+# - periodic validation every 10 steps
+#
+# Gradient-analysis policy:
+# - run gradient analysis every 50 steps
+# - current trainer trigger is step 1, 51, 101, ...
+# - do not exit after the analysis step
+
+set -euo pipefail
+
+STEPS=101
+SAVE_FREQ=100
+GPU_MEMORY_UTILIZATION=0.3
+RAY_NUM_CPUS=16
+PPO_MICRO_BATCH_SIZE_PER_GPU=4
+LOG_PROB_MICRO_BATCH_SIZE_PER_GPU=4
+GPUS=()
+GPUS_PROVIDED=false
+
+MODEL_NAME="Qwen2.5-3B"
+MODEL_PATH="Qwen/${MODEL_NAME}"
+TASK="sokoban"
+ALGO="PPO"
+FILTER_LABEL="filter"
+FILTER_VALUE="0.9"
+GROUP_SIZE=16
+ENV_GROUPS=8
+ANALYSIS_GROUP_SIZE=16
+ANALYSIS_ENV_GROUPS=128
+CONFIG_NAME="_2_sokoban"
+
+normalize_algo() {
+    case "$1" in
+        PPO|ppo) echo "PPO" ;;
+        GRPO|grpo) echo "GRPO" ;;
+        *)
+            echo "Error: unsupported --algo '$1'. Supported values: PPO, GRPO" >&2
+            exit 1
+            ;;
+    esac
+}
+
+get_algo_overrides() {
+    case "$1" in
+        PPO)
+            echo "algorithm.adv_estimator=gae actor_rollout_ref.actor.loss_agg_mode=token-mean"
+            ;;
+        GRPO)
+            echo "algorithm.adv_estimator=grpo algorithm.norm_adv_by_std_in_grpo=True actor_rollout_ref.actor.loss_agg_mode=seq-mean-token-mean"
+            ;;
+        *)
+            echo "Error: unsupported algorithm '$1'" >&2
+            exit 1
+            ;;
+    esac
+}
+
+usage() {
+    echo "Usage: $0 [options]"
+    echo "Options:"
+    echo "  --algo NAME                 Training algorithm: PPO or GRPO (default: PPO)"
+    echo "  --steps N                    Training steps (default: 101)"
+    echo "  --gpus LIST                  Comma-separated GPU IDs (default: auto-detect)"
+    echo "  --gpu-memory-utilization V   Rollout gpu_memory_utilization (default: 0.3)"
+    echo "  --ray-num-cpus N             Max CPUs for ray.init (default: 16)"
+    echo "  --ppo-micro-batch-size-per-gpu N      PPO micro batch size per GPU for actor/critic (default: 4)"
+    echo "  --log-prob-micro-batch-size-per-gpu N log-prob micro batch size per GPU for ref/rollout (default: 4)"
+    echo "  --save-freq N                Checkpoint save frequency (default: 100)"
+    echo "  -h, --help                   Show this help"
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --algo) ALGO="$(normalize_algo "$2")"; shift 2 ;;
+        --algo=*) ALGO="$(normalize_algo "${1#*=}")"; shift ;;
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --ray-num-cpus) RAY_NUM_CPUS="$2"; shift 2 ;;
+        --ray-num-cpus=*) RAY_NUM_CPUS="${1#*=}"; shift ;;
+        --ppo-micro-batch-size-per-gpu) PPO_MICRO_BATCH_SIZE_PER_GPU="$2"; shift 2 ;;
+        --ppo-micro-batch-size-per-gpu=*) PPO_MICRO_BATCH_SIZE_PER_GPU="${1#*=}"; shift ;;
+        --log-prob-micro-batch-size-per-gpu) LOG_PROB_MICRO_BATCH_SIZE_PER_GPU="$2"; shift 2 ;;
+        --log-prob-micro-batch-size-per-gpu=*) LOG_PROB_MICRO_BATCH_SIZE_PER_GPU="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$RAY_NUM_CPUS" =~ ^[0-9]+$ ]] || [ "$RAY_NUM_CPUS" -lt 1 ]; then
+    echo "Error: --ray-num-cpus must be a positive integer" >&2
+    exit 1
+fi
+if ! [[ "$PPO_MICRO_BATCH_SIZE_PER_GPU" =~ ^[0-9]+$ ]] || [ "$PPO_MICRO_BATCH_SIZE_PER_GPU" -lt 1 ]; then
+    echo "Error: --ppo-micro-batch-size-per-gpu must be a positive integer" >&2
+    exit 1
+fi
+if ! [[ "$LOG_PROB_MICRO_BATCH_SIZE_PER_GPU" =~ ^[0-9]+$ ]] || [ "$LOG_PROB_MICRO_BATCH_SIZE_PER_GPU" -lt 1 ]; then
+    echo "Error: --log-prob-micro-batch-size-per-gpu must be a positive integer" >&2
+    exit 1
+fi
+
+ALGO_OVERRIDES=$(get_algo_overrides "$ALGO")
+read -r -a ALGO_ARGS <<< "$ALGO_OVERRIDES"
+
+GPU_LIST=$(IFS=,; echo "${GPUS[*]}")
+NUM_GPUS=${#GPUS[@]}
+if [ "$NUM_GPUS" -lt 1 ]; then
+    echo "Error: no GPUs available" >&2
+    exit 1
+fi
+
+EXP_NAME="${TASK}-${ALGO}-${FILTER_LABEL}-topp09-${MODEL_NAME}-train${ENV_GROUPS}x${GROUP_SIZE}-analysis${ANALYSIS_ENV_GROUPS}x${ANALYSIS_GROUP_SIZE}-grad-every50"
+LOG_DIR="logs/gradient_analysis_${TASK}_${MODEL_NAME}"
+CHECKPOINT_DIR="model_saving/gradient_analysis/${TASK}/${ALGO}/${FILTER_LABEL}/${EXP_NAME}"
+LOG_PATH="${LOG_DIR}/${EXP_NAME}.log"
+
+mkdir -p "$LOG_DIR"
+mkdir -p "$CHECKPOINT_DIR"
+
+echo "=== Gradient Analysis Runner: $(date) ===" | tee "$LOG_PATH"
+echo "task=${TASK} algo=${ALGO} filter=top_p:${FILTER_VALUE} model=${MODEL_NAME} steps=${STEPS} gpus=${GPU_LIST}" | tee -a "$LOG_PATH"
+echo "train_group_size=${GROUP_SIZE} train_env_groups=${ENV_GROUPS} analysis_group_size=${ANALYSIS_GROUP_SIZE} analysis_env_groups=${ANALYSIS_ENV_GROUPS} gradient_analysis_every=50 exit_after_gradient_analysis=False" | tee -a "$LOG_PATH"
+echo "ppo_micro_batch_per_gpu=${PPO_MICRO_BATCH_SIZE_PER_GPU} log_prob_micro_batch_per_gpu=${LOG_PROB_MICRO_BATCH_SIZE_PER_GPU}" | tee -a "$LOG_PATH"
+
+CUDA_VISIBLE_DEVICES="${GPU_LIST}" python train.py --config-name "${CONFIG_NAME}" \
+    model_path="${MODEL_PATH}" \
+    trainer.project_name="ragen_gradient_analysis" \
+    trainer.experiment_name="${EXP_NAME}" \
+    trainer.total_training_steps="${STEPS}" \
+    trainer.save_freq="${SAVE_FREQ}" \
+    trainer.default_local_dir="${CHECKPOINT_DIR}" \
+    trainer.logger="['console','wandb']" \
+    trainer.val_before_train=True \
+    trainer.test_freq=10 \
+    trainer.n_gpus_per_node="${NUM_GPUS}" \
+    ray_kwargs.ray_init.num_cpus="${RAY_NUM_CPUS}" \
+    system.CUDA_VISIBLE_DEVICES="'${GPU_LIST}'" \
+    es_manager.train.env_groups="${ENV_GROUPS}" \
+    es_manager.train.group_size="${GROUP_SIZE}" \
+    es_manager.train.env_configs.n_groups="[${ENV_GROUPS}]" \
+    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu="${PPO_MICRO_BATCH_SIZE_PER_GPU}" \
+    actor_rollout_ref.actor.use_kl_loss=False \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.entropy_coeff=0.001 \
+    actor_rollout_ref.actor.entropy_from_logits_with_chunking=True \
+    actor_rollout_ref.actor.filter_loss_scaling=none \
+    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu="${LOG_PROB_MICRO_BATCH_SIZE_PER_GPU}" \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu="${LOG_PROB_MICRO_BATCH_SIZE_PER_GPU}" \
+    actor_rollout_ref.rollout.gpu_memory_utilization="${GPU_MEMORY_UTILIZATION}" \
+    actor_rollout_ref.rollout.rollout_filter_value="${FILTER_VALUE}" \
+    actor_rollout_ref.rollout.rollout_filter_strategy=top_p \
+    actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=linear \
+    actor_rollout_ref.rollout.rollout_filter_type=largest \
+    actor_rollout_ref.rollout.rollout_filter_metric=reward_variance \
+    actor_rollout_ref.rollout.rollout_filter_include_zero=False \
+    actor_rollout_ref.actor.checkpoint.save_contents=[model] \
+    critic.ppo_micro_batch_size_per_gpu="${PPO_MICRO_BATCH_SIZE_PER_GPU}" \
+    critic.checkpoint.save_contents=[model] \
+    trainer.gradient_analysis_mode=True \
+    trainer.gradient_analysis_every=50 \
+    trainer.gradient_analysis_env_groups="${ANALYSIS_ENV_GROUPS}" \
+    trainer.gradient_analysis_group_size="${ANALYSIS_GROUP_SIZE}" \
+    trainer.gradient_analysis_log_prefilter=True \
+    trainer.exit_after_gradient_analysis=False \
+    actor_rollout_ref.rollout.gradient_analysis_num_buckets=6 \
+    actor_rollout_ref.rollout.gradient_analysis_bucket_mode=quantile \
+    "${ALGO_ARGS[@]}" \
+    2>&1 | tee -a "$LOG_PATH"

scripts/runs/run_sokoban_ppo_filter_grad_analysis_probe_ckpt.sh

@@ -0,0 +1,244 @@
+#!/bin/bash
+# Probe a saved Sokoban PPO/GRPO checkpoint with gradient analysis only.
+#
+# This script resumes from an existing `global_step_*` checkpoint directory and:
+# - does not update critic or actor
+# - runs one gradient-analysis pass
+# - logs both post-filter and pre-filter grad metrics
+#
+# By default it targets the checkpoint layout produced by
+# scripts/runs/run_sokoban_ppo_filter_grad_analysis.sh.
+
+set -euo pipefail
+
+GPU_MEMORY_UTILIZATION=0.3
+RAY_NUM_CPUS=16
+PPO_MICRO_BATCH_SIZE_PER_GPU=4
+LOG_PROB_MICRO_BATCH_SIZE_PER_GPU=4
+GPUS=()
+GPUS_PROVIDED=false
+
+MODEL_NAME="Qwen2.5-3B"
+MODEL_PATH="Qwen/${MODEL_NAME}"
+TASK="sokoban"
+ALGO="PPO"
+FILTER_LABEL="filter"
+FILTER_VALUE="0.9"
+GROUP_SIZE=16
+ENV_GROUPS=8
+ANALYSIS_GROUP_SIZE=16
+ANALYSIS_ENV_GROUPS=128
+CONFIG_NAME="_2_sokoban"
+
+CHECKPOINT_STEP=101
+CHECKPOINT_ROOT=""
+RESUME_FROM_PATH=""
+VAL_BEFORE_TRAIN=false
+
+normalize_algo() {
+    case "$1" in
+        PPO|ppo) echo "PPO" ;;
+        GRPO|grpo) echo "GRPO" ;;
+        *)
+            echo "Error: unsupported --algo '$1'. Supported values: PPO, GRPO" >&2
+            exit 1
+            ;;
+    esac
+}
+
+get_algo_overrides() {
+    case "$1" in
+        PPO)
+            echo "algorithm.adv_estimator=gae actor_rollout_ref.actor.loss_agg_mode=token-mean"
+            ;;
+        GRPO)
+            echo "algorithm.adv_estimator=grpo algorithm.norm_adv_by_std_in_grpo=True actor_rollout_ref.actor.loss_agg_mode=seq-mean-token-mean"
+            ;;
+        *)
+            echo "Error: unsupported algorithm '$1'" >&2
+            exit 1
+            ;;
+    esac
+}
+
+usage() {
+    echo "Usage: $0 [options]"
+    echo "Options:"
+    echo "  --algo NAME                        Training algorithm of the checkpoint: PPO or GRPO (default: PPO)"
+    echo "  --checkpoint-step N                Checkpoint step to probe (default: 101)"
+    echo "  --checkpoint-root DIR              Root directory containing global_step_* checkpoints"
+    echo "  --resume-from-path DIR             Exact global_step_* directory to probe"
+    echo "  --with-val                         Run validation before the gradient-analysis probe"
+    echo "  --gpus LIST                        Comma-separated GPU IDs (default: auto-detect)"
+    echo "  --gpu-memory-utilization V         Rollout gpu_memory_utilization (default: 0.3)"
+    echo "  --ray-num-cpus N                   Max CPUs for ray.init (default: 16)"
+    echo "  --ppo-micro-batch-size-per-gpu N   PPO micro batch size per GPU for actor/critic (default: 4)"
+    echo "  --log-prob-micro-batch-size-per-gpu N"
+    echo "                                     Log-prob micro batch size per GPU for ref/rollout (default: 4)"
+    echo "  -h, --help                         Show this help"
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --algo) ALGO="$(normalize_algo "$2")"; shift 2 ;;
+        --algo=*) ALGO="$(normalize_algo "${1#*=}")"; shift ;;
+        --checkpoint-step) CHECKPOINT_STEP="$2"; shift 2 ;;
+        --checkpoint-step=*) CHECKPOINT_STEP="${1#*=}"; shift ;;
+        --checkpoint-root) CHECKPOINT_ROOT="$2"; shift 2 ;;
+        --checkpoint-root=*) CHECKPOINT_ROOT="${1#*=}"; shift ;;
+        --resume-from-path) RESUME_FROM_PATH="$2"; shift 2 ;;
+        --resume-from-path=*) RESUME_FROM_PATH="${1#*=}"; shift ;;
+        --with-val) VAL_BEFORE_TRAIN=true; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --ray-num-cpus) RAY_NUM_CPUS="$2"; shift 2 ;;
+        --ray-num-cpus=*) RAY_NUM_CPUS="${1#*=}"; shift ;;
+        --ppo-micro-batch-size-per-gpu) PPO_MICRO_BATCH_SIZE_PER_GPU="$2"; shift 2 ;;
+        --ppo-micro-batch-size-per-gpu=*) PPO_MICRO_BATCH_SIZE_PER_GPU="${1#*=}"; shift ;;
+        --log-prob-micro-batch-size-per-gpu) LOG_PROB_MICRO_BATCH_SIZE_PER_GPU="$2"; shift 2 ;;
+        --log-prob-micro-batch-size-per-gpu=*) LOG_PROB_MICRO_BATCH_SIZE_PER_GPU="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$CHECKPOINT_STEP" =~ ^[0-9]+$ ]]; then
+    echo "Error: --checkpoint-step must be a non-negative integer" >&2
+    exit 1
+fi
+if ! [[ "$RAY_NUM_CPUS" =~ ^[0-9]+$ ]] || [ "$RAY_NUM_CPUS" -lt 1 ]; then
+    echo "Error: --ray-num-cpus must be a positive integer" >&2
+    exit 1
+fi
+if ! [[ "$PPO_MICRO_BATCH_SIZE_PER_GPU" =~ ^[0-9]+$ ]] || [ "$PPO_MICRO_BATCH_SIZE_PER_GPU" -lt 1 ]; then
+    echo "Error: --ppo-micro-batch-size-per-gpu must be a positive integer" >&2
+    exit 1
+fi
+if ! [[ "$LOG_PROB_MICRO_BATCH_SIZE_PER_GPU" =~ ^[0-9]+$ ]] || [ "$LOG_PROB_MICRO_BATCH_SIZE_PER_GPU" -lt 1 ]; then
+    echo "Error: --log-prob-micro-batch-size-per-gpu must be a positive integer" >&2
+    exit 1
+fi
+
+SOURCE_EXP_NAME="${TASK}-${ALGO}-${FILTER_LABEL}-topp09-${MODEL_NAME}-train${ENV_GROUPS}x${GROUP_SIZE}-analysis${ANALYSIS_ENV_GROUPS}x${ANALYSIS_GROUP_SIZE}-grad-every50"
+
+if [ -z "$CHECKPOINT_ROOT" ]; then
+    CHECKPOINT_ROOT="model_saving/gradient_analysis/${TASK}/${ALGO}/${FILTER_LABEL}/${SOURCE_EXP_NAME}"
+fi
+
+if [ -z "$RESUME_FROM_PATH" ]; then
+    RESUME_FROM_PATH="${CHECKPOINT_ROOT}/global_step_${CHECKPOINT_STEP}"
+fi
+
+if [[ "$RESUME_FROM_PATH" =~ global_step_([0-9]+)$ ]]; then
+    CHECKPOINT_STEP="${BASH_REMATCH[1]}"
+else
+    echo "Error: --resume-from-path must point to a global_step_* directory" >&2
+    exit 1
+fi
+
+if [ ! -d "$RESUME_FROM_PATH" ]; then
+    echo "Error: checkpoint directory not found: $RESUME_FROM_PATH" >&2
+    exit 1
+fi
+
+ALGO_OVERRIDES=$(get_algo_overrides "$ALGO")
+read -r -a ALGO_ARGS <<< "$ALGO_OVERRIDES"
+
+GPU_LIST=$(IFS=,; echo "${GPUS[*]}")
+NUM_GPUS=${#GPUS[@]}
+if [ "$NUM_GPUS" -lt 1 ]; then
+    echo "Error: no GPUs available" >&2
+    exit 1
+fi
+
+PROBE_TOTAL_STEPS="$CHECKPOINT_STEP"
+if [ "$PROBE_TOTAL_STEPS" -lt 1 ]; then
+    PROBE_TOTAL_STEPS=1
+fi
+
+VAL_LABEL="noval"
+if [ "$VAL_BEFORE_TRAIN" = true ]; then
+    VAL_LABEL="withval"
+fi
+
+EXP_NAME="${TASK}-${ALGO}-${FILTER_LABEL}-topp09-${MODEL_NAME}-probe-ckpt${CHECKPOINT_STEP}-${VAL_LABEL}-kl_and_entropy"
+LOG_DIR="logs/gradient_analysis_probe_${TASK}_${MODEL_NAME}"
+OUTPUT_DIR="model_saving/gradient_analysis_probe/${TASK}/${ALGO}/${FILTER_LABEL}/${EXP_NAME}"
+LOG_PATH="${LOG_DIR}/${EXP_NAME}.log"
+
+mkdir -p "$LOG_DIR"
+mkdir -p "$OUTPUT_DIR"
+
+echo "=== Gradient Analysis Checkpoint Probe: $(date) ===" | tee "$LOG_PATH"
+echo "task=${TASK} algo=${ALGO} model=${MODEL_NAME} ckpt=${RESUME_FROM_PATH} gpus=${GPU_LIST}" | tee -a "$LOG_PATH"
+echo "train_group_size=${GROUP_SIZE} train_env_groups=${ENV_GROUPS} analysis_group_size=${ANALYSIS_GROUP_SIZE} analysis_env_groups=${ANALYSIS_ENV_GROUPS}" | tee -a "$LOG_PATH"
+echo "gradient_analysis_only=True gradient_analysis_every=1 exit_after_gradient_analysis=True val_before_train=${VAL_BEFORE_TRAIN}" | tee -a "$LOG_PATH"
+echo "ppo_micro_batch_per_gpu=${PPO_MICRO_BATCH_SIZE_PER_GPU} log_prob_micro_batch_per_gpu=${LOG_PROB_MICRO_BATCH_SIZE_PER_GPU}" | tee -a "$LOG_PATH"
+
+CUDA_VISIBLE_DEVICES="${GPU_LIST}" python3 train.py --config-name "${CONFIG_NAME}" \
+    model_path="${MODEL_PATH}" \
+    trainer.project_name="ragen_gradient_analysis_probe" \
+    trainer.experiment_name="${EXP_NAME}" \
+    trainer.total_training_steps="${PROBE_TOTAL_STEPS}" \
+    trainer.save_freq=-1 \
+    trainer.default_local_dir="${OUTPUT_DIR}" \
+    trainer.logger="['console','wandb']" \
+    trainer.resume_mode=resume_path \
+    trainer.resume_from_path="${RESUME_FROM_PATH}" \
+    trainer.val_before_train="${VAL_BEFORE_TRAIN}" \
+    trainer.test_freq=0 \
+    trainer.n_gpus_per_node="${NUM_GPUS}" \
+    ray_kwargs.ray_init.num_cpus="${RAY_NUM_CPUS}" \
+    system.CUDA_VISIBLE_DEVICES="'${GPU_LIST}'" \
+    es_manager.train.env_groups="${ENV_GROUPS}" \
+    es_manager.train.group_size="${GROUP_SIZE}" \
+    es_manager.train.env_configs.n_groups="[${ENV_GROUPS}]" \
+    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu="${PPO_MICRO_BATCH_SIZE_PER_GPU}" \
+    actor_rollout_ref.actor.use_kl_loss=True \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.entropy_coeff=0.001 \
+    actor_rollout_ref.actor.entropy_from_logits_with_chunking=True \
+    actor_rollout_ref.actor.filter_loss_scaling=none \
+    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu="${LOG_PROB_MICRO_BATCH_SIZE_PER_GPU}" \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu="${LOG_PROB_MICRO_BATCH_SIZE_PER_GPU}" \
+    actor_rollout_ref.rollout.gpu_memory_utilization="${GPU_MEMORY_UTILIZATION}" \
+    actor_rollout_ref.rollout.rollout_filter_value="${FILTER_VALUE}" \
+    actor_rollout_ref.rollout.rollout_filter_strategy=top_p \
+    actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=linear \
+    actor_rollout_ref.rollout.rollout_filter_type=largest \
+    actor_rollout_ref.rollout.rollout_filter_metric=reward_variance \
+    actor_rollout_ref.rollout.rollout_filter_include_zero=False \
+    actor_rollout_ref.actor.checkpoint.save_contents=[model] \
+    critic.ppo_micro_batch_size_per_gpu="${PPO_MICRO_BATCH_SIZE_PER_GPU}" \
+    critic.checkpoint.save_contents=[model] \
+    trainer.gradient_analysis_mode=True \
+    trainer.gradient_analysis_every=1 \
+    trainer.gradient_analysis_env_groups="${ANALYSIS_ENV_GROUPS}" \
+    trainer.gradient_analysis_group_size="${ANALYSIS_GROUP_SIZE}" \
+    trainer.gradient_analysis_log_prefilter=True \
+    trainer.gradient_analysis_only=True \
+    trainer.exit_after_gradient_analysis=True \
+    actor_rollout_ref.rollout.gradient_analysis_num_buckets=6 \
+    actor_rollout_ref.rollout.gradient_analysis_bucket_mode=quantile \
+    "${ALGO_ARGS[@]}" \
+    2>&1 | tee -a "$LOG_PATH"

scripts/runs/run_top_p_sweep.sh

@@ -0,0 +1,446 @@
+#!/bin/bash
+# Top-p sweep for sokoban using GAE with the requested filter overrides.
+set -euo pipefail
+
+# Defaults
+STEPS=400
+MODEL_NAME="Qwen2.5-3B"
+MODEL_PATH="Qwen/${MODEL_NAME}"
+PROJECT_NAME="ragen_release_top_p_sweep"
+CONFIG_NAME="_2_sokoban"
+SAVE_FREQ=-1
+ROLL_FILTER_VALUES="1.0,0.98,0.95,0.9,0.8,0.6,0.4,nofilter"
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=0
+GPU_MEMORY_UTILIZATION=0.5
+RAY_NUM_CPUS=16
+
+usage() {
+    cat <<'EOF'
+Usage: $0 [options]
+Options:
+  --steps N                     Training steps (default: 400)
+  --rollout_filter_value LIST   Comma-separated top_p values (default: 1.0,0.98,0.95,0.9,0.8,0.6,0.4,nofilter)
+  --gpus LIST                   Comma-separated GPU IDs (auto-detected if omitted)
+  --gpus-per-exp N              GPUs per experiment (default: 1)
+  --ray-num-cpus N              Max CPUs per task for ray.init (default: 16)
+  --gpu-memory-utilization V    GPU memory utilization for rollouts (default: 0.5)
+  --save-freq N                 Checkpoint save frequency (default: -1)
+  -h, --help                    Show this help message
+EOF
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --rollout_filter_value) ROLL_FILTER_VALUES="$2"; shift 2 ;;
+        --rollout_filter_value=*) ROLL_FILTER_VALUES="${1#*=}"; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --ray-num-cpus) RAY_NUM_CPUS="$2"; shift 2 ;;
+        --ray-num-cpus=*) RAY_NUM_CPUS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+if ! [[ "$RAY_NUM_CPUS" =~ ^[0-9]+$ ]] || [ "$RAY_NUM_CPUS" -lt 1 ]; then
+    echo "Error: --ray-num-cpus must be a positive integer"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+NUM_SLOTS=${#GPU_GROUPS[@]}
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+declare -A GPU_LABELS
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_model_label() {
+    local models=()
+    local id label model
+    for id in "${GPUS[@]}"; do
+        label=$(get_gpu_label "$id")
+        model="${label#1x}"
+        models+=("$model")
+    done
+    local unique_models=()
+    local m found
+    for m in "${models[@]}"; do
+        found=false
+        for u in "${unique_models[@]}"; do
+            if [ "$u" = "$m" ]; then
+                found=true
+                break
+            fi
+        done
+        if [ "$found" = false ]; then
+            unique_models+=("$m")
+        fi
+    done
+    if [ ${#unique_models[@]} -eq 1 ]; then
+        echo "${unique_models[0]}"
+    else
+        echo "mixed"
+    fi
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+GPU_MODEL_LABEL=$(get_gpu_model_label)
+GPU_LOG_LABEL="${GPUS_PER_EXP}x${GPU_MODEL_LABEL}"
+LOG_FILE="logs/top_p_sweep_${MODEL_NAME}.log"
+RESULT_ROOT="logs/top_p_sweep_${MODEL_NAME}"
+CHECKPOINT_ROOT="model_saving/top_p_sweep_${MODEL_NAME}"
+
+mkdir -p logs
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== Top-p Sweep Runner (${MODEL_NAME}): $(date) ===" | tee "$LOG_FILE"
+echo "Values: ${ROLL_FILTER_VALUES} | Steps: ${STEPS} | GPUs per exp: ${GPU_LOG_LABEL}" | tee -a "$LOG_FILE"
+echo "Groups: ${GPU_GROUPS[*]} | GPU memory util: ${GPU_MEMORY_UTILIZATION} | ray_num_cpus: ${RAY_NUM_CPUS} | save_freq: ${SAVE_FREQ}" | tee -a "$LOG_FILE"
+
+parse_rollout_values() {
+    IFS=',' read -r -a raw <<< "$ROLL_FILTER_VALUES"
+    ROLL_VALUES=()
+    for token in "${raw[@]}"; do
+        token="${token// /}"
+        if [ -n "$token" ]; then
+            ROLL_VALUES+=("$token")
+        fi
+    done
+    if [ ${#ROLL_VALUES[@]} -eq 0 ]; then
+        echo "Error: no rollout_filter_value entries provided" >&2
+        exit 1
+    fi
+}
+
+parse_rollout_values
+EXPERIMENTS=("${ROLL_VALUES[@]}")
+
+run_experiment() {
+    local value="$1"
+    local gpu_list="$2"
+
+    local include_zero="False"
+    local filter_value="$value"
+    if [ "$value" = "nofilter" ]; then
+        include_zero="True"
+        filter_value="1.0"
+    fi
+    local safe_label="${value//./}"
+    safe_label="${safe_label,,}"
+
+    local name="sokoban_top_p_${safe_label}-${MODEL_NAME}"
+    local task_dir="${RESULT_ROOT}/${safe_label}"
+    local log_path="${task_dir}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${safe_label}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    mkdir -p "$task_dir"
+    mkdir -p "$checkpoint_dir"
+
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$CONFIG_NAME" \
+        model_path="${MODEL_PATH}" \
+        trainer.project_name="${PROJECT_NAME}" \
+        trainer.experiment_name="${name}" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        ray_kwargs.ray_init.num_cpus="${RAY_NUM_CPUS}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        algorithm.adv_estimator=gae \
+        actor_rollout_ref.actor.use_kl_loss=True \
+        actor_rollout_ref.actor.kl_loss_coef=0.001 \
+        actor_rollout_ref.actor.entropy_coeff=0.001 \
+        actor_rollout_ref.actor.entropy_from_logits_with_chunking=True \
+        actor_rollout_ref.actor.filter_loss_scaling=none \
+        actor_rollout_ref.actor.ppo_mini_batch_size=32 \
+        actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
+        actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=8 \
+        critic.ppo_mini_batch_size=32 \
+        critic.ppo_micro_batch_size_per_gpu=4 \
+        ppo_mini_batch_size=32 \
+        actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=8 \
+        actor_rollout_ref.rollout.rollout_filter_strategy=top_p \
+        actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=softmax \
+        actor_rollout_ref.rollout.rollout_filter_value=${filter_value} \
+        actor_rollout_ref.rollout.rollout_filter_include_zero=${include_zero} \
+        actor_rollout_ref.rollout.rollout_filter_type=largest \
+        actor_rollout_ref.rollout.gpu_memory_utilization="${GPU_MEMORY_UTILIZATION}" \
+        actor_rollout_ref.rollout.rollout_filter_metric=reward_variance \
+        es_manager.train.env_groups=8 \
+        es_manager.train.group_size=16 \
+        es_manager.train.env_configs.n_groups='[8]' \
+        es_manager.val.env_groups=512 \
+        es_manager.val.group_size=1 \
+        es_manager.val.env_configs.n_groups='[512]' \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+
+    END=$(date +%s)
+    TOTAL_TIME=$((END - START))
+
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\s]+([\d.]+)",
+    r"timing_s/eval_total[:\s]+([\d.]+)",
+    r"timing_s/total[:\s]+([\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="value=${value} | include_zero=${include_zero} | filter_value=${filter_value} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${task_dir}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+}
+
+EXPERIMENT_COUNT=${#EXPERIMENTS[@]}
+if [ $EXPERIMENT_COUNT -eq 0 ]; then
+    echo "No experiments to run" >&2
+    exit 1
+fi
+
+QUEUE_FILE=$(mktemp -t ragen_top_p_queue.XXXXXX)
+echo 0 > "$QUEUE_FILE"
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+USE_FLOCK=false
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$MAIN_PID" != "$$" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list="$1"
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        local value="${EXPERIMENTS[$idx]}"
+        run_experiment "$value" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== Top-p Sweep Summary ==="
+    echo "Project: ${PROJECT_NAME} | Steps: ${STEPS} | GPU per exp: ${GPU_LOG_LABEL}"
+    for val in "${EXPERIMENTS[@]}"; do
+        safe_label="${val//./p}"
+        safe_label="${safe_label,,}"
+        name="sokoban_top_p_${safe_label}-${MODEL_NAME}"
+        task_dir="${RESULT_ROOT}/${safe_label}"
+        if [ -f "${task_dir}/${name}.result" ]; then
+            cat "${task_dir}/${name}.result"
+        else
+            echo "value=${val} | status=missing"
+        fi
+    done
+} | tee -a "$LOG_FILE"

scripts/runs/run_webshop_release_combos.sh

@@ -0,0 +1,565 @@
+#!/bin/bash
+# Webshop (release): GRPO with 4 release filter modes
+# Model:
+#   - Qwen2.5-3B-Instruct
+# Filter modes:
+#   - topk25  => top_k, keep top 25% of groups
+#   - topp09  => top_p=0.9 with linear aggregation
+#   - topp095 => top_p=0.95 with linear aggregation
+#   - nofilter => top_p=1.0 with linear aggregation
+
+set -euo pipefail
+
+# Defaults
+STEPS=100
+TASK="webshop-release"
+CONFIG="_6_webshop"
+SAVE_FREQ=100
+NUM_GROUPS=16
+GROUP_SIZE=8
+MODEL_NAME="Qwen2.5-3B-Instruct"
+ALGO="GRPO"
+FILTER_MODES=("topk25" "topp09" "topp095" "nofilter")
+FILTERS_OPTION="all"
+SELECTED_FILTERS=("${FILTER_MODES[@]}")
+
+# GPU settings
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=30
+GPU_MEMORY_UTILIZATION=0.3
+declare -A GPU_LABELS
+
+usage() {
+    echo "Usage: $0 [options]"
+    echo "Options:"
+    echo "  --steps N             Training steps (default: 100)"
+    echo "  --gpus LIST           Comma-separated GPU IDs (default: auto-detect)"
+    echo "  --gpus-per-exp N      GPUs per experiment (default: 1)"
+    echo "  --cooldown SECONDS    Cooldown between runs on the same GPU group (default: 30)"
+    echo "  --gpu-memory-utilization V  Rollout gpu_memory_utilization (default: 0.3)"
+    echo "  --save-freq N         Checkpoint save frequency (default: 100)"
+    echo "  --filters LIST        Comma-separated filter modes (topk25,topp09,topp095,nofilter,all). Default: all"
+    echo "  -h, --help            Show this help"
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --cooldown) COOLDOWN_SECONDS="$2"; shift 2 ;;
+        --cooldown=*) COOLDOWN_SECONDS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        --filters) FILTERS_OPTION="$2"; shift 2 ;;
+        --filters=*) FILTERS_OPTION="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+# Map model names to HuggingFace paths
+get_model_path() {
+    if [[ "$1" == *"/"* ]]; then
+        echo "$1"
+        return
+    fi
+    case "$1" in
+        Qwen2.5-3B-Instruct) echo "Qwen/Qwen2.5-3B-Instruct" ;;
+        Qwen2.5-7B-Instruct) echo "Qwen/Qwen2.5-7B-Instruct" ;;
+        Llama-3.2-3B-Instruct) echo "meta-llama/Llama-3.2-3B-Instruct" ;;
+        *) echo "Qwen/$1" ;;
+    esac
+}
+
+get_algo_overrides() {
+    case "$1" in
+        PPO)
+            echo "algorithm.adv_estimator=gae actor_rollout_ref.actor.loss_agg_mode=token-mean"
+            ;;
+        GRPO)
+            echo "algorithm.adv_estimator=grpo algorithm.norm_adv_by_std_in_grpo=True actor_rollout_ref.actor.loss_agg_mode=seq-mean-token-mean"
+            ;;
+        *)
+            echo ""
+            ;;
+    esac
+}
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+NUM_SLOTS=${#GPU_GROUPS[@]}
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_model_label() {
+    local models=()
+    local id label model
+    for id in "${GPUS[@]}"; do
+        label=$(get_gpu_label "$id")
+        model="${label#1x}"
+        models+=("$model")
+    done
+    local unique_models=()
+    local m found
+    for m in "${models[@]}"; do
+        found=false
+        for u in "${unique_models[@]}"; do
+            if [ "$u" = "$m" ]; then
+                found=true
+                break
+            fi
+        done
+        if [ "$found" = false ]; then
+            unique_models+=("$m")
+        fi
+    done
+    if [ ${#unique_models[@]} -eq 1 ]; then
+        echo "${unique_models[0]}"
+    else
+        echo "mixed"
+    fi
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+GPU_MODEL_LABEL=$(get_gpu_model_label)
+GPU_LOG_LABEL="${GPUS_PER_EXP}x${GPU_MODEL_LABEL}"
+LOG_FILE="logs/webshop_release_combos.log"
+RESULT_ROOT="logs"
+CHECKPOINT_ROOT="model_saving/webshop_release_combos"
+
+mkdir -p logs
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== Webshop Release Combos Runner: $(date) ===" | tee "$LOG_FILE"
+echo "Task: ${TASK} | Steps: ${STEPS} | GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL}" | tee -a "$LOG_FILE"
+echo "GPUS: ${GPUS[*]} | groups: ${GPU_GROUPS[*]} | cooldown=${COOLDOWN_SECONDS}s" | tee -a "$LOG_FILE"
+
+get_filter_config() {
+    case "$1" in
+        topk25)
+            echo "top_k|0.25|linear|topk25"
+            ;;
+        topp09)
+            echo "top_p|0.9|linear|topp09-linear"
+            ;;
+        topp095)
+            echo "top_p|0.95|linear|topp095-linear"
+            ;;
+        nofilter)
+            echo "top_p|1.0|linear|nofilter"
+            ;;
+        *)
+            echo "Unknown filter mode: $1" >&2
+            return 1
+            ;;
+    esac
+}
+
+build_experiment_name() {
+    local task=$1
+    local algo=$2
+    local filter_suffix=$3
+    local model_name=$4
+    echo "${task}-${algo}-${filter_suffix}-${model_name}-${NUM_GROUPS}x${GROUP_SIZE}"
+}
+
+run_experiment() {
+    local task=$1
+    local model_name=$2
+    local algo=$3
+    local filter=$4
+    local config=$5
+    local gpu_list=$6
+
+    local model_path
+    model_path=$(get_model_path "$model_name")
+
+    local filter_strategy
+    local filter_value
+    local filter_prob_mode
+    local filter_suffix
+    IFS='|' read -r filter_strategy filter_value filter_prob_mode filter_suffix <<< "$(get_filter_config "$filter")"
+    local include_zero=True
+
+    local common_overrides=(
+        "actor_rollout_ref.actor.use_kl_loss=False"
+        "actor_rollout_ref.actor.kl_loss_type=low-var-kl"
+        "actor_rollout_ref.actor.kl_loss_coef=0.001"
+        "actor_rollout_ref.actor.entropy_coeff=0.001"
+        "actor_rollout_ref.actor.entropy_from_logits_with_chunking=True"
+        "actor_rollout_ref.actor.filter_loss_scaling=none"
+        "actor_rollout_ref.rollout.gpu_memory_utilization=${GPU_MEMORY_UTILIZATION}"
+        "actor_rollout_ref.rollout.rollout_filter_strategy=${filter_strategy}"
+        "actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=${filter_prob_mode}"
+        "actor_rollout_ref.rollout.rollout_filter_type=largest"
+        "actor_rollout_ref.rollout.rollout_filter_metric=reward_variance"
+        "actor_rollout_ref.rollout.rollout_filter_include_zero=${include_zero}"
+    )
+
+    local env_overrides=()
+
+    local checkpoint_overrides=(
+        "actor_rollout_ref.actor.checkpoint.save_contents=[model]"
+        "critic.checkpoint.save_contents=[model]"
+    )
+
+    local algo_overrides
+    algo_overrides=$(get_algo_overrides "$algo")
+    read -r -a algo_args <<< "$algo_overrides"
+
+    local name
+    name=$(build_experiment_name "$task" "$algo" "$filter_suffix" "$model_name")
+    local task_dir="${RESULT_ROOT}/webshop_release_combos"
+    local log_path="${task_dir}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${model_name}/${algo}/${filter}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    mkdir -p "$task_dir"
+    mkdir -p "${checkpoint_dir}"
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$config" \
+        model_path="${model_path}" \
+        micro_batch_size_per_gpu=1 \
+        log_prob_micro_batch_size_per_gpu=1 \
+        agent_proxy.max_turn=9 \
+        trainer.project_name="main_webshop" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.experiment_name="${name}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        actor_rollout_ref.rollout.rollout_filter_value="${filter_value}" \
+        actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=1 \
+        actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=1 \
+        ppo_mini_batch_size=32 \
+        es_manager.train.env_groups=${NUM_GROUPS} \
+        es_manager.train.group_size=${GROUP_SIZE} \
+        es_manager.train.env_configs.n_groups="[${NUM_GROUPS}]" \
+        es_manager.val.env_groups=256 \
+        es_manager.val.group_size=1 \
+        es_manager.val.env_configs.n_groups="[256]" \
+        "${common_overrides[@]}" \
+        "${env_overrides[@]}" \
+        "${checkpoint_overrides[@]}" \
+        "${algo_args[@]}" \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+    END=$(date +%s)
+
+    TOTAL_TIME=$((END - START))
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\s]+([\d.]+)",
+    r"timing_s/eval_total[:\s]+([\d.]+)",
+    r"timing_s/total[:\s]+([\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="task=${task} | algo=${algo} | filter=${filter} | model=${model_name} | steps=${STEPS} | filter=${filter_strategy}:${filter_value} | prob_mode=${filter_prob_mode} | include_zero=${include_zero} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${task_dir}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+    return 0
+}
+
+EXPERIMENTS=()
+GROUP_LABELS=()
+CURRENT_GROUP=""
+
+resolve_filter_selection() {
+    local raw="$1"
+    if [ -z "$raw" ] || [ "$raw" = "all" ]; then
+        SELECTED_FILTERS=("${FILTER_MODES[@]}")
+        return
+    fi
+    IFS=',' read -r -a candidates <<< "$raw"
+    SELECTED_FILTERS=()
+    for candidate in "${candidates[@]}"; do
+        candidate="${candidate// /}"
+        case "$candidate" in
+            topk25|topp09|topp095|nofilter)
+                SELECTED_FILTERS+=("$candidate")
+                ;;
+            "")
+                continue
+                ;;
+            *)
+                echo "Unknown filter mode: $candidate" >&2
+                exit 1
+                ;;
+        esac
+    done
+    if [ ${#SELECTED_FILTERS[@]} -eq 0 ]; then
+        echo "No valid filters selected via --filters" >&2
+        exit 1
+    fi
+}
+
+set_group() {
+    CURRENT_GROUP="$1"
+    GROUP_LABELS+=("$1")
+}
+
+add_experiment() {
+    local model_name=$1
+    local algo=$2
+    local filter=$3
+    EXPERIMENTS+=("${CURRENT_GROUP}|${TASK}|${model_name}|${algo}|${filter}|${CONFIG}")
+}
+
+resolve_filter_selection "$FILTERS_OPTION"
+
+set_group "${MODEL_NAME} + ${ALGO}"
+for filter in "${SELECTED_FILTERS[@]}"; do
+    add_experiment "$MODEL_NAME" "$ALGO" "$filter"
+done
+
+QUEUE_FILE=$(mktemp -t ragen_webshop_small_queue.XXXXXX)
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+echo 0 > "$QUEUE_FILE"
+USE_FLOCK=false
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$$" -ne "$MAIN_PID" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list=$1
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        local exp="${EXPERIMENTS[$idx]}"
+        IFS='|' read -r exp_group task model_name algo filter config <<< "$exp"
+        run_experiment "$task" "$model_name" "$algo" "$filter" "$config" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== Grouped Summary ==="
+    echo "GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL} | Task: ${TASK} | Steps: ${STEPS}"
+    for group_label in "${GROUP_LABELS[@]}"; do
+        echo "=== ${group_label} ==="
+        for exp in "${EXPERIMENTS[@]}"; do
+            IFS='|' read -r exp_group task model_name algo filter config <<< "$exp"
+            if [ "$exp_group" != "$group_label" ]; then
+                continue
+            fi
+            IFS='|' read -r _filter_strategy _filter_value _filter_prob_mode filter_suffix <<< "$(get_filter_config "$filter")"
+            name=$(build_experiment_name "$task" "$algo" "$filter_suffix" "$model_name")
+            task_dir="${RESULT_ROOT}/webshop_release_combos"
+            if [ -f "${task_dir}/${name}.result" ]; then
+                cat "${task_dir}/${name}.result"
+            else
+                echo "task=${task} | algo=${algo} | filter=${filter} | model=${model_name} | status=missing"
+            fi
+        done
+    done
+} | tee -a "$LOG_FILE"

scripts/runs/run_webshop_small_combos.sh

@@ -0,0 +1,563 @@
+#!/bin/bash
+# Webshop (small): 4 model×algo combos with filter/no-filter
+#   1) Qwen2.5-3B-Instruct  + PPO
+#   2) Qwen2.5-3B-Instruct  + GRPO
+#   3) Qwen2.5-7B-Instruct  + PPO
+#   4) Llama-3.2-3B-Instruct + PPO
+# Filtering rule: filter => top_p=0.9, nofilter => top_p=1.0
+
+set -euo pipefail
+
+# Defaults
+STEPS=100
+TASK="webshop"
+CONFIG="_6_webshop"
+SAVE_FREQ=100
+NUM_GROUPS=8
+GROUP_SIZE=16
+COMBOS_SELECTION=""
+FILTER_MODES=("filter" "nofilter")
+FILTERS_OPTION="all"
+SELECTED_FILTERS=("${FILTER_MODES[@]}")
+
+# GPU settings
+GPUS=()
+GPUS_PROVIDED=false
+GPUS_PER_EXP=1
+COOLDOWN_SECONDS=30
+GPU_MEMORY_UTILIZATION=0.3
+declare -A GPU_LABELS
+
+usage() {
+    echo "Usage: $0 [options]"
+    echo "Options:"
+    echo "  --steps N             Training steps (default: 100)"
+    echo "  --gpus LIST           Comma-separated GPU IDs (default: auto-detect)"
+    echo "  --gpus-per-exp N      GPUs per experiment (default: 1)"
+    echo "  --cooldown SECONDS    Cooldown between runs on the same GPU group (default: 30)"
+    echo "  --gpu-memory-utilization V  Rollout gpu_memory_utilization (default: 0.3)"
+    echo "  --save-freq N         Checkpoint save frequency (default: 100)"
+    echo "  --filters LIST        Comma-separated filter modes (filter,nofilter,all). Default: all"
+    echo "  --combos LIST         Comma-separated 1-based combo indices to run (e.g., 2,3,4). Default: all"
+    echo "  -h, --help            Show this help"
+    exit 0
+}
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --steps) STEPS="$2"; shift 2 ;;
+        --steps=*) STEPS="${1#*=}"; shift ;;
+        --gpus) IFS=',' read -r -a GPUS <<< "$2"; GPUS_PROVIDED=true; shift 2 ;;
+        --gpus=*) IFS=',' read -r -a GPUS <<< "${1#*=}"; GPUS_PROVIDED=true; shift ;;
+        --gpus-per-exp) GPUS_PER_EXP="$2"; shift 2 ;;
+        --gpus-per-exp=*) GPUS_PER_EXP="${1#*=}"; shift ;;
+        --cooldown) COOLDOWN_SECONDS="$2"; shift 2 ;;
+        --cooldown=*) COOLDOWN_SECONDS="${1#*=}"; shift ;;
+        --gpu-memory-utilization) GPU_MEMORY_UTILIZATION="$2"; shift 2 ;;
+        --gpu-memory-utilization=*) GPU_MEMORY_UTILIZATION="${1#*=}"; shift ;;
+        --save-freq) SAVE_FREQ="$2"; shift 2 ;;
+        --save-freq=*) SAVE_FREQ="${1#*=}"; shift ;;
+        --filters) FILTERS_OPTION="$2"; shift 2 ;;
+        --filters=*) FILTERS_OPTION="${1#*=}"; shift ;;
+        --combos) COMBOS_SELECTION="$2"; shift 2 ;;
+        --combos=*) COMBOS_SELECTION="${1#*=}"; shift ;;
+        -h|--help) usage ;;
+        *) echo "Unknown argument: $1"; usage ;;
+    esac
+done
+
+# Map model names to HuggingFace paths
+get_model_path() {
+    if [[ "$1" == *"/"* ]]; then
+        echo "$1"
+        return
+    fi
+    case "$1" in
+        Qwen2.5-3B-Instruct) echo "Qwen/Qwen2.5-3B-Instruct" ;;
+        Qwen2.5-7B-Instruct) echo "Qwen/Qwen2.5-7B-Instruct" ;;
+        Llama-3.2-3B-Instruct) echo "meta-llama/Llama-3.2-3B-Instruct" ;;
+        *) echo "Qwen/$1" ;;
+    esac
+}
+
+# Algorithm overrides (from diff_algo script)
+get_algo_overrides() {
+    case "$1" in
+        PPO)
+            echo "algorithm.adv_estimator=gae actor_rollout_ref.actor.loss_agg_mode=token-mean"
+            ;;
+        GRPO)
+            echo "algorithm.adv_estimator=grpo algorithm.norm_adv_by_std_in_grpo=True actor_rollout_ref.actor.loss_agg_mode=seq-mean-token-mean"
+            ;;
+        *)
+            echo ""
+            ;;
+    esac
+}
+
+if [ "$GPUS_PROVIDED" = false ]; then
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        GPU_COUNT=$(nvidia-smi -L 2>/dev/null | wc -l | tr -d ' ')
+        if [[ "$GPU_COUNT" =~ ^[0-9]+$ ]] && [ "$GPU_COUNT" -gt 0 ]; then
+            GPUS=()
+            for ((i=0; i<GPU_COUNT; i++)); do
+                GPUS+=("$i")
+            done
+        fi
+    fi
+    if [ ${#GPUS[@]} -eq 0 ]; then
+        echo "Warning: failed to auto-detect GPUs, falling back to 0-7" >&2
+        GPUS=(0 1 2 3 4 5 6 7)
+    fi
+fi
+
+if ! [[ "$GPUS_PER_EXP" =~ ^[0-9]+$ ]] || [ "$GPUS_PER_EXP" -lt 1 ]; then
+    echo "Error: --gpus-per-exp must be a positive integer"
+    exit 1
+fi
+if (( ${#GPUS[@]} < GPUS_PER_EXP )); then
+    echo "Error: --gpus-per-exp (${GPUS_PER_EXP}) exceeds available GPUs (${#GPUS[@]})"
+    exit 1
+fi
+if (( ${#GPUS[@]} % GPUS_PER_EXP != 0 )); then
+    echo "Error: GPU count (${#GPUS[@]}) must be divisible by --gpus-per-exp (${GPUS_PER_EXP})"
+    exit 1
+fi
+
+GPU_GROUPS=()
+for ((i=0; i<${#GPUS[@]}; i+=GPUS_PER_EXP)); do
+    group="${GPUS[$i]}"
+    for ((j=1; j<GPUS_PER_EXP; j++)); do
+        group+=",${GPUS[$((i+j))]}"
+    done
+    GPU_GROUPS+=("$group")
+done
+NUM_SLOTS=${#GPU_GROUPS[@]}
+
+short_gpu_name() {
+    local name="$1"
+    local cleaned
+    cleaned=$(echo "$name" | sed -E 's/^NVIDIA //; s/^Tesla //; s/^GeForce //; s/^Quadro //; s/^RTX //')
+    if [[ "$cleaned" =~ (B[0-9]{2,3}|H[0-9]{2,3}|A[0-9]{2,3}|L[0-9]{2,3}|V100|T4|P100|K80) ]]; then
+        echo "${BASH_REMATCH[1]}"
+        return
+    fi
+    echo "${cleaned%% *}"
+}
+
+get_gpu_label() {
+    local gpu_id="$1"
+    if [ -n "${GPU_LABELS[$gpu_id]+x}" ]; then
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local name=""
+    if command -v nvidia-smi >/dev/null 2>&1; then
+        name=$(nvidia-smi --query-gpu=name --format=csv,noheader -i "$gpu_id" 2>/dev/null | head -1)
+    fi
+    if [ -z "$name" ]; then
+        GPU_LABELS[$gpu_id]="1xGPU${gpu_id}"
+        echo "${GPU_LABELS[$gpu_id]}"
+        return
+    fi
+    local short
+    short=$(short_gpu_name "$name")
+    GPU_LABELS[$gpu_id]="1x${short}"
+    echo "${GPU_LABELS[$gpu_id]}"
+}
+
+get_gpu_model_label() {
+    local models=()
+    local id label model
+    for id in "${GPUS[@]}"; do
+        label=$(get_gpu_label "$id")
+        model="${label#1x}"
+        models+=("$model")
+    done
+    local unique_models=()
+    local m found
+    for m in "${models[@]}"; do
+        found=false
+        for u in "${unique_models[@]}"; do
+            if [ "$u" = "$m" ]; then
+                found=true
+                break
+            fi
+        done
+        if [ "$found" = false ]; then
+            unique_models+=("$m")
+        fi
+    done
+    if [ ${#unique_models[@]} -eq 1 ]; then
+        echo "${unique_models[0]}"
+    else
+        echo "mixed"
+    fi
+}
+
+get_gpu_label_for_list() {
+    local gpu_list="$1"
+    IFS=',' read -r -a ids <<< "$gpu_list"
+    local count=${#ids[@]}
+    if [ "$count" -eq 0 ]; then
+        echo "0xGPU"
+        return
+    fi
+    local first_model
+    first_model="$(get_gpu_label "${ids[0]}")"
+    first_model="${first_model#1x}"
+    local id model
+    for id in "${ids[@]:1}"; do
+        model="$(get_gpu_label "$id")"
+        model="${model#1x}"
+        if [ "$model" != "$first_model" ]; then
+            echo "${count}xmixed"
+            return
+        fi
+    done
+    echo "${count}x${first_model}"
+}
+
+GPU_MODEL_LABEL=$(get_gpu_model_label)
+GPU_LOG_LABEL="${GPUS_PER_EXP}x${GPU_MODEL_LABEL}"
+LOG_FILE="logs/webshop_small_combos.log"
+RESULT_ROOT="logs"
+CHECKPOINT_ROOT="model_saving/webshop_small_combos"
+
+mkdir -p logs
+mkdir -p "$RESULT_ROOT"
+mkdir -p "$CHECKPOINT_ROOT"
+
+echo "=== Webshop Small Combos Runner: $(date) ===" | tee "$LOG_FILE"
+echo "Task: ${TASK} | Steps: ${STEPS} | GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL}" | tee -a "$LOG_FILE"
+echo "GPUS: ${GPUS[*]} | groups: ${GPU_GROUPS[*]} | cooldown=${COOLDOWN_SECONDS}s" | tee -a "$LOG_FILE"
+
+run_experiment() {
+    local task=$1
+    local model_name=$2
+    local algo=$3
+    local filter=$4
+    local config=$5
+    local gpu_list=$6
+
+    local model_path
+    model_path=$(get_model_path "$model_name")
+
+    local filter_value
+    local include_zero
+    if [ "$filter" = "filter" ]; then
+        filter_value=0.9
+        include_zero=False
+    else
+        filter_value=1.0
+        include_zero=True
+    fi
+    local filter_strategy="top_p"
+
+    local common_overrides=(
+        "actor_rollout_ref.actor.use_kl_loss=False"
+        "actor_rollout_ref.actor.kl_loss_type=low-var-kl"
+        "actor_rollout_ref.actor.kl_loss_coef=0.001"
+        "actor_rollout_ref.actor.entropy_coeff=0.001"
+        "actor_rollout_ref.actor.entropy_from_logits_with_chunking=True"
+        "actor_rollout_ref.actor.filter_loss_scaling=none"
+        "actor_rollout_ref.rollout.gpu_memory_utilization=${GPU_MEMORY_UTILIZATION}"
+        "actor_rollout_ref.rollout.rollout_filter_strategy=${filter_strategy}"
+        "actor_rollout_ref.rollout.rollout_filter_top_p_prob_mode=linear"
+        "actor_rollout_ref.rollout.rollout_filter_type=largest"
+        "actor_rollout_ref.rollout.rollout_filter_metric=reward_variance"
+        "actor_rollout_ref.rollout.rollout_filter_include_zero=${include_zero}"
+    )
+
+    local env_overrides=()
+
+    local checkpoint_overrides=(
+        "actor_rollout_ref.actor.checkpoint.save_contents=[model]"
+        "critic.checkpoint.save_contents=[model]"
+    )
+
+    local algo_overrides
+    algo_overrides=$(get_algo_overrides "$algo")
+    read -r -a algo_args <<< "$algo_overrides"
+
+    local name="${task}-${algo}-${filter}-${model_name}-${NUM_GROUPS}x${GROUP_SIZE}-linear"
+    local task_dir="${RESULT_ROOT}/webshop_small_combos"
+    local log_path="${task_dir}/${name}.log"
+    local checkpoint_dir="${CHECKPOINT_ROOT}/${model_name}/${algo}/${filter}/${name}"
+    local gpus_per_exp
+    IFS=',' read -r -a gpu_ids <<< "$gpu_list"
+    gpus_per_exp=${#gpu_ids[@]}
+
+    mkdir -p "$task_dir"
+    mkdir -p "${checkpoint_dir}"
+    START=$(date +%s)
+    CUDA_VISIBLE_DEVICES="${gpu_list}" python train.py --config-name "$config" \
+        model_path="${model_path}" \
+        micro_batch_size_per_gpu=1 \
+        agent_proxy.max_turn=12 \
+        trainer.project_name="main_webshop" \
+        trainer.total_training_steps="${STEPS}" \
+        trainer.experiment_name="${name}" \
+        trainer.save_freq="${SAVE_FREQ}" \
+        trainer.default_local_dir="${checkpoint_dir}" \
+        trainer.logger="['console','wandb']" \
+        trainer.val_before_train=True \
+        trainer.n_gpus_per_node="${gpus_per_exp}" \
+        system.CUDA_VISIBLE_DEVICES="'${gpu_list}'" \
+        actor_rollout_ref.rollout.rollout_filter_value="${filter_value}" \
+        ppo_mini_batch_size=16 \
+        es_manager.train.env_groups=${NUM_GROUPS} \
+        es_manager.train.group_size=${GROUP_SIZE} \
+        es_manager.train.env_configs.n_groups="[${NUM_GROUPS}]" \
+        es_manager.val.env_groups=128 \
+        es_manager.val.group_size=1 \
+        es_manager.val.env_configs.n_groups="[128]" \
+        "${common_overrides[@]}" \
+        "${env_overrides[@]}" \
+        "${checkpoint_overrides[@]}" \
+        "${algo_args[@]}" \
+        2>&1 | tee "$log_path"
+    EXIT_CODE=${PIPESTATUS[0]}
+    END=$(date +%s)
+
+    TOTAL_TIME=$((END - START))
+    timing_values=()
+    mapfile -t timing_values < <(
+        python - "$log_path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+def last(pattern, text):
+    matches = re.findall(pattern, text)
+    return matches[-1] if matches else ""
+
+try:
+    text = Path(sys.argv[1]).read_text(errors="ignore")
+except Exception:
+    text = ""
+
+patterns = [
+    r"timing_s/train_total[:\s]+([\d.]+)",
+    r"timing_s/eval_total[:\s]+([\d.]+)",
+    r"timing_s/total[:\s]+([\d.]+)",
+]
+
+for pattern in patterns:
+    print(last(pattern, text))
+PY
+    )
+    TRAIN_TIME_RAW="${timing_values[0]:-}"
+    EVAL_TIME_RAW="${timing_values[1]:-}"
+    TOTAL_TIME_RAW="${timing_values[2]:-}"
+    TRAIN_TIME=$([ -n "$TRAIN_TIME_RAW" ] && printf "%.2f" "$TRAIN_TIME_RAW" || echo "N/A")
+    EVAL_TIME=$([ -n "$EVAL_TIME_RAW" ] && printf "%.2f" "$EVAL_TIME_RAW" || echo "N/A")
+    TOTAL_TIME_METRIC=$([ -n "$TOTAL_TIME_RAW" ] && printf "%.2f" "$TOTAL_TIME_RAW" || echo "N/A")
+
+    local status="success"
+    local error_line=""
+    if [ $EXIT_CODE -ne 0 ]; then
+        status="fail"
+        error_line=$(tail -2 "$log_path" | tr '\n' ' ')
+    fi
+
+    local gpu_label
+    gpu_label=$(get_gpu_label_for_list "$gpu_list")
+    local summary_line="task=${task} | algo=${algo} | filter=${filter} | model=${model_name} | steps=${STEPS} | filter=${filter_strategy}:${filter_value} | train_time=${TRAIN_TIME}s | eval_time=${EVAL_TIME}s | total_time=${TOTAL_TIME_METRIC}s | wall_time=${TOTAL_TIME}s | gpu=${gpu_label} | status=${status}"
+    echo "${summary_line}" > "${task_dir}/${name}.result"
+    echo "${summary_line}" | tee -a "$LOG_FILE"
+    if [ "$status" = "fail" ]; then
+        echo "  error: ${error_line}" | tee -a "$LOG_FILE"
+    fi
+    return 0
+}
+
+EXPERIMENTS=()
+GROUP_LABELS=()
+CURRENT_GROUP=""
+
+resolve_filter_selection() {
+    local raw="$1"
+    if [ -z "$raw" ] || [ "$raw" = "all" ]; then
+        SELECTED_FILTERS=("${FILTER_MODES[@]}")
+        return
+    fi
+    IFS=',' read -r -a candidates <<< "$raw"
+    SELECTED_FILTERS=()
+    for candidate in "${candidates[@]}"; do
+        candidate="${candidate// /}"
+        case "$candidate" in
+            filter|nofilter)
+                SELECTED_FILTERS+=("$candidate")
+                ;;
+            "")
+                continue
+                ;;
+            *)
+                echo "Unknown filter mode: $candidate" >&2
+                exit 1
+                ;;
+        esac
+    done
+    if [ ${#SELECTED_FILTERS[@]} -eq 0 ]; then
+        echo "No valid filters selected via --filters" >&2
+        exit 1
+    fi
+}
+
+set_group() {
+    CURRENT_GROUP="$1"
+    GROUP_LABELS+=("$1")
+}
+
+add_experiment() {
+    local model_name=$1
+    local algo=$2
+    local filter=$3
+    EXPERIMENTS+=("${CURRENT_GROUP}|${TASK}|${model_name}|${algo}|${filter}|${CONFIG}")
+}
+
+resolve_filter_selection "$FILTERS_OPTION"
+
+# Define the 4 combos: (model, algo)
+ALL_COMBOS=(
+    "Qwen2.5-3B-Instruct|PPO"
+    "Qwen2.5-3B-Instruct|GRPO"
+    "Qwen2.5-7B-Instruct|PPO"
+    "Llama-3.2-3B-Instruct|PPO"
+)
+
+if [ -n "$COMBOS_SELECTION" ]; then
+    COMBOS=()
+    IFS=',' read -r -a combo_indices <<< "$COMBOS_SELECTION"
+    for ci in "${combo_indices[@]}"; do
+        idx=$((ci - 1))
+        if [ "$idx" -ge 0 ] && [ "$idx" -lt ${#ALL_COMBOS[@]} ]; then
+            COMBOS+=("${ALL_COMBOS[$idx]}")
+        else
+            echo "Invalid combo index: $ci (valid: 1-${#ALL_COMBOS[@]})" >&2
+            exit 1
+        fi
+    done
+else
+    COMBOS=("${ALL_COMBOS[@]}")
+fi
+
+for combo in "${COMBOS[@]}"; do
+    IFS='|' read -r model_name algo <<< "$combo"
+    set_group "${model_name} + ${algo}"
+    for filter in "${SELECTED_FILTERS[@]}"; do
+        add_experiment "$model_name" "$algo" "$filter"
+    done
+done
+
+QUEUE_FILE=$(mktemp -t ragen_webshop_small_queue.XXXXXX)
+QUEUE_LOCK="${QUEUE_FILE}.lock"
+echo 0 > "$QUEUE_FILE"
+USE_FLOCK=false
+QUEUE_LOCK_DIR="${QUEUE_LOCK}.d"
+MAIN_PID=$$
+
+cleanup_queue() {
+    if [ "$$" -ne "$MAIN_PID" ]; then
+        return
+    fi
+    rm -f "$QUEUE_FILE" "$QUEUE_LOCK"
+    rmdir "$QUEUE_LOCK_DIR" 2>/dev/null || true
+}
+trap cleanup_queue EXIT
+
+if command -v flock >/dev/null 2>&1; then
+    USE_FLOCK=true
+fi
+
+next_experiment_index() {
+    local idx
+    if [ "$USE_FLOCK" = true ]; then
+        flock -x "$QUEUE_LOCK_FD"
+        idx=$(cat "$QUEUE_FILE")
+        if [ -z "$idx" ]; then
+            idx=0
+        fi
+        if (( idx >= ${#EXPERIMENTS[@]} )); then
+            flock -u "$QUEUE_LOCK_FD"
+            echo -1
+            return
+        fi
+        echo $((idx + 1)) > "$QUEUE_FILE"
+        flock -u "$QUEUE_LOCK_FD"
+        echo "$idx"
+        return
+    fi
+
+    while ! mkdir "$QUEUE_LOCK_DIR" 2>/dev/null; do
+        sleep 0.05
+    done
+    idx=$(cat "$QUEUE_FILE")
+    if [ -z "$idx" ]; then
+        idx=0
+    fi
+    if (( idx >= ${#EXPERIMENTS[@]} )); then
+        rmdir "$QUEUE_LOCK_DIR"
+        echo -1
+        return
+    fi
+    echo $((idx + 1)) > "$QUEUE_FILE"
+    rmdir "$QUEUE_LOCK_DIR"
+    echo "$idx"
+}
+
+run_queue_for_slot() {
+    local gpu_list=$1
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>"$QUEUE_LOCK"
+    fi
+    while true; do
+        local idx
+        idx=$(next_experiment_index)
+        if [ "$idx" -lt 0 ]; then
+            break
+        fi
+        local exp="${EXPERIMENTS[$idx]}"
+        IFS='|' read -r exp_group task model_name algo filter config <<< "$exp"
+        run_experiment "$task" "$model_name" "$algo" "$filter" "$config" "$gpu_list" || true
+        if [ "$COOLDOWN_SECONDS" -gt 0 ]; then
+            sleep "$COOLDOWN_SECONDS"
+        fi
+    done
+    if [ "$USE_FLOCK" = true ]; then
+        exec {QUEUE_LOCK_FD}>&-
+    fi
+}
+
+pids=()
+for idx in "${!GPU_GROUPS[@]}"; do
+    run_queue_for_slot "${GPU_GROUPS[$idx]}" &
+    pids+=("$!")
+done
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done
+
+{
+    echo ""
+    echo "=== Grouped Summary ==="
+    echo "GPU per exp: ${GPUS_PER_EXP}x${GPU_MODEL_LABEL} | Task: ${TASK} | Steps: ${STEPS}"
+    for group_label in "${GROUP_LABELS[@]}"; do
+        echo "=== ${group_label} ==="
+        for exp in "${EXPERIMENTS[@]}"; do
+            IFS='|' read -r exp_group task model_name algo filter config <<< "$exp"
+            if [ "$exp_group" != "$group_label" ]; then
+                continue
+            fi
+            name="${task}-${algo}-${filter}-${model_name}-${NUM_GROUPS}x${GROUP_SIZE}-linear"
+            task_dir="${RESULT_ROOT}/webshop_small_combos"
+            if [ -f "${task_dir}/${name}.result" ]; then
+                cat "${task_dir}/${name}.result"
+            else
+                echo "task=${task} | algo=${algo} | filter=${filter} | model=${model_name} | status=missing"
+            fi
+        done
+    done
+} | tee -a "$LOG_FILE"

scripts/setup_ragen.md

@@ -0,0 +1,47 @@
+# Scripts README
+
+## Release Environment Setup
+
+Use the setup flow below for the validated RAGEN environment.
+
+This setup has been validated on `H100`, `H200`, and `B200`, and supports:
+- `bandit`
+- `sokoban`
+- `frozenlake`
+- `metamathqa`
+- `countdown`
+- `deepcoder`
+
+## Requirements
+
+- `CUDA >= 12.8`
+
+### 1. Clone the repository
+
+```bash
+git clone https://github.com/CHIGUI0/RAGEN.git
+cd RAGEN
+```
+
+### 2. Create and activate the conda environment
+
+```bash
+conda create -n ragen python=3.12 -y
+conda activate ragen
+```
+
+### 3. Run the environment setup script
+
+```bash
+bash scripts/setup_ragen.sh
+```
+
+If you want to install the `search` environment, use the following command:
+
+```bash
+bash scripts/setup_ragen.sh --with-search
+```
+
+This release setup does not install `webshop`. If you need `webshop`, use its separate setup flow instead of `setup_ragen.sh`.
+
+If you want to run WebShop experiments, see [docs/experiment_webshop_release.md](../docs/experiment_webshop_release.md).

scripts/visualize.py

@@ -0,0 +1,692 @@
+#!/usr/bin/env python3
+"""
+Local rollout visualizer.
+
+Usage:
+    python scripts/visualize.py --rollout_path results/ [--host 127.0.0.1] [--port 8000]
+
+The script launches a small HTTP server that lets you inspect .pkl files
+(containing verl.DataProto dumps) inside the rollout path. Open the printed
+URL in a browser to explore directories, select a file, and view its
+meta information and non-tensor batches entry by entry.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import logging
+import threading
+from functools import lru_cache
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from pathlib import Path
+from typing import Any, Dict, List
+from urllib.parse import parse_qs, urlparse
+import webbrowser
+
+import numpy as np
+
+from verl import DataProto
+
+LOGGER = logging.getLogger(__name__)
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Launch a local rollout visualizer")
+    parser.add_argument("--rollout_path", required=True, help="Directory containing rollout .pkl files")
+    parser.add_argument("--host", default="127.0.0.1", help="Host to bind (default: 127.0.0.1)")
+    parser.add_argument("--port", type=int, default=8000, help="Port to bind (default: 8000)")
+    parser.add_argument("--no-browser", action="store_true", help="Do not attempt to open a browser automatically")
+    return parser.parse_args()
+
+
+def ensure_within(path: Path, root: Path) -> Path:
+    resolved = path.resolve()
+    try:
+        resolved.relative_to(root)
+    except ValueError as exc:
+        raise ValueError(f"Path {path} escapes the rollout root {root}") from exc
+    return resolved
+
+
+def numpy_summary(array: np.ndarray) -> Dict[str, Any]:
+    array = np.asarray(array)
+    summary: Dict[str, Any] = {
+        "__type__": "ndarray",
+        "dtype": str(array.dtype),
+        "shape": list(array.shape),
+        "size": int(array.size),
+    }
+    preview_limit = 32
+    flat = array.reshape(-1)
+    preview = flat[:preview_limit].tolist()
+    summary["preview"] = preview
+    summary["preview_count"] = len(preview)
+    if array.size <= preview_limit and array.size <= 10_000:
+        summary["values"] = array.tolist()
+    return summary
+
+
+def serialize_for_view(value: Any, depth: int = 0) -> Any:
+    if depth > 6:
+        return repr(value)
+
+    if value is None or isinstance(value, (str, int, float, bool)):
+        return value
+
+    if isinstance(value, (np.integer, np.floating, np.bool_)):
+        return value.item()
+
+    if isinstance(value, dict):
+        return {str(key): serialize_for_view(val, depth + 1) for key, val in value.items()}
+
+    if isinstance(value, (list, tuple, set)):
+        return [serialize_for_view(val, depth + 1) for val in value]
+
+    if isinstance(value, np.ndarray):
+        return numpy_summary(value)
+
+    return repr(value)
+
+
+def build_tree(root: Path) -> Dict[str, Any]:
+    root_node: Dict[str, Any] = {"name": root.name, "path": "", "type": "dir", "children": []}
+    nodes: Dict[str, Dict[str, Any]] = {"": root_node}
+
+    for file_path in sorted(root.rglob("*.pkl")):
+        rel_path = file_path.relative_to(root)
+        rel_path_posix = rel_path.as_posix()
+        parts = rel_path.parts
+        if not parts:
+            continue
+
+        cumulative = []
+        for part in parts[:-1]:
+            cumulative.append(part)
+            current_key = "/".join(cumulative)
+            parent_key = "/".join(cumulative[:-1]) if len(cumulative) > 1 else ""
+            if current_key not in nodes:
+                node = {"name": part, "path": current_key, "type": "dir", "children": []}
+                nodes[current_key] = node
+                nodes[parent_key]["children"].append(node)
+        file_parent_key = "/".join(parts[:-1]) if len(parts) > 1 else ""
+        file_node = {"name": parts[-1], "path": rel_path_posix, "type": "file"}
+        nodes[file_parent_key]["children"].append(file_node)
+
+    def sort_children(node: Dict[str, Any]) -> None:
+        children = node.get("children")
+        if not children:
+            return
+        children.sort(key=lambda item: (item.get("type") != "dir", item.get("name", "")))
+        for child in children:
+            if child.get("type") == "dir":
+                sort_children(child)
+
+    sort_children(root_node)
+    return root_node
+
+
+def data_proto_to_payload(file_path: Path) -> Dict[str, Any]:
+    data = DataProto.load_from_disk(str(file_path))
+    length = len(data)
+    items: List[Dict[str, Any]] = []
+    for idx in range(length):
+        try:
+            item = data[idx]
+        except Exception as exc:  # pragma: no cover - defensive guard
+            LOGGER.warning("Failed to read item %s from %s: %s", idx, file_path, exc)
+            continue
+        items.append(
+            {
+                "index": idx,
+                "meta_info": serialize_for_view(item.meta_info),
+                "non_tensor_batch": serialize_for_view(item.non_tensor_batch),
+            }
+        )
+
+    return {
+        "path": str(file_path),
+        "length": length,
+        "meta_info": serialize_for_view(data.meta_info),
+        "items": items,
+    }
+
+
+class RolloutExplorer:
+    def __init__(self, root: Path):
+        self.root = root
+        self._tree_cache: Dict[str, Any] | None = None
+        self._lock = threading.Lock()
+
+    def tree(self) -> Dict[str, Any]:
+        with self._lock:
+            if self._tree_cache is None:
+                self._tree_cache = build_tree(self.root)
+            return self._tree_cache
+
+    @lru_cache(maxsize=32)
+    def load_file(self, relative_path: str) -> Dict[str, Any]:
+        normalized_path = Path(relative_path)
+        target = ensure_within(self.root / normalized_path, self.root)
+        if not target.exists() or not target.is_file():
+            raise FileNotFoundError(f"File {relative_path} not found under {self.root}")
+        payload = data_proto_to_payload(target)
+        payload["relative_path"] = target.relative_to(self.root).as_posix()
+        return payload
+
+
+HTML_PAGE = """<!DOCTYPE html>
+<html lang=\"en\">
+<head>
+    <meta charset=\"utf-8\" />
+    <title>Rollout Visualizer</title>
+    <style>
+        :root {
+            color-scheme: light dark;
+            --bg: #f7f7fb;
+            --panel: #ffffffcc;
+            --accent: #4a6cff;
+            --accent-soft: #e6ebff;
+            --text: #1d1d25;
+            --border: #d9d9e3;
+        }
+        * { box-sizing: border-box; }
+        body {
+            margin: 0;
+            font-family: "Segoe UI", Tahoma, sans-serif;
+            background: var(--bg);
+            color: var(--text);
+        }
+        header {
+            padding: 14px 24px;
+            background: linear-gradient(135deg, var(--accent), #7f9bff);
+            color: white;
+            font-weight: 600;
+            letter-spacing: 0.4px;
+        }
+        #layout {
+            display: flex;
+            height: calc(100vh - 56px);
+        }
+        #sidebar {
+            width: 28%;
+            max-width: 360px;
+            min-width: 240px;
+            border-right: 1px solid var(--border);
+            background: var(--panel);
+            padding: 12px 16px;
+            overflow-y: auto;
+        }
+        #content {
+            flex: 1;
+            overflow-y: auto;
+            padding: 20px 28px;
+        }
+        .tree-node {
+            margin-left: 12px;
+        }
+        .tree-toggle {
+            cursor: pointer;
+            user-select: none;
+            display: inline-flex;
+            align-items: center;
+            gap: 6px;
+            padding: 4px 6px;
+            border-radius: 6px;
+        }
+        .tree-toggle:hover {
+            background: var(--accent-soft);
+        }
+        .file-entry {
+            cursor: pointer;
+            display: block;
+            padding: 4px 8px;
+            margin: 2px 0;
+            border-radius: 6px;
+        }
+        .file-entry:hover,
+        .file-entry.active {
+            background: var(--accent-soft);
+            color: var(--accent);
+        }
+        .panel {
+            background: var(--panel);
+            border: 1px solid var(--border);
+            border-radius: 12px;
+            padding: 16px 20px;
+            box-shadow: 0 4px 16px rgba(76, 96, 255, 0.05);
+        }
+        .section-title {
+            font-weight: 600;
+            margin-bottom: 12px;
+            font-size: 18px;
+        }
+        .meta-grid {
+            display: grid;
+            grid-template-columns: repeat(auto-fit, minmax(220px, 1fr));
+            gap: 12px;
+        }
+        .section-subtitle {
+            font-weight: 600;
+            margin: 18px 0 10px;
+            color: var(--accent);
+            font-size: 16px;
+        }
+        .kv-block {
+            border: 1px solid var(--border);
+            border-radius: 10px;
+            padding: 12px;
+            background: #fff;
+        }
+        .kv-header {
+            font-weight: 600;
+            margin-bottom: 8px;
+            color: var(--accent);
+        }
+        .kv-body {
+            font-size: 14px;
+            line-height: 1.5;
+            white-space: pre-wrap;
+        }
+        details {
+            border: 1px solid var(--border);
+            border-radius: 10px;
+            padding: 10px 14px;
+            margin-bottom: 10px;
+            background: #fff;
+        }
+        details[open] {
+            border-color: var(--accent);
+            box-shadow: 0 4px 12px rgba(74, 108, 255, 0.08);
+        }
+        summary {
+            cursor: pointer;
+            font-weight: 600;
+            color: var(--accent);
+        }
+        .message-card {
+            border: 1px solid var(--border);
+            border-radius: 8px;
+            padding: 10px 12px;
+            margin: 6px 0;
+            background: #fbfbff;
+        }
+        .message-meta {
+            font-size: 13px;
+            opacity: 0.7;
+            margin-bottom: 4px;
+        }
+        .message-content {
+            white-space: pre-wrap;
+            font-family: "Fira Code", "Consolas", monospace;
+            font-size: 14px;
+        }
+        .messages-section {
+            border: 1px solid var(--border);
+            border-radius: 12px;
+            padding: 14px 16px;
+            background: #f3f5ff;
+            margin-bottom: 14px;
+            box-shadow: inset 0 0 0 1px rgba(74, 108, 255, 0.05);
+        }
+        .messages-section .section-subtitle {
+            margin-top: 0;
+            color: var(--accent);
+        }
+        .placeholder {
+            opacity: 0.6;
+            font-style: italic;
+        }
+    </style>
+</head>
+<body>
+    <header>Rollout Visualizer</header>
+    <div id="layout">
+        <aside id="sidebar">
+            <div id="tree"></div>
+        </aside>
+        <main id="content">
+            <div class="panel placeholder">Select a file to inspect its rollout details.</div>
+        </main>
+    </div>
+    <script>
+        let activePath = null;
+
+        async function fetchJSON(url) {
+            const res = await fetch(url);
+            if (!res.ok) {
+                const text = await res.text();
+                throw new Error(text || res.statusText);
+            }
+            return res.json();
+        }
+
+        function createElement(tag, options = {}) {
+            const el = document.createElement(tag);
+            if (options.className) el.className = options.className;
+            if (options.text) el.textContent = options.text;
+            if (options.html) el.innerHTML = options.html;
+            return el;
+        }
+
+        function renderTree(node, container) {
+            const wrapper = createElement('div', { className: 'tree-node' });
+            const hasChildren = Array.isArray(node.children) && node.children.length > 0;
+
+            if (node.type === 'dir') {
+                const summary = createElement('div', { className: 'tree-toggle' });
+                const icon = createElement('span', { text: hasChildren ? '▸' : '•' });
+                icon.dataset.state = 'collapsed';
+                const label = createElement('span', { text: node.name || '(root)' });
+                summary.append(icon, label);
+                wrapper.appendChild(summary);
+                const childrenContainer = createElement('div');
+                childrenContainer.style.display = 'none';
+                if (node.path === '') {
+                    childrenContainer.style.display = 'block';
+                    icon.textContent = '▾';
+                }
+                summary.addEventListener('click', () => {
+                    if (!hasChildren) return;
+                    if (childrenContainer.style.display === 'none') {
+                        childrenContainer.style.display = 'block';
+                        icon.textContent = '▾';
+                    } else {
+                        childrenContainer.style.display = 'none';
+                        icon.textContent = '▸';
+                    }
+                });
+                wrapper.appendChild(childrenContainer);
+                node.children.forEach(child => renderTree(child, childrenContainer));
+            } else if (node.type === 'file') {
+                const entry = createElement('div', { className: 'file-entry', text: node.name });
+                entry.dataset.path = node.path;
+                entry.addEventListener('click', () => loadFile(node.path, entry));
+                wrapper.appendChild(entry);
+            }
+            container.appendChild(wrapper);
+        }
+
+        function renderKeyValue(container, key, value) {
+            const block = createElement('div', { className: 'kv-block' });
+            block.appendChild(createElement('div', { className: 'kv-header', text: key }));
+            const body = createElement('div', { className: 'kv-body' });
+            body.appendChild(renderValue(value, key));
+            block.appendChild(body);
+            container.appendChild(block);
+        }
+
+        function renderMessages(messages) {
+            const wrapper = createElement('div');
+            messages.forEach((msg, idx) => {
+                const card = createElement('div', { className: 'message-card' });
+                const role = msg.role || msg.author || `Message ${idx}`;
+                const meta = createElement('div', { className: 'message-meta', text: `${role}` });
+                if (msg.timestamp) {
+                    meta.textContent += ` · ${msg.timestamp}`;
+                }
+                const contentContainer = createElement('div', { className: 'message-content' });
+                let content = msg.content;
+                if (Array.isArray(content)) {
+                    content = content.map(part => typeof part === 'string' ? part : JSON.stringify(part, null, 2)).join('\\n');
+                }
+                contentContainer.textContent = content ?? '';
+                card.append(meta, contentContainer);
+                wrapper.appendChild(card);
+            });
+            return wrapper;
+        }
+
+        function renderNdArray(info) {
+            const wrapper = createElement('div');
+            const summary = `dtype=${info.dtype} · shape=[${info.shape.join(', ')}] · size=${info.size}`;
+            wrapper.appendChild(createElement('div', { text: summary }));
+            if (info.preview && info.preview.length) {
+                const preview = createElement('pre');
+                preview.textContent = JSON.stringify(info.preview, null, 2);
+                wrapper.appendChild(preview);
+            }
+            if (info.values) {
+                const details = document.createElement('details');
+                details.appendChild(createElement('summary', { text: 'Show full values' }));
+                const pre = createElement('pre');
+                pre.textContent = JSON.stringify(info.values, null, 2);
+                details.appendChild(pre);
+                wrapper.appendChild(details);
+            }
+            return wrapper;
+        }
+
+        function renderValue(value, key = '') {
+            if (value === null || typeof value === 'undefined') {
+                return createElement('span', { text: '—' });
+            }
+            if (typeof value !== 'object') {
+                return createElement('span', { text: String(value) });
+            }
+            if (Array.isArray(value)) {
+                if (key === 'messages') {
+                    return renderMessages(value.map(item => typeof item === 'object' ? item : { content: String(item) }));
+                }
+                const details = document.createElement('details');
+                details.appendChild(createElement('summary', { text: `List [${value.length}]` }));
+                value.forEach((item, idx) => {
+                    const line = createElement('div');
+                    line.appendChild(createElement('strong', { text: `#${idx}` }));
+                    line.appendChild(createElement('div', { className: 'kv-body' }));
+                    line.lastChild.appendChild(renderValue(item));
+                    details.appendChild(line);
+                });
+                return details;
+            }
+            if (value.__type__ === 'ndarray') {
+                return renderNdArray(value);
+            }
+            const entries = Object.entries(value);
+            const container = createElement('div');
+            entries.forEach(([childKey, childValue]) => {
+                const block = createElement('div');
+                block.appendChild(createElement('strong', { text: childKey }));
+                const inner = createElement('div', { className: 'kv-body' });
+                inner.appendChild(renderValue(childValue, childKey));
+                block.appendChild(inner);
+                container.appendChild(block);
+            });
+            return container;
+        }
+
+        function markActive(entry) {
+            document.querySelectorAll('.file-entry.active').forEach(el => el.classList.remove('active'));
+            entry.classList.add('active');
+        }
+
+        async function loadFile(path, entryEl) {
+            try {
+                activePath = path;
+                markActive(entryEl);
+                const data = await fetchJSON(`/api/file?path=${encodeURIComponent(path)}`);
+                renderContent(data);
+            } catch (error) {
+                console.error(error);
+                const panel = createElement('div', { className: 'panel' });
+                panel.appendChild(createElement('h2', { text: 'Failed to load file' }));
+                panel.appendChild(createElement('pre', { text: error.message }));
+                const content = document.getElementById('content');
+                content.innerHTML = '';
+                content.appendChild(panel);
+            }
+        }
+
+        function renderContent(data) {
+            const content = document.getElementById('content');
+            content.innerHTML = '';
+            const panel = createElement('div', { className: 'panel' });
+            const title = createElement('div', { className: 'section-title', text: data.relative_path || data.path });
+            panel.appendChild(title);
+            panel.appendChild(createElement('div', { text: `Entries: ${data.length}` }));
+
+            const metaSection = createElement('div', { className: 'section-title', text: 'Meta Info (global)' });
+            panel.appendChild(metaSection);
+            const metaGrid = createElement('div', { className: 'meta-grid' });
+            Object.entries(data.meta_info || {}).forEach(([key, value]) => {
+                renderKeyValue(metaGrid, key, value);
+            });
+            if (!Object.keys(data.meta_info || {}).length) {
+                metaGrid.appendChild(createElement('div', { className: 'placeholder', text: 'No meta info available.' }));
+            }
+            panel.appendChild(metaGrid);
+
+            const itemsSection = createElement('div', { className: 'section-title', text: 'Entries' });
+            panel.appendChild(itemsSection);
+            if (!data.items.length) {
+                panel.appendChild(createElement('div', { className: 'placeholder', text: 'No entries in this DataProto.' }));
+            }
+            data.items.forEach(item => {
+                const details = document.createElement('details');
+                const summary = createElement('summary', { text: `Item #${item.index}` });
+                details.appendChild(summary);
+
+                const metaBlock = createElement('div', { className: 'meta-grid' });
+                Object.entries(item.meta_info || {}).forEach(([key, value]) => {
+                    renderKeyValue(metaBlock, key, value);
+                });
+                if (!Object.keys(item.meta_info || {}).length) {
+                    metaBlock.appendChild(createElement('div', { className: 'placeholder', text: 'No item-level meta info.' }));
+                }
+                details.appendChild(metaBlock);
+
+                const nonTensorEntries = Object.entries(item.non_tensor_batch || {});
+                let messagesHandled = false;
+                if (nonTensorEntries.length) {
+                    nonTensorEntries.forEach(([key, value]) => {
+                        if (key === 'messages_list') {
+                            const messagesSection = createElement('div', { className: 'messages-section' });
+                            messagesSection.appendChild(createElement('div', { className: 'section-subtitle', text: 'Messages' }));
+                            messagesSection.appendChild(renderValue(value, key));
+                            details.appendChild(messagesSection);
+                            messagesHandled = true;
+                        }
+                    });
+                    const others = nonTensorEntries.filter(([key]) => key !== 'messages_list');
+                    if (others.length) {
+                        const ntBlock = createElement('div', { className: 'meta-grid' });
+                        others.forEach(([key, value]) => {
+                            renderKeyValue(ntBlock, key, value);
+                        });
+                        details.appendChild(ntBlock);
+                    }
+                    if (!messagesHandled && !others.length) {
+                        details.appendChild(createElement('div', { className: 'placeholder', text: 'No non-tensor batch data.' }));
+                    }
+                } else {
+                    details.appendChild(createElement('div', { className: 'placeholder', text: 'No non-tensor batch data.' }));
+                }
+
+                panel.appendChild(details);
+            });
+
+            content.appendChild(panel);
+        }
+
+        async function init() {
+            try {
+                const treeData = await fetchJSON('/api/tree');
+                const treeRoot = document.getElementById('tree');
+                treeRoot.innerHTML = '';
+                renderTree(treeData, treeRoot);
+            } catch (error) {
+                const treeRoot = document.getElementById('tree');
+                treeRoot.textContent = 'Failed to load file tree.';
+                console.error(error);
+            }
+        }
+
+        init();
+    </script>
+</body>
+</html>
+"""
+
+
+class VisualizerHandler(BaseHTTPRequestHandler):
+    explorer: RolloutExplorer
+
+    def do_GET(self) -> None:  # noqa: N802 - http.server signature
+        parsed = urlparse(self.path)
+        if parsed.path == "/":
+            self.respond_html(HTML_PAGE)
+            return
+        if parsed.path == "/api/tree":
+            payload = VisualizerHandler.explorer.tree()
+            self.respond_json(payload)
+            return
+        if parsed.path == "/api/file":
+            query = parse_qs(parsed.query)
+            relative = query.get("path", [None])[0]
+            if not relative:
+                self.respond_json({"error": "Missing path query parameter"}, status=400)
+                return
+            try:
+                payload = VisualizerHandler.explorer.load_file(relative)
+            except FileNotFoundError:
+                self.respond_json({"error": "File not found"}, status=404)
+                return
+            except Exception as exc:  # pragma: no cover - defensive guard
+                LOGGER.exception("Failed to load %s", relative)
+                self.respond_json({"error": str(exc)}, status=500)
+                return
+            self.respond_json(payload)
+            return
+
+        self.respond_json({"error": "Not found"}, status=404)
+
+    def log_message(self, format: str, *args: Any) -> None:  # noqa: A003 - inherited name
+        LOGGER.info("%s - %s", self.address_string(), format % args)
+
+    def respond_json(self, payload: Any, status: int = 200) -> None:
+        body = json.dumps(payload).encode("utf-8")
+        self.send_response(status)
+        self.send_header("Content-Type", "application/json; charset=utf-8")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+    def respond_html(self, html: str, status: int = 200) -> None:
+        body = html.encode("utf-8")
+        self.send_response(status)
+        self.send_header("Content-Type", "text/html; charset=utf-8")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+
+def main() -> None:
+    logging.basicConfig(level=logging.INFO, format="[%(levelname)s] %(message)s")
+    args = parse_args()
+    root = Path(args.rollout_path).expanduser().resolve()
+    if not root.exists() or not root.is_dir():
+        raise SystemExit(f"Rollout path {root} does not exist or is not a directory")
+
+    explorer = RolloutExplorer(root)
+    VisualizerHandler.explorer = explorer
+
+    server = ThreadingHTTPServer((args.host, args.port), VisualizerHandler)
+
+    address = f"http://{args.host}:{args.port}/"
+    print(f"Serving rollout visualizer for {root} at {address}")
+    if not args.no_browser:
+        try:
+            webbrowser.open(address)
+        except Exception as exc:  # pragma: no cover - best effort
+            LOGGER.info("Could not open browser automatically: %s", exc)
+
+    try:
+        server.serve_forever()
+    except KeyboardInterrupt:
+        print("\nShutting down...")
+    finally:
+        server.server_close()
+
+
+if __name__ == "__main__":
+    main()

tests/env/test_sokoban_render.py

@@ -0,0 +1,41 @@
+import re
+
+from ragen.env.sokoban.config import SokobanEnvConfig
+from ragen.env.sokoban.env import SokobanEnv
+
+
+def test_sokoban_render_supports_grid_and_coord():
+    seed = 1234
+    grid_config = SokobanEnvConfig(
+        dim_room=(5, 5),
+        num_boxes=1,
+        max_steps=10,
+        search_depth=20,
+        observation_format="grid",
+    )
+    coord_config = SokobanEnvConfig(
+        dim_room=(5, 5),
+        num_boxes=1,
+        max_steps=10,
+        search_depth=20,
+        observation_format="coord",
+    )
+
+    grid_env = SokobanEnv(grid_config)
+    coord_env = SokobanEnv(coord_config)
+
+    try:
+        grid_obs = grid_env.reset(seed=seed)
+        coord_obs = coord_env.reset(seed=seed)
+
+        assert isinstance(grid_obs, str)
+        assert isinstance(coord_obs, str)
+
+        assert "Board size:" in coord_obs
+        assert re.search(r"Walls: \(\d+, \d+\)", coord_obs)
+
+        assert isinstance(coord_env.render(mode="grid"), str)
+        assert "Board size:" in grid_env.render(mode="coord")
+    finally:
+        grid_env.close()
+        coord_env.close()

tests/es_manager/test_seed_iteration.py

@@ -0,0 +1,34 @@
+import pytest
+from omegaconf import OmegaConf
+from ragen.llm_agent.es_manager import EnvStateManager
+
+
+def make_cfg():
+    return OmegaConf.create({
+        'seed': {'train': 7},
+        'es_manager': {
+            'train': {
+                'env_groups': 1,
+                'group_size': 1,
+                'env_configs': {'tags': ['Bandit'], 'n_groups': [1]},
+            }
+        },
+        'custom_envs': {
+            'Bandit': {
+                'env_type': 'bandit',
+                'max_actions_per_traj': 1,
+                'env_config': None
+            }
+        }
+    })
+
+
+def test_seed_iteration():
+    cfg = make_cfg()
+    es = EnvStateManager(cfg, mode='train')
+    es.reset()
+    first_seed = es.envs[0]['status'].seed
+    es.reset()
+    second_seed = es.envs[0]['status'].seed
+    assert first_seed == 7
+    assert second_seed == 8

tests/llm_agent/test_context_window.py

@@ -0,0 +1,84 @@
+import pytest
+from ragen.llm_agent.ctx_manager import ContextManager
+from omegaconf import OmegaConf
+from verl.verl.protocol import DataProto
+
+class DummyTokenizer:
+    name_or_path = "qwen"  # or "llama-3" or any string your code expects
+
+    def apply_chat_template(self, messages, add_generation_prompt, tokenize):
+        return " ".join([msg["content"] for msg in messages])
+
+    def __call__(self, texts, return_tensors, padding, padding_side, truncation):
+        import torch
+        class DummyOutput:
+            input_ids = torch.tensor([[1, 2, 3]])
+            attention_mask = torch.tensor([[1, 1, 1]])
+        return DummyOutput()
+
+    def encode(self, text):
+        # Return a dummy list of token ids; must be at least length 1 for [0] indexing
+        return [42, 43]
+
+@pytest.fixture
+def dummy_config():
+    cfg = OmegaConf.create({
+        "agent_proxy": {
+            "max_context_window": 2,
+            "enable_think": False,
+            "use_turn_scores": False,
+            "action_sep": "|",
+            "reward_normalization": {
+                "grouping": "batch",
+                "method": "identity"
+            }
+        },
+        "enable_response_mask": False,
+        "es_manager": {
+            "train": {
+                "env_configs": {
+                    "n_groups": [1],
+                    "tags": ["sokoban"]
+                },
+                "group_size": 1
+            }
+        },
+        "custom_envs": {
+            "sokoban": {
+                "env_type": "sokoban",
+                "max_actions_per_traj": 10
+            }
+        },
+        "actor_rollout_ref": {
+            "rollout": {
+                "response_length": 128
+            }
+        }
+    })
+    return cfg
+
+def test_context_window_truncation(dummy_config):
+    tokenizer = DummyTokenizer()
+    ctx = ContextManager(config=dummy_config, tokenizer=tokenizer, mode="train")
+    ctx.prefix_lookup = {0: "Initial prompt"}
+    ctx.env_config_lookup = {0: {"max_tokens": 128}}
+    ctx.env_nums = {"": 1}  # For metrics
+
+    env_outputs = [{
+        "env_id": 0,
+        "group_id": 0,
+        "history": [
+            {"state": "S1", "llm_response": "R1", "reward": 0.1, "actions_left": 5},
+            {"state": "S2", "llm_response": "R2", "reward": 0.2, "actions_left": 4},
+            {"state": "S3", "llm_response": "R3", "reward": 0.3, "actions_left": 3},
+        ],
+        "metrics": {},
+    }]
+
+    lm_inputs: DataProto = ctx.get_lm_inputs(env_outputs, prepare_for_update=True)
+    messages = lm_inputs.non_tensor_batch["messages_list"][0]
+
+    # Ensure only last 2 turns are present
+    assert "S1" not in str(messages)
+    assert "S2" in str(messages)
+    assert "S3" in str(messages)

verl/.gemini/config.yaml

@@ -0,0 +1,10 @@
+have_fun: false
+code_review:
+  disable: false
+  comment_severity_threshold: HIGH
+  max_review_comments: -1
+  pull_request_opened:
+    help: false
+    summary: false
+    code_review: true
+ignore_patterns: []

verl/.github/CODEOWNERS

@@ -0,0 +1,30 @@
+/docs @eric-haibin-lin @zhaochenyang20 @hongpeng-guo
+/docs/amd_tutorial @yushengsu-thu
+/docs/slang_multiturn @zhaochenyang20 @SwordFaith
+/docs/ascend_tutorial @FightingZhen
+
+/recipe/dapo @tongyx361 @PeterSH6 @vermouth1992 @tardis-key @FightingZhen @ji-huazhong
+/recipe/spin @zhaochenyang20
+/recipe/sppo @zhaochenyang20
+
+/third_party/sglang @zhaochenyang20 @SwordFaith
+/third_party/vllm @PeterSH6 @wuxibin89
+
+/examples/grpo_trainer @vermouth1992 @PeterSH6 @tardis-key @FightingZhen @ji-huazhong
+
+/verl/single_controller @zw0610 @wuxibin89 @hongpeng-guo
+/verl/trainer @eric-haibin-lin @vermouth1992 @tongyx361 @PeterSH6
+/verl/models/mcore @ISEEKYAN @vermouth1992
+/verl/models/transformers @vermouth1992 @PeterSH6 @tardis-key @FightingZhen @ji-huazhong
+/verl/workers/engine @eric-haibin-lin @vermouth1992 @ZihengJiang
+/verl/workers/roles @eric-haibin-lin @vermouth1992 @ZihengJiang
+/verl/workers/engine/fsdp @eric-haibin-lin @vermouth1992 @ZihengJiang
+/verl/workers/rollout/vllm_rollout @wuxibin89 @PeterSH6 @chenhaiq
+/verl/workers/rollout/sglang_rollout @zhaochenyang20 @SwordFaith @chenhaiq
+/verl/workers/actor/megatron_actor.py @ISEEKYAN @vermouth1992
+/verl/workers/critic/megatron_critic.py @ISEEKYAN @vermouth1992
+/verl/workers/megatron_workers.py @ISEEKYAN @vermouth1992
+
+/tests/single_controller @zw0610 @wuxibin89
+/tests/trainer @eric-haibin-lin @vermouth1992 @tongyx361 @PeterSH6
+/tests/workers/rollout/vllm_rollout @wuxibin89 @PeterSH6 @chenhaiq

verl/.github/ISSUE_TEMPLATE/bug-report.yml

@@ -0,0 +1,65 @@
+# modified from https://github.com/huggingface/transformers/blob/main/.github/ISSUE_TEMPLATE/bug-report.yml?plain=1
+name: "\U0001F41B Bug Report"
+description: Submit a bug report to help us improve verl
+labels: [ "bug" ]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for taking the time to fill out this bug report! 🤗
+
+  - type: textarea
+    id: system-info
+    attributes:
+      label: System Info
+      description: Please share your system info with us. You can run the command `python scripts/diagnose.py` and copy-paste its output below.
+      placeholder: verl version, platform, python version, ...
+    validations:
+      required: true
+
+  - type: checkboxes
+    id: information-scripts-examples
+    attributes:
+      label: Information
+      description: 'The problem arises when using:'
+      options:
+        - label: "The official example scripts"
+        - label: "My own modified scripts"
+
+  - type: checkboxes
+    id: information-tasks
+    attributes:
+      label: Tasks
+      description: "The tasks I am working on are:"
+      options:
+        - label: "An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)"
+        - label: "My own task or dataset (give details below)"
+
+  - type: textarea
+    id: reproduction
+    validations:
+      required: true
+    attributes:
+      label: Reproduction
+      description: |
+        Please provide a code sample that reproduces the problem you ran into. It can be a Colab link or just a code snippet.
+        Please include relevant config information with your code.
+        If you have code snippets, error messages, stack traces please provide them here as well.
+        Important! Use code tags to correctly format your code. See https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
+        Do not use screenshots, as they are hard to read and (more importantly) don't allow others to copy-and-paste your code.
+
+      placeholder: |
+        Steps to reproduce the behavior:
+
+          1.
+          2.
+          3.
+
+
+  - type: textarea
+    id: expected-behavior
+    validations:
+      required: true
+    attributes:
+      label: Expected behavior
+      description: "A clear and concise description of what you would expect to happen."

verl/.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,2 @@
+blank_issues_enabled: true
+version: 0.1

verl/.github/ISSUE_TEMPLATE/feature-request.yml

@@ -0,0 +1,32 @@
+# modified from https://github.com/huggingface/transformers/blob/main/.github/ISSUE_TEMPLATE/feature-request.yml?plain=1
+name: "\U0001F680 Feature request"
+description: Submit a proposal/request for a new verl feature
+labels: [ "Feature request" ]
+body:
+  - type: textarea
+    id: feature-request
+    validations:
+      required: true
+    attributes:
+      label: Feature request
+      description: |
+        A clear and concise description of the feature proposal. Please provide a link to the paper and code in case they exist.
+
+  - type: textarea
+    id: motivation
+    validations:
+      required: true
+    attributes:
+      label: Motivation
+      description: |
+        Please outline the motivation for the proposal. Is your feature request related to a problem? e.g., I'm always frustrated when [...]. If this is related to another GitHub issue, please link here too.
+
+
+  - type: textarea
+    id: contribution
+    validations:
+      required: true
+    attributes:
+      label: Your contribution
+      description: |
+        Is there any way that you could help, e.g. by submitting a PR? Make sure to read the CONTRIBUTING.MD [readme](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md)

verl/.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,40 @@
+### What does this PR do?
+
+> Add **concise** overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review.
+
+### Checklist Before Starting
+
+- [ ] Search for similar PRs. Paste at least one query link here: ...
+- [ ] Format the PR title as `[{modules}] {type}: {description}` (This will be checked by the CI)
+  - `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`, `trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`, `ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`, `env`, `tool`, `ckpt`, `doc`, `data`
+  - If this PR involves multiple modules, separate them with `,` like `[megatron, fsdp, doc]`
+  - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test`
+  - If this PR breaks any API (CLI arguments, config, function signature, etc.), add `[BREAKING]` to the beginning of the title.
+  - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching`
+
+### Test
+
+> For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc.
+
+### API and Usage Example
+
+> Demonstrate how the API changes if any, and provide usage example(s) if possible.
+
+```python
+# Add code snippet or script demonstrating how to use this
+```
+
+### Design & Code Changes
+
+> Demonstrate the high-level design if this PR is complex, and list the specific changes.
+
+### Checklist Before Submitting
+
+> [!IMPORTANT]
+> Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.
+
+- [ ] Read the [Contribute Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md).
+- [ ] Apply [pre-commit checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting): `pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always`
+- [ ] Add / Update [the documentation](https://github.com/volcengine/verl/tree/main/docs).
+- [ ] Add unit or end-to-end test(s) to [the CI workflow](https://github.com/volcengine/verl/tree/main/.github/workflows) to cover all the code. If not feasible, explain why: ...
+- [ ] Once your PR is ready for CI, send a message in [the `ci-request` channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the `verl` Slack workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ). (If not accessible, please try [the Feishu group (飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)

verl/.github/dependabot.yml

@@ -0,0 +1,9 @@
+## Enabled the dependabot to check the dependencies of the project
+## Dependabot will open pull requests to update dependencies automatically
+
+version: 2
+updates:
+  - package-ecosystem: pip
+    directory: "/"
+    schedule:
+      interval: weekly

verl/.github/workflows/.deprecate/e2e_eval_aime24.yml

@@ -0,0 +1,147 @@
+# # Tests layout
+
+# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:
+# - `tests/trainer` for testing functionality related to `verl/trainer`
+# - `tests/models` for testing functionality related to `verl/models`
+# - ...
+
+# There are a few folders with `special_` prefix, created for special purposes:
+# - `special_distributed`: unit tests that must run with multiple GPUs
+# - `special_e2e`: end-to-end tests with training/generation scripts
+# - `special_npu`: tests for NPUs
+# - `special_sanity`: a suite of quick sanity tests
+# - `special_standalone`: a set of test that are designed to run in dedicated environments
+
+# Accelerators for tests 
+# - By default tests are run with GPU available, except for the ones under `special_npu`, and any test script whose name ends with `on_cpu.py`.
+# - For test scripts with `on_cpu.py` name suffix would be tested on CPU resources in linux environment.
+
+# # Workflow layout
+
+# All CI tests are configured by yaml files in `.github/workflows/`. Here's an overview of all test configs:
+# 1. A list of always triggered CPU sanity tests: `check-pr-title.yml`, `secrets_scan.yml`, `check-pr-title,yml`, `pre-commit.yml`, `doc.yml`
+# 2. Some heavy multi-GPU unit tests, such as `model.yml`, `vllm.yml`, `sgl.yml`
+# 3. End-to-end tests: `e2e_*.yml`
+# 4. Unit tests
+#   - `cpu_unit_tests.yml`, run pytest on all scripts with file name pattern `tests/**/test_*_on_cpu.py`
+#   - `gpu_unit_tests.yml`, run pytest on all scripts with file without the `on_cpu.py` suffix.
+#   - Since cpu/gpu unit tests by default runs all tests under `tests`, please make sure tests are manually excluded in them when
+#     - new workflow yaml is added to `.github/workflows`
+#     - new tests are added to workflow mentioned in 2.
+
+
+name: e2e_eval_aime24
+
+on:
+  # Trigger the workflow on push or pull request,
+  # but only for the main branch
+  # For push, for now only anti-patterns are specified so it is more conservative
+  # and achieves higher coverage.
+  push:
+    branches:
+      - main
+      - v0.*
+    paths:
+      - "**/*.py"
+      # Other entrypoints
+      - "!*.md"
+      - "!docker/**"
+      - "!docs/**"
+      - "!examples/**"
+      - "!tests/**"
+      - "!verl/trainer/main_*.py"
+      - "!verl/trainer/fsdp_sft_trainer.py"
+      - "!recipe/**"
+      - "recipe/r1"
+      - "!recipe/r1/README.md"
+  pull_request:
+    branches:
+      - main
+    paths:
+      - "**/*.py"
+      # Other entrypoints
+      - "!*.md"
+      - "!docker/**"
+      - "!docs/**"
+      - "!examples/**"
+      - "!tests/**"
+      - "!verl/trainer/main_*.py"
+      - "!verl/trainer/fsdp_sft_trainer.py"
+      # Home
+      - "recipe/r1"
+      - "!recipe/r1/README.md"
+      # Other recipes
+      - "!recipe/**"
+      # Entrypoints
+      - ".github/workflows/e2e_eval_aime24.yml"
+      - "tests/special_e2e/run_r1_distill_qwen_aime24_eval.sh"
+      - "verl/trainer/main_generation.py"
+      - "verl/trainer/config/generation.yaml"
+
+# Cancel jobs on the same ref if a new one is triggered
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+
+# Declare permissions just read content.
+permissions:
+  contents: read
+
+env:
+  IMAGE: "verl-ci-cn-beijing.cr.volces.com/verlai/verl:app-verl0.5-transformers4.55.4-vllm0.10.0-mcore0.13.0-te2.2"
+  DYNAMIC_RUNNER_ENDPOINT: "https://sd10g3clalm04ug7alq90.apigateway-cn-beijing.volceapi.com/runner"
+
+jobs:
+  setup:
+    if: github.repository_owner == 'volcengine'
+    runs-on: ubuntu-latest
+    outputs:
+      runner-label: ${{ steps.create-runner.outputs.runner-label }}
+      mlp-task-id: ${{ steps.create-runner.outputs.mlp-task-id }}
+    steps:
+      - uses: actions/checkout@v4
+      - id: create-runner
+        uses: volcengine/vemlp-github-runner@v1 
+        with:
+          mode: "create"
+          faas-url: "${{ env.DYNAMIC_RUNNER_ENDPOINT }}"
+          mlp-image: "${{ env.IMAGE }}"
+  
+  e2e_eval_aime24:
+    needs: setup
+    runs-on: ["${{ needs.setup.outputs.runner-label || 'L20x8' }}"]
+    timeout-minutes: 40 # Increase this timeout value as needed
+    env:
+      HTTP_PROXY: ${{ secrets.PROXY_HTTP }}
+      HTTPS_PROXY: ${{ secrets.PROXY_HTTPS }}
+      NO_PROXY: "localhost,127.0.0.1,hf-mirror.com"
+      HF_ENDPOINT: "https://hf-mirror.com"
+      HF_HUB_ENABLE_HF_TRANSFER: "0" # This is more stable
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+      - name: Install the current repository
+        run: |
+          pip3 install --no-deps -e .[test,gpu,math]
+          pip3 install math-verify transformers==4.56.2
+      - name: Prepare aime24 dataset
+        run: |
+          ray stop --force
+          python3 recipe/r1/data_process.py --task aime2024
+      - name: Running generation and evaluation in AIME 2024
+        run: |
+          ray stop --force
+          bash tests/special_e2e/run_r1_distill_qwen_aime24_eval.sh
+
+  cleanup:
+      runs-on: ubuntu-latest
+      needs: [setup, e2e_eval_aime24]
+      if: always()
+      steps:
+        - id: destroy-runner
+          uses: volcengine/vemlp-github-runner@v1
+          with:
+            mode: "destroy"
+            faas-url: "${{ env.DYNAMIC_RUNNER_ENDPOINT }}"
+            mlp-task-id: "${{ needs.setup.outputs.mlp-task-id }}"