Add qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0 step200 LoRA adapter and run config

results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/checkpoints/global_step_200/actor/lora_adapter/adapter_config.json

@@ -0,0 +1,51 @@
+{
+    "task_type": "CAUSAL_LM",
+    "peft_type": "LORA",
+    "auto_mapping": null,
+    "peft_version": "0.19.1",
+    "base_model_name_or_path": "/home/yuxinr/.cache/huggingface/hub/models--Qwen--Qwen2.5-1.5B-Instruct/snapshots/989aa7980e4cf806f80c7fef2b1adb7bc71aa306",
+    "revision": null,
+    "inference_mode": false,
+    "r": 64,
+    "target_modules": [
+        "down_proj",
+        "gate_proj",
+        "k_proj",
+        "o_proj",
+        "q_proj",
+        "up_proj",
+        "v_proj"
+    ],
+    "exclude_modules": null,
+    "lora_alpha": 64,
+    "lora_dropout": 0.0,
+    "fan_in_fan_out": false,
+    "bias": "none",
+    "use_rslora": false,
+    "modules_to_save": null,
+    "init_lora_weights": true,
+    "layers_to_transform": null,
+    "layers_pattern": null,
+    "rank_pattern": {},
+    "alpha_pattern": {},
+    "megatron_config": null,
+    "megatron_core": "megatron.core",
+    "trainable_token_indices": null,
+    "loftq_config": {},
+    "eva_config": null,
+    "corda_config": null,
+    "lora_ga_config": null,
+    "use_dora": false,
+    "alora_invocation_tokens": null,
+    "use_qalora": false,
+    "qalora_group_size": 16,
+    "layer_replication": null,
+    "runtime_config": {
+        "ephemeral_gpu_offload": false
+    },
+    "lora_bias": false,
+    "target_parameters": null,
+    "use_bdlora": null,
+    "arrow_config": null,
+    "ensure_weight_tying": false
+}

results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/checkpoints/global_step_200/actor/lora_adapter/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89aa2560d7d6dbe211e89031626edaa838c8310d5e0a915115de17713beff3a0
+size 295488904

results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/tensorboard/events.out.tfevents.1790134325.train8.3168242.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f479579cd00b1c75908cbdabc26b342fb06b9b2f47b99cafde40f5369c10fb06
+size 38386

results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/tensorboard/events.out.tfevents.1790146713.train8.3292090.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a517c55f5e7af8a117e2816200c2301ef0d85bfadc757b7b8fd85b42d6f20e61
+size 1898681

results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/run_config.json

@@ -0,0 +1,276 @@
+{
+  "base_model": "989aa7980e4cf806f80c7fef2b1adb7bc71aa306",
+  "environment": "webshop",
+  "experiment_name": "qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0",
+  "implementation_sha256": {
+    "scripts/train/clean/run_webshop_1expert_1k_tmux.sh": "505c382630dd089246d8cf7f2101254c932f762338f86291e52118c7eb0d7785",
+    "scripts/train/robust/gaussian/run_training.sh": "3c6e237d1a6d27f01d8aef836ae591f1644c4806abefbef27ed453cdfa0c0c8e",
+    "verl/models/adaptive_ffn_channel_noise.py": "9b5865cbb6142e06a97f20677369dfe50a885641882231f241b1d48f115f9249",
+    "verl/trainer/config/ppo_trainer.yaml": "f0b5654b0bbdf585bad56259942301121d5d694d4bf94aca3f459dd3a6d292e6",
+    "verl/workers/actor/dp_actor.py": "fbaa9cb65975acf52404db4443119118419804098fc6d1e18c7b9417e82ea1b8"
+  },
+  "model_path": "/home/yuxinr/.cache/huggingface/hub/models--Qwen--Qwen2.5-1.5B-Instruct/snapshots/989aa7980e4cf806f80c7fef2b1adb7bc71aa306",
+  "perturbation": {
+    "actor_weights_perturbed": false,
+    "channel_ema_symbol": "lambda_EMA",
+    "checkpoint_state_file": "adaptive_ffn_noise.pt",
+    "distribution": "multiplicative_normal_scale_1_plus_channel_std_times_epsilon",
+    "empty_response_sensitivity_contribution": 0.0,
+    "extra_actor_backward_passes_per_rollout_batch": 1,
+    "extra_environment_rollouts_per_step": 0,
+    "extra_forward_backward_beyond_existing_sensitivity_prepass": 0,
+    "family": "stable_gaussian_noise",
+    "global_scale_estimator": "squared_L2_norm_of_full_batch_mean_PPO_parameter_gradient_from_sensitivity_prepass",
+    "global_scale_used_for": "next_global_step_noise_standard_deviation_only",
+    "global_std_cap_after_warmup": 0.4,
+    "global_std_cap_multiplier": 2.0,
+    "global_std_formula": "sigma_base * warmup * min(updated_policy_full_batch_mean_gradient_squared, clip_multiplier)",
+    "global_weight_clipping": true,
+    "global_weight_ema": false,
+    "gradient_source": "clipped_PPO_policy_objective_only_with_GiGPO_advantages",
+    "initial_variance_allocation": "isotropic_before_first_sensitivity_estimate",
+    "outer_gradient": "PPO_loss_of_perturbed_policy_on_its_noisy_rollout_against_noisy_vllm_behavior_log_probs_with_identical_channel_scale",
+    "rollout_fixed_scale_path_at_engine_initialization": "/home/yuxinr/yuanzhe/RobustLLMAgent/results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/adaptive_rollout_clean_scale.pt",
+    "rollout_perturbation": true,
+    "rollout_scale_path": "/home/yuxinr/yuanzhe/RobustLLMAgent/results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/adaptive_rollout_scale.pt",
+    "rollout_sigma": 0.2,
+    "rollout_sigma_schedule": "warmup_times_base_sigma_times_clipped_post_update_gradient_squared",
+    "sampling_mode": "step",
+    "scope": "one_per_layer_channel_vector_shared_by_all_rollouts_and_actor_mini_batches_in_one_global_step",
+    "sensitivity_estimator": "mean_over_rollout_rows_of_squared_channel_gain_gradient_of_sequence_mean_policy_loss",
+    "sensitivity_is_fisher": false,
+    "sensitivity_normalization": "per_layer_mean_before_EMA",
+    "sensitivity_prepass": "updated_noisy_policy_over_entire_existing_rollout_batch_after_all_optimizer_updates",
+    "sensitivity_used_for": "next_global_step_variance_only; current_rollout_and_PPO_scale_remains_frozen",
+    "settings": {
+      "allow_schedule_override": false,
+      "alpha": 0.0,
+      "damping": 0.01,
+      "ema_decay": 0.9,
+      "enabled": true,
+      "global_scale_clip_multiplier": 2.0,
+      "global_scale_fixed_multiplier": -1.0,
+      "global_scale_floor": 0.0,
+      "global_scale_mode": "prepass_grad_squared",
+      "initial_global_scale": 1.0,
+      "max_variance_ratio": 4.0,
+      "min_variance_ratio": 0.25,
+      "seed": 0,
+      "sigma": 0.2,
+      "statistics_timing": "post_update",
+      "warmup_steps": 20
+    },
+    "statistics_gradient_clipped": false,
+    "statistics_timing": "post_update",
+    "stop_gradient": [
+      "sensitivity",
+      "EMA",
+      "variance",
+      "sampled_channel_noise"
+    ],
+    "type": "adaptive_gaussian_ffn_channel_noise",
+    "update_mode": "lagged_rollout_adaptive",
+    "validation_clean_scale_path": "/home/yuxinr/yuanzhe/RobustLLMAgent/results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/adaptive_rollout_clean_scale.pt",
+    "validation_perturbation": false,
+    "variance_allocation": "inverse_damped_EMA_sensitivity_power_alpha_with_bounded_unit_layer_mean_variance_ratios",
+    "variance_projection": "scalar_rescale_with_bounds_and_exact_layer_mean_ratio_1",
+    "variance_weight_formula": "(layer_normalized_EMA_sensitivity+damping)^(-alpha)",
+    "warmup_formula": "sigma*min((completed_adaptive_updates+1)/warmup_steps,1); full_sigma_when_warmup_steps_is_0"
+  },
+  "result_category": "robust_training",
+  "resume_history": [
+    {
+      "architecture": "x86_64",
+      "dataloader_num_workers": 0,
+      "expected_torch": "2.8.0+cu128",
+      "free_cache_engine": false,
+      "git_commit_at_launch": "6b2abd6d16a4bdfc0987712e01a603a9fd50e324",
+      "git_worktree_dirty_at_launch": true,
+      "host": "train8",
+      "launcher_log": "/home/yuxinr/yuanzhe/RobustLLMAgent/results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0.log",
+      "launcher_pid": 3126931,
+      "plot_tmux_session": "ws_stable_v9_s200_seed0_plots",
+      "project_python": "/home/yuxinr/miniforge3/envs/verl-agent/bin/python",
+      "python_no_user_site": true,
+      "ray_override_resources": null,
+      "ray_temp_dir": "/tmp/ray_wsstablev9200",
+      "requested_start_mode": "fresh",
+      "resume_checkpoint": null,
+      "slurm_cpus_per_task": null,
+      "slurm_job_id": null,
+      "slurm_mem_per_node": null,
+      "tmux_session": "ws_stable_v9_s200_seed0",
+      "tmux_socket": null,
+      "torch": "2.8.0+cu128",
+      "vllm": "0.11.0",
+      "vllm_enable_prefix_caching": false,
+      "vllm_enforce_eager": false,
+      "vllm_load_format": "dummy_dtensor"
+    },
+    {
+      "architecture": "x86_64",
+      "dataloader_num_workers": 0,
+      "expected_torch": "2.8.0+cu128",
+      "free_cache_engine": false,
+      "git_commit_at_launch": "6b2abd6d16a4bdfc0987712e01a603a9fd50e324",
+      "git_worktree_dirty_at_launch": true,
+      "host": "train8",
+      "launcher_log": "/home/yuxinr/yuanzhe/RobustLLMAgent/results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0.log",
+      "launcher_pid": 3128191,
+      "plot_tmux_session": "ws_stable_v9_s200_seed0_plots",
+      "project_python": "/home/yuxinr/miniforge3/envs/verl-agent/bin/python",
+      "python_no_user_site": true,
+      "ray_override_resources": null,
+      "ray_temp_dir": "/tmp/ray_wsstablev9200",
+      "requested_start_mode": "auto",
+      "resume_checkpoint": null,
+      "slurm_cpus_per_task": null,
+      "slurm_job_id": null,
+      "slurm_mem_per_node": null,
+      "tmux_session": "ws_stable_v9_s200_seed0",
+      "tmux_socket": null,
+      "torch": "2.8.0+cu128",
+      "vllm": "0.11.0",
+      "vllm_enable_prefix_caching": false,
+      "vllm_enforce_eager": false,
+      "vllm_load_format": "dummy_dtensor"
+    },
+    {
+      "architecture": "x86_64",
+      "dataloader_num_workers": 0,
+      "expected_torch": "2.8.0+cu128",
+      "free_cache_engine": false,
+      "git_commit_at_launch": "6b2abd6d16a4bdfc0987712e01a603a9fd50e324",
+      "git_worktree_dirty_at_launch": true,
+      "host": "train8",
+      "launcher_log": "/home/yuxinr/yuanzhe/RobustLLMAgent/results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0.log",
+      "launcher_pid": 3145217,
+      "plot_tmux_session": "ws_stable_v9_s200_seed0_plots",
+      "project_python": "/home/yuxinr/miniforge3/envs/verl-agent/bin/python",
+      "python_no_user_site": true,
+      "ray_override_resources": null,
+      "ray_temp_dir": "/tmp/ray_wsstablev9200",
+      "requested_start_mode": "auto",
+      "resume_checkpoint": null,
+      "slurm_cpus_per_task": null,
+      "slurm_job_id": null,
+      "slurm_mem_per_node": null,
+      "tmux_session": "ws_stable_v9_s200_seed0",
+      "tmux_socket": null,
+      "torch": "2.8.0+cu128",
+      "vllm": "0.11.0",
+      "vllm_enable_prefix_caching": false,
+      "vllm_enforce_eager": false,
+      "vllm_load_format": "dummy_dtensor"
+    },
+    {
+      "architecture": "x86_64",
+      "dataloader_num_workers": 0,
+      "expected_torch": "2.8.0+cu128",
+      "free_cache_engine": false,
+      "git_commit_at_launch": "6b2abd6d16a4bdfc0987712e01a603a9fd50e324",
+      "git_worktree_dirty_at_launch": true,
+      "host": "train8",
+      "launcher_log": "/home/yuxinr/yuanzhe/RobustLLMAgent/results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0.log",
+      "launcher_pid": 3154938,
+      "plot_tmux_session": "ws_stable_v9_s200_seed0_plots",
+      "project_python": "/home/yuxinr/miniforge3/envs/verl-agent/bin/python",
+      "python_no_user_site": true,
+      "ray_override_resources": null,
+      "ray_temp_dir": "/tmp/ray_wsstablev9200",
+      "requested_start_mode": "auto",
+      "resume_checkpoint": null,
+      "slurm_cpus_per_task": null,
+      "slurm_job_id": null,
+      "slurm_mem_per_node": null,
+      "tmux_session": "ws_stable_v9_s200_seed0",
+      "tmux_socket": null,
+      "torch": "2.8.0+cu128",
+      "vllm": "0.11.0",
+      "vllm_enable_prefix_caching": false,
+      "vllm_enforce_eager": false,
+      "vllm_load_format": "dummy_dtensor"
+    }
+  ],
+  "runtime": {
+    "architecture": "x86_64",
+    "dataloader_num_workers": 0,
+    "expected_torch": "2.8.0+cu128",
+    "free_cache_engine": false,
+    "git_commit_at_launch": "6b2abd6d16a4bdfc0987712e01a603a9fd50e324",
+    "git_worktree_dirty_at_launch": true,
+    "host": "train8",
+    "launcher_log": "/home/yuxinr/yuanzhe/RobustLLMAgent/results/webshop/robust_training/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0/logs/qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0.log",
+    "launcher_pid": 3278823,
+    "plot_tmux_session": "ws_stable_v9_s200_seed0_plots",
+    "project_python": "/home/yuxinr/miniforge3/envs/verl-agent/bin/python",
+    "python_no_user_site": true,
+    "ray_override_resources": null,
+    "ray_temp_dir": "/tmp/ray_wsstablev9200",
+    "requested_start_mode": "auto",
+    "resume_checkpoint": null,
+    "slurm_cpus_per_task": null,
+    "slurm_job_id": null,
+    "slurm_mem_per_node": null,
+    "tmux_session": "ws_stable_v9_s200_seed0",
+    "tmux_socket": null,
+    "torch": "2.8.0+cu128",
+    "vllm": "0.11.0",
+    "vllm_enable_prefix_caching": false,
+    "vllm_enforce_eager": false,
+    "vllm_load_format": "dummy_dtensor"
+  },
+  "source_checkpoint": null,
+  "status": "active",
+  "training": {
+    "actor_micro_batch_size_per_gpu": 4,
+    "algorithm": "GiGPO",
+    "cleanup_non_interval_checkpoints_on_completion": true,
+    "clip_ratio_high": 0.2,
+    "clip_ratio_low": 0.2,
+    "completed_checkpoint_keep_interval": 100,
+    "cuda_visible_devices": "0,1",
+    "entropy_coeff": 0.001,
+    "gamma": 0.95,
+    "history_length": 2,
+    "invalid_action_penalty_coef": 0.1,
+    "learning_rate": 3e-06,
+    "lora_alpha": 64,
+    "lora_rank": 64,
+    "max_environment_steps": 15,
+    "max_prompt_length": 4096,
+    "max_response_length": 512,
+    "n_gpus_per_node": 2,
+    "old_log_probs_source": "noisy_vllm_rollout_with_same_channel_scale_as_PPO",
+    "optimizer_grad_clip": 1.0,
+    "ppo_epochs": 1,
+    "ppo_mini_batch_size": 128,
+    "resume_mode": "auto",
+    "robust_update_count": 200,
+    "rollout_gpu_memory_utilization": 0.3,
+    "rollout_n": 8,
+    "rollout_temperature": 1.0,
+    "rollout_tensor_model_parallel_size": 1,
+    "rollout_top_p": 1.0,
+    "save_frequency_steps": 5,
+    "seed": 0,
+    "start_mode": "fresh",
+    "target_global_step": 200,
+    "thinking_mode": "off",
+    "thinking_mode_launcher": "nothinking",
+    "train_batch_size": 16,
+    "train_sessions_per_worker": 8,
+    "use_kl_loss": false,
+    "val_sessions_per_worker": 8
+  },
+  "validation": {
+    "channel_scale": "explicit_all_ones_then_restore_training_scale",
+    "do_sample": true,
+    "frequency_steps": 20,
+    "official_evaluation_goals": "0..499 (separate post-training evaluation)",
+    "perturbation_enabled": false,
+    "sampled_goal_count": 128,
+    "step_0_enabled": true,
+    "temperature": 0.4
+  }
+}