Publish S1: M5 GDN S sweep, completed epoch 1 (step 3852), verified serving weights

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+S1/tokenizer.json filter=lfs diff=lfs merge=lfs -text

S1/README.md

@@ -0,0 +1,39 @@
+# S1 — M5 GDN-only S sweep, completed epoch 1
+
+Memory priority: **mean over heads of the unscaled delta-rule residual ‖e‖, e = v − S̄ᵀk (token)**.
+Score `S1`, unit `token`, head reduction `mean`,
+backend `fp32_torch_unscaled_residual_block64_v1`. Everything else is M5 (`episodic_window_v1` at R4/B1200):
+4 recent observation groups including the current one plus a
+1200-token episodic budget, native K/V rows kept at their original positions,
+canonical text action history and full-episode BPTT.
+
+This folder is the arm's serving model; `provenance/` holds the evidence and `SHA256SUMS.json` the published hashes.
+
+## Training (matched to M5_E1)
+
+The recipe is M5_E1's apart from the memory priority and two loader-only settings (asserted before launch):
+original Qwen3.5-4B revision `851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a`, joint R2R + English RxR15deg
+(30,815 episodes), one epoch of 3,852 optimizer updates, global batch 8 (4 H100 x 1 episode x accumulation 2),
+seed 429, backbone LR 5e-6 with 116 warmup updates and cosine decay to 0.1, Adam 0.9/0.95, weight decay 0.01,
+clip 1, ZeRO-2 with CPU Adam. Every rank saw exactly M5_E1's episodes in M5_E1's order.
+
+| | S1 | M5_E1 |
+|---|---:|---:|
+| Action-weighted epoch loss | 0.12647 | 0.12658 |
+| Final update loss | 0.07020 | |
+
+Training loss is not navigation success. Wall time 14.09 h including load and saves; no GPU or host OOM.
+
+## Verification
+
+The lane controller declared the run complete; the bound identity re-derives from the sealed source; all
+3,852 updates and exposures are covered on every rank with finite losses; checkpoint-3852 is a complete
+epoch-boundary recovery checkpoint. Serving tensors are bitwise identical to it and the real navigation model
+strictly reloads on CPU with exact serializer and memory metadata. No completed-model GPU replay is claimed.
+
+## Use
+
+Code: SimpleMemVLN branch `m5_gdn_s_sweep`, source `dabf4ad085d20512640fe0ba3bda72ad4724dc76` (memory mode
+`episodic_window_gdn_sweep_v1`). Load with
+`qwen_vl.train.vln_runtime.load_checkpoint('S1', base_model_path)` after downloading this folder.
+These are navigation-wrapper weights for inference; optimizer/RNG state stays on the cluster.

S1/SHA256SUMS.json

@@ -0,0 +1,82 @@
+{
+  "S1/README.md": {
+    "sha256": "7698b305c5a560d1f2aa4dcd129df1cb7ce2a3b9ccf3ea97267686e27aa211f2",
+    "size": 2200
+  },
+  "S1/chat_template.jinja": {
+    "sha256": "a4aee8afcf2e0711942cf848899be66016f8d14a889ff9ede07bca099c28f715",
+    "size": 7756
+  },
+  "S1/config.json": {
+    "sha256": "2451a8786918f4c348acb32c438d9a127098c022334ae1a7713ce5be588a7df1",
+    "size": 2903
+  },
+  "S1/navigation.json": {
+    "sha256": "7345f6b2618b4ded781d839dacfc9ca1e038629c6340ede5b735b431fb1cd4b3",
+    "size": 5040
+  },
+  "S1/processor_config.json": {
+    "sha256": "d89ef49ce9cd37fbf510158e13c1ef063d9286411c1ec9049932dbe0487143b1",
+    "size": 1191
+  },
+  "S1/provenance/arm_S1.yaml": {
+    "sha256": "aab3a2c6bba624a5172fa716a1ccabeb7d0509576c122cc8468a0a7ddcac1c7e",
+    "size": 747
+  },
+  "S1/provenance/completion_resources.json": {
+    "sha256": "5ae199582a5e945385f2eb3c8d5455c7a208c86e4f670fcc04e220e600c46a38",
+    "size": 1982
+  },
+  "S1/provenance/epoch-1.json": {
+    "sha256": "8975908fd4ed2a8e95026c956d9cf000d756697cacfbf985992c8e34450da428",
+    "size": 396567
+  },
+  "S1/provenance/publication_validation.json": {
+    "sha256": "1aeb1c77be35df4702593e2819220235893eec3ad848840455038be3c8315b57",
+    "size": 290
+  },
+  "S1/provenance/qualification_pass.json": {
+    "sha256": "9469b364e3e652b9b2daa0d1ad47f710747f725d30112fcd8b93ba477fbd0522",
+    "size": 26644
+  },
+  "S1/provenance/resolved_config.json": {
+    "sha256": "9245a9be2ec5690f3faa54baa64e093fbd51fa6576ad3c3a9ca7e6728e90e7ad",
+    "size": 2787
+  },
+  "S1/provenance/run_identity.json": {
+    "sha256": "e2dbe48780a1ec242f3cd4aa8bb93bf70b4582a4987b13c5cf07c9d8679d7f93",
+    "size": 3599
+  },
+  "S1/provenance/run_timing.json": {
+    "sha256": "28433f3fb0afd3a7e4963ffcd63d9f17ed934e6ff8da5da424f3ce2fa093fc6a",
+    "size": 353
+  },
+  "S1/provenance/runtime.yaml": {
+    "sha256": "abc762cc00a8ada3f7d7c21dfafb5bee4e72474acaa272d0c0fd4ab2793cd101",
+    "size": 647
+  },
+  "S1/provenance/source_and_completion.json": {
+    "sha256": "ee4e9d75139062669b3597e799b1eded389c1b4c83a49971ac6c136a4e175da9",
+    "size": 1971
+  },
+  "S1/provenance/trainer_state.json": {
+    "sha256": "5e4e7ee00ddeb30869aa748af096388a7a09d0057fc9ab3fc51356834437978e",
+    "size": 719101
+  },
+  "S1/provenance/training_validation.json": {
+    "sha256": "c6e9153ba45c783e42fa8d52765e1831b95b8b9c0ee33a414eee91fd7bcacf23",
+    "size": 158
+  },
+  "S1/pytorch_model.bin": {
+    "sha256": "dd5876f50647fac8bfeb09bb4c7f6c8590dc796aa96db96642ec4b7af4e228df",
+    "size": 10350179723
+  },
+  "S1/tokenizer.json": {
+    "sha256": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523",
+    "size": 19989325
+  },
+  "S1/tokenizer_config.json": {
+    "sha256": "7488550a894dcf741616126e322ea31d66cf9457b59ec5825f4c7efdced421f8",
+    "size": 1164
+  }
+}

S1/chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}

S1/config.json

@@ -0,0 +1,112 @@
+{
+  "architectures": [
+    "Qwen3_5ForConditionalGeneration"
+  ],
+  "dtype": "bfloat16",
+  "eos_token_id": 248046,
+  "image_token_id": 248056,
+  "model_type": "qwen3_5",
+  "pad_token_id": 248044,
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attn_output_gate": true,
+    "bos_token_id": null,
+    "dtype": "bfloat16",
+    "eos_token_id": 248044,
+    "full_attention_interval": 4,
+    "head_dim": 256,
+    "hidden_act": "silu",
+    "hidden_size": 2560,
+    "initializer_range": 0.02,
+    "intermediate_size": 9216,
+    "layer_types": [
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention"
+    ],
+    "linear_conv_kernel_dim": 4,
+    "linear_key_head_dim": 128,
+    "linear_num_key_heads": 16,
+    "linear_num_value_heads": 32,
+    "linear_value_head_dim": 128,
+    "mamba_ssm_dtype": "float32",
+    "max_position_embeddings": 262144,
+    "mlp_only_layers": [],
+    "model_type": "qwen3_5_text",
+    "mtp_num_hidden_layers": 1,
+    "mtp_use_dedicated_embeddings": false,
+    "num_attention_heads": 16,
+    "num_hidden_layers": 32,
+    "num_key_value_heads": 4,
+    "pad_token_id": null,
+    "partial_rotary_factor": 0.25,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "mrope_interleaved": true,
+      "mrope_section": [
+        11,
+        11,
+        10
+      ],
+      "partial_rotary_factor": 0.25,
+      "rope_theta": 10000000,
+      "rope_type": "default"
+    },
+    "tie_word_embeddings": true,
+    "use_cache": true,
+    "vocab_size": 248320
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.11.0",
+  "use_cache": false,
+  "video_token_id": 248057,
+  "vision_config": {
+    "deepstack_visual_indexes": [],
+    "depth": 24,
+    "dtype": "bfloat16",
+    "hidden_act": "gelu_pytorch_tanh",
+    "hidden_size": 1024,
+    "in_channels": 3,
+    "initializer_range": 0.02,
+    "intermediate_size": 4096,
+    "model_type": "qwen3_5_vision",
+    "num_heads": 16,
+    "num_position_embeddings": 2304,
+    "out_hidden_size": 2560,
+    "patch_size": 16,
+    "spatial_merge_size": 2,
+    "temporal_patch_size": 2
+  },
+  "vision_end_token_id": 248054,
+  "vision_start_token_id": 248053
+}

S1/navigation.json

@@ -0,0 +1,198 @@
+{
+  "serializer": "vln_append_only_chat_v1",
+  "template_sha256": "fb79cf3922f74475e9a5faaac8c08d2f79a3ca7bef5dbf434f3f7726119229b8",
+  "actions": [
+    "MOVE_FORWARD",
+    "TURN_LEFT",
+    "TURN_RIGHT",
+    "STOP"
+  ],
+  "habitat_ids": [
+    1,
+    2,
+    3,
+    0
+  ],
+  "action_token_ids": [
+    [
+      28164,
+      61846,
+      248046
+    ],
+    [
+      15834,
+      18440,
+      248046
+    ],
+    [
+      15834,
+      20549,
+      248046
+    ],
+    [
+      48964,
+      248046
+    ]
+  ],
+  "eos": 248046,
+  "separator_ids": [
+    198
+  ],
+  "rgb_shape": [
+    480,
+    640,
+    3
+  ],
+  "position_convention": "logical_text_plus_native_mrope_v1",
+  "rotary_precision": "fp32_preserved_across_module_casts_v1",
+  "config": {
+    "model": {
+      "backbone": "Qwen/Qwen3.5-4B",
+      "revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
+      "output_mode": "qwen_text"
+    },
+    "observations": {
+      "serializer_version": "vln_append_only_chat_v1",
+      "append_action_tokens": true,
+      "expected_rgb_shape": [
+        480,
+        640,
+        3
+      ],
+      "expected_visual_tokens": 300,
+      "max_prefix_tokens": 2048,
+      "max_step_group_tokens": 384
+    },
+    "generation": {
+      "max_response_tokens_including_eos": 16,
+      "fixed_separator": "\n"
+    },
+    "training": {
+      "use_cache": false,
+      "microbatch_episodes_per_rank": 1,
+      "nominal_episodes_per_update": 8,
+      "model_max_length": 262144,
+      "epochs": 1,
+      "backbone_lr": 5e-06,
+      "classifier_lr": 5e-06,
+      "weight_decay": 0.01,
+      "warmup_steps": 116,
+      "expected_total_steps": 3852,
+      "expected_warmup_steps": 116,
+      "seed": 429,
+      "checkpoint_decoder_layers": true,
+      "activation_offload_min_tokens": 65536,
+      "length_bucket_pool_episodes": 64,
+      "tail_policy": "accelerate_repeat",
+      "supervise_assistant_terminator": true,
+      "class_weighting": "none",
+      "save_strategy": "epoch",
+      "save_total_limit": null,
+      "activation_offload_backend": "pinned_arena_v1",
+      "cpu_optimizer_impl": "deepspeed_cpu_adam",
+      "dataloader_num_workers": 4,
+      "dataloader_persistent_workers": true,
+      "dataloader_pin_memory": false,
+      "dataloader_prefetch_factor": 1,
+      "gradient_accumulation_steps": 2,
+      "optimizer_group_chunks": 1
+    },
+    "runtime": {
+      "text_attention": "simplememvln_step_attention",
+      "vision_attention": "flash_attention_2",
+      "vision_microbatch_images": 8,
+      "max_logical_context_tokens": 262144,
+      "window_attention_batch_steps": 32,
+      "profile_components": false,
+      "action_diagnostics": false,
+      "episodic_diagnostics": false,
+      "malloc_trim_each_update": false,
+      "stage_pixels_bf16": true
+    },
+    "memory": {
+      "mode": "episodic_window_gdn_sweep_v1",
+      "recent_groups_including_current": 4,
+      "episodic_budget_tokens": 1200,
+      "selection_unit": "token",
+      "score_id": "S1",
+      "head_reduction": "mean",
+      "scorer_backend": "fp32_torch_unscaled_residual_block64_v1",
+      "score_segmentation": "prefix_obs_response_block64_v1",
+      "selector_seed": 429,
+      "evict_before_new_step": true,
+      "kv_guard_tokens": 4784,
+      "priority_timing": "birth_fixed",
+      "score_dtype": "float32",
+      "key_normalization": "native_additive_eps_1e-6",
+      "cosine_epsilon": 1e-06,
+      "zero_value_threshold": 1e-06,
+      "state_layout": "value_heads_key_by_value",
+      "attention_backend": "episodic_fa2_varlen_gather_fp32acc_v1",
+      "logical_position_policy": "preserve",
+      "candidate_policy": "resident_plus_newly_eligible",
+      "tie_break": "priority_desc_original_id_asc"
+    }
+  },
+  "episodic": {
+    "version": "episodic_window_gdn_sweep_v1",
+    "recent_groups": 4,
+    "budget_tokens": 1200,
+    "granularity": "token",
+    "score": "S1",
+    "scoring_pairs": [
+      [
+        2,
+        3
+      ],
+      [
+        6,
+        7
+      ],
+      [
+        10,
+        11
+      ],
+      [
+        14,
+        15
+      ],
+      [
+        18,
+        19
+      ],
+      [
+        22,
+        23
+      ],
+      [
+        26,
+        27
+      ],
+      [
+        30,
+        31
+      ]
+    ],
+    "selector_seed": 429,
+    "per_source_cap": null,
+    "score_backend": "fp32_torch_unscaled_residual_block64_v1",
+    "score_segmentation": "prefix_obs_response_block64_v1",
+    "attention_backend": "episodic_fa2_varlen_gather_fp32acc_v1",
+    "head_reduction": "mean",
+    "statistic": "error_norm",
+    "priority_timing": "birth_fixed",
+    "score_dtype": "float32",
+    "key_normalization": "native_additive_eps_1e-6",
+    "cosine_epsilon": 1e-06,
+    "zero_value_threshold": 1e-06,
+    "state_layout": "value_heads_key_by_value",
+    "logical_position_policy": "preserve",
+    "candidate_policy": "resident_plus_newly_eligible",
+    "tie_break": "priority_desc_original_id_asc",
+    "slow_head_mask": null,
+    "slow_head_mask_sha256": null,
+    "drift_floor": null,
+    "drift_floor_sha256": null,
+    "calibration_sha256": null
+  }
+}

S1/processor_config.json

@@ -0,0 +1,60 @@
+{
+  "image_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "Qwen2VLImageProcessor",
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "merge_size": 2,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "longest_edge": 16777216,
+      "shortest_edge": 65536
+    },
+    "temporal_patch_size": 2
+  },
+  "processor_class": "Qwen3VLProcessor",
+  "video_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "do_sample_frames": true,
+    "fps": 2,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "max_frames": 768,
+    "merge_size": 2,
+    "min_frames": 4,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "return_metadata": false,
+    "size": {
+      "longest_edge": 25165824,
+      "shortest_edge": 4096
+    },
+    "temporal_patch_size": 2,
+    "video_processor_type": "Qwen3VLVideoProcessor"
+  }
+}

S1/provenance/arm_S1.yaml

@@ -0,0 +1,22 @@
+memory:
+  mode: episodic_window_gdn_sweep_v1
+  recent_groups_including_current: 4
+  episodic_budget_tokens: 1200
+  selection_unit: token
+  score_id: S1
+  head_reduction: mean
+  scorer_backend: fp32_torch_unscaled_residual_block64_v1
+  score_segmentation: prefix_obs_response_block64_v1
+  selector_seed: 429
+  evict_before_new_step: true
+  kv_guard_tokens: 4784
+  priority_timing: birth_fixed
+  score_dtype: float32
+  key_normalization: native_additive_eps_1e-6
+  cosine_epsilon: 1.0e-06
+  zero_value_threshold: 1.0e-06
+  state_layout: value_heads_key_by_value
+  attention_backend: episodic_fa2_varlen_gather_fp32acc_v1
+  logical_position_policy: preserve
+  candidate_policy: resident_plus_newly_eligible
+  tie_break: priority_desc_original_id_asc

S1/provenance/completion_resources.json

@@ -0,0 +1,58 @@
+{
+  "status": "PASS",
+  "exit_code": 0,
+  "host_guard_triggered": false,
+  "host_guard_gib": 432.0,
+  "job_limit_gib": 480.0,
+  "sampled_peak_job_usage_gib": 480.0001220703125,
+  "sampled_peak_working_set_gib": 361.42921447753906,
+  "sampled_peak_anon_gib": 240.22570037841797,
+  "cgroup_recorded_peak_gib": 198.49213790893555,
+  "new_oom_events": 0,
+  "new_oom_kill_events": 0,
+  "new_limit_events": 19134302,
+  "start_usage_gib": 169.14991760253906,
+  "nvidia_smi_peak_used_mib": {
+    "0": 72455,
+    "1": 68295,
+    "2": 65093,
+    "3": 68093
+  },
+  "seconds": 50731.11613750458,
+  "command": [
+    "torchrun",
+    "--standalone",
+    "--nproc_per_node=4",
+    "-m",
+    "qwen_vl.train.train_qwen",
+    "--vln_config",
+    "configs/episodic/science_common.yaml",
+    "--output_config",
+    "configs/vln_empty.yaml",
+    "--memory_config",
+    "/mnt/data/vmo-ai-task/anhdh35/m5-gdn-s-sweep-dabf4ad-20261020/configs/arm_S1.yaml",
+    "--policy_config",
+    "configs/vggt/g3_joint_one_epoch.yaml",
+    "--runtime_profile",
+    "/mnt/data/vmo-ai-task/anhdh35/m5-gdn-s-sweep-dabf4ad-20261020/configs/runtime.yaml",
+    "--deepspeed",
+    "configs/deepspeed_zero2_cpu_optimizer.json",
+    "--manifest",
+    "/mnt/data/vmo-ai-task/anhdh35/SimpleMemVLN/artifacts/r2r_rxr15_train_20260929.jsonl",
+    "--model_name_or_path",
+    "/mnt/data/vmo-ai-task/anhdh35/cache/huggingface/hub/models--Qwen--Qwen3.5-4B/snapshots/851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
+    "--output_dir",
+    "/mnt/data/vmo-ai-task/anhdh35/m5-gdn-s-sweep-dabf4ad-20261020/runs/S1_seed429_e1/train",
+    "--run-name",
+    "s1_seed429_e1",
+    "--recovery-save-steps",
+    "100",
+    "--first-recovery-save-step",
+    "10",
+    "--resume_from_checkpoint",
+    "auto"
+  ],
+  "cgroup": "/sys/fs/cgroup/kubepods.slice/kubepods-burstable.slice/kubepods-burstable-podf16b1000_b339_46e0_b01c_db5bc40a1e72.slice/system.slice/slurmstepd.scope/job_5248",
+  "cgroup_version": 2,
+  "free_disk_gib_at_start": 2717.384765625
+}

S1/provenance/publication_validation.json

@@ -0,0 +1,11 @@
+{
+  "all_model_tensors_finite": true,
+  "completed_model_gpu_replay": false,
+  "cpu_only": true,
+  "exact_serving_tensors": 724,
+  "memory_metadata_exact": true,
+  "passed": true,
+  "removed_training_tensors": [],
+  "serializer_metadata_exact": true,
+  "strict_architecture_reload": true
+}

S1/provenance/qualification_pass.json

@@ -0,0 +1,695 @@
+{
+  "calibration_sha256": "7aafb0e25f35d453d0e4553939c9a29f82ed40fec59b0291cb6116b59c79792d",
+  "gates": {
+    "calibration": {
+      "passed": true,
+      "result": {
+        "calibration_sha256": "7aafb0e25f35d453d0e4553939c9a29f82ed40fec59b0291cb6116b59c79792d",
+        "configs": {
+          "S6": "331a3c28246acb4ed2aaadb05139a44ad43db604d90e7ecca2dca6a563a46ec9",
+          "S8N": "0a11379bebae99407c671a88536001c7f1fdde9cff9f97c007f2645e97234d42",
+          "S8W": "90a4f012fc97c4f5e4b736b1b11da5e1d84b050e93125ccd2aaa0829c4a5f0ca"
+        }
+      },
+      "utc": "2026-10-20T21:08:41Z"
+    },
+    "cpu-suite": {
+      "passed": true,
+      "result": "pass",
+      "utc": "2026-10-20T21:04:03Z"
+    },
+    "provenance": {
+      "passed": true,
+      "result": {
+        "base_model_revision": true,
+        "baseline_ancestor": true,
+        "fixture_images_present": true,
+        "images_present_sample": true,
+        "m5_e1_is_plain_m5": true,
+        "m5_e1_matches_publication": true,
+        "m5_e1_weights": true,
+        "manifest": true,
+        "manifest_counts": true,
+        "native_forward": true,
+        "source_commit": true,
+        "transformers_pinned": true
+      },
+      "utc": "2026-10-20T21:04:29Z"
+    },
+    "recovery": {
+      "passed": true,
+      "result": {
+        "checkpoints": {
+          "/mnt/data/vmo-ai-task/anhdh35/m5-gdn-s-sweep-dabf4ad-20261020/shared/recovery/reference/train/checkpoint-2": {
+            "files": 22,
+            "step": 2
+          },
+          "/mnt/data/vmo-ai-task/anhdh35/m5-gdn-s-sweep-dabf4ad-20261020/shared/recovery/resumed/train/checkpoint-3": {
+            "files": 22,
+            "step": 3
+          }
+        },
+        "losses": [
+          0.4781087040901184,
+          0.4781087040901184
+        ],
+        "passed": true,
+        "weights": {
+          "bitwise_equal": false,
+          "changed_elements": 57105,
+          "elements": 5174964736,
+          "max_abs": 2.384185791015625e-07,
+          "relative_l2": 2.5299356095180488e-08
+        }
+      },
+      "utc": "2026-10-20T21:18:05Z"
+    },
+    "serving-replay": {
+      "passed": true,
+      "result": {
+        "records": {
+          "C_RANDOM/r2rce": {
+            "action_nll": 0.1149396151304245,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.625,
+            "max_relative_score_error": null,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 1.0,
+            "min_selected_overlap": 1.0,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.17261256479075685,
+              "mean_overlap": 0.2416336805555556,
+              "min_jaccard": 0.029601029601029602,
+              "min_overlap": 0.0575,
+              "p10_overlap": 0.0775
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 0,
+            "rms_logit_error": 0.04785158857703209
+          },
+          "C_RANDOM/rxrce": {
+            "action_nll": 0.3265545964241028,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.423828125,
+            "max_relative_score_error": null,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 1.0,
+            "min_selected_overlap": 1.0,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.17126053905958816,
+              "mean_overlap": 0.23905555555555555,
+              "min_jaccard": 0.026518391787852865,
+              "min_overlap": 0.051666666666666666,
+              "p10_overlap": 0.07241666666666667
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 0,
+            "rms_logit_error": 0.03253242000937462
+          },
+          "C_RECENCY/r2rce": {
+            "action_nll": 0.1217651516199112,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.5625,
+            "max_relative_score_error": null,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 1.0,
+            "min_selected_overlap": 1.0,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.17122633829691541,
+              "mean_overlap": 0.23387499999999997,
+              "min_jaccard": 0.005025125628140704,
+              "min_overlap": 0.01,
+              "p10_overlap": 0.051666666666666666
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 0,
+            "rms_logit_error": 0.047868385910987854
+          },
+          "C_RECENCY/rxrce": {
+            "action_nll": 0.31977471709251404,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.5,
+            "max_relative_score_error": null,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 1.0,
+            "min_selected_overlap": 1.0,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.1624957751944985,
+              "mean_overlap": 0.22520659722222222,
+              "min_jaccard": 0.0046044370029300966,
+              "min_overlap": 0.009166666666666667,
+              "p10_overlap": 0.04575
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 0,
+            "rms_logit_error": 0.03140942007303238
+          },
+          "S0/r2rce": {
+            "action_nll": 0.11342943459749222,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.7265625,
+            "max_relative_score_error": 0.09263685345649719,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9924583333333333,
+            "min_selected_overlap": 0.9791666666666666,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 1.0,
+              "mean_overlap": 1.0,
+              "min_jaccard": 1.0,
+              "min_overlap": 1.0,
+              "p10_overlap": 1.0
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.05344880744814873
+          },
+          "S0/rxrce": {
+            "action_nll": 0.30024611949920654,
+            "argmax_agreement": 0.9947643876075745,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 1.9775390625,
+            "max_relative_score_error": 0.17983831465244293,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9908906249999999,
+            "min_selected_overlap": 0.9733333333333334,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 1.0,
+              "mean_overlap": 1.0,
+              "min_jaccard": 1.0,
+              "min_overlap": 1.0,
+              "p10_overlap": 1.0
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.04369400069117546
+          },
+          "S0_legacy/r2rce": {
+            "action_nll": 0.11342943459749222,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.7265625,
+            "max_relative_score_error": 0.09263685345649719,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9924583333333333,
+            "min_selected_overlap": 0.9791666666666666,
+            "overlap_vs_S0_legacy": null,
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.05344880744814873
+          },
+          "S0_legacy/rxrce": {
+            "action_nll": 0.30024611949920654,
+            "argmax_agreement": 0.9947643876075745,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 1.9775390625,
+            "max_relative_score_error": 0.17983831465244293,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9908906249999999,
+            "min_selected_overlap": 0.9733333333333334,
+            "overlap_vs_S0_legacy": null,
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.04369400069117546
+          },
+          "S1/r2rce": {
+            "action_nll": 0.1120259165763855,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 1.1484375,
+            "max_relative_score_error": 0.1060464084148407,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9846649305555556,
+            "min_selected_overlap": 0.9633333333333334,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.3866098169453221,
+              "mean_overlap": 0.52659375,
+              "min_jaccard": 0.15886045388701112,
+              "min_overlap": 0.27416666666666667,
+              "p10_overlap": 0.324
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 447,
+            "rms_logit_error": 0.06233717128634453
+          },
+          "S1/rxrce": {
+            "action_nll": 0.32432419061660767,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.767578125,
+            "max_relative_score_error": 0.05822468921542168,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9859722222222221,
+            "min_selected_overlap": 0.9683333333333334,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.3777660367637328,
+              "mean_overlap": 0.5164253472222222,
+              "min_jaccard": 0.11265646731571627,
+              "min_overlap": 0.2025,
+              "p10_overlap": 0.31483333333333335
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.03846774250268936
+          },
+          "S2/r2rce": {
+            "action_nll": 0.11628837883472443,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 2.544921875,
+            "max_relative_score_error": 0.09035920351743698,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9929409722222222,
+            "min_selected_overlap": 0.9808333333333333,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.7660236294968146,
+              "mean_overlap": 0.8612951388888888,
+              "min_jaccard": 0.48606811145510836,
+              "min_overlap": 0.6541666666666667,
+              "p10_overlap": 0.7475
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 447,
+            "rms_logit_error": 0.06610462814569473
+          },
+          "S2/rxrce": {
+            "action_nll": 0.3019610643386841,
+            "argmax_agreement": 0.9895288348197937,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.5,
+            "max_relative_score_error": 0.1436961144208908,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9910347222222222,
+            "min_selected_overlap": 0.9708333333333333,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.7076658749036285,
+              "mean_overlap": 0.8211145833333333,
+              "min_jaccard": 0.4625228519195612,
+              "min_overlap": 0.6325,
+              "p10_overlap": 0.674
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.03577248379588127
+          },
+          "S3/r2rce": {
+            "action_nll": 0.11478417366743088,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 1.0,
+            "max_relative_score_error": 0.07442871481180191,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9874444444444443,
+            "min_selected_overlap": 0.97,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.24667344633986443,
+              "mean_overlap": 0.3432795138888889,
+              "min_jaccard": 0.010101010101010102,
+              "min_overlap": 0.02,
+              "p10_overlap": 0.048083333333333346
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.055965375155210495
+          },
+          "S3/rxrce": {
+            "action_nll": 0.3225160539150238,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.796875,
+            "max_relative_score_error": 0.07373587042093277,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9891371527777778,
+            "min_selected_overlap": 0.9708333333333333,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.31110058721781103,
+              "mean_overlap": 0.43771701388888895,
+              "min_jaccard": 0.059602649006622516,
+              "min_overlap": 0.1125,
+              "p10_overlap": 0.22416666666666665
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.04278301075100899
+          },
+          "S4/r2rce": {
+            "action_nll": 0.10958902537822723,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 3.265625,
+            "max_relative_score_error": 0.3189394474029541,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9654670138888889,
+            "min_selected_overlap": 0.9166666666666666,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.2688000117274076,
+              "mean_overlap": 0.3763767361111111,
+              "min_jaccard": 0.03359173126614987,
+              "min_overlap": 0.065,
+              "p10_overlap": 0.10916666666666666
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.0864478126168251
+          },
+          "S4/rxrce": {
+            "action_nll": 0.32661473751068115,
+            "argmax_agreement": 0.9947643876075745,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 2.421875,
+            "max_relative_score_error": 0.29020625352859497,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9707222222222222,
+            "min_selected_overlap": 0.9216666666666666,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.20183128095855002,
+              "mean_overlap": 0.2847152777777778,
+              "min_jaccard": 0.01694915254237288,
+              "min_overlap": 0.03333333333333333,
+              "p10_overlap": 0.07075000000000001
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.06290886551141739
+          },
+          "S5/r2rce": {
+            "action_nll": 0.1149848997592926,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 1.0,
+            "max_relative_score_error": 0.11874917894601822,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9842760416666666,
+            "min_selected_overlap": 0.9591666666666666,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.2585834286963461,
+              "mean_overlap": 0.35568055555555556,
+              "min_jaccard": 0.02564102564102564,
+              "min_overlap": 0.05,
+              "p10_overlap": 0.07325000000000001
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.05545856058597565
+          },
+          "S5/rxrce": {
+            "action_nll": 0.3232060670852661,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.78125,
+            "max_relative_score_error": 0.1603846698999405,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9858003472222222,
+            "min_selected_overlap": 0.9658333333333333,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.18739605051804029,
+              "mean_overlap": 0.2598576388888889,
+              "min_jaccard": 0.007556675062972292,
+              "min_overlap": 0.015,
+              "p10_overlap": 0.044083333333333335
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 447,
+            "rms_logit_error": 0.03974205255508423
+          },
+          "S6/r2rce": {
+            "action_nll": 0.11945445090532303,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.849609375,
+            "max_relative_score_error": 0.15499961376190186,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9919809027777778,
+            "min_selected_overlap": 0.9733333333333334,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.5175073471249387,
+              "mean_overlap": 0.6551319444444444,
+              "min_jaccard": 0.08892921960072596,
+              "min_overlap": 0.16333333333333333,
+              "p10_overlap": 0.33775
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.05567092075943947
+          },
+          "S6/rxrce": {
+            "action_nll": 0.31513679027557373,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.53125,
+            "max_relative_score_error": 0.3851288855075836,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9895694444444445,
+            "min_selected_overlap": 0.97,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.3923391338578999,
+              "mean_overlap": 0.5378124999999999,
+              "min_jaccard": 0.09140518417462483,
+              "min_overlap": 0.1675,
+              "p10_overlap": 0.3790833333333333
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.03548096492886543
+          },
+          "S7/r2rce": {
+            "action_nll": 0.11049564927816391,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.78125,
+            "max_relative_score_error": 0.01943712956192002,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9859375,
+            "min_selected_overlap": 0.75,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.23436715243887823,
+              "mean_overlap": 0.3336163194444444,
+              "min_jaccard": 0.05726872246696035,
+              "min_overlap": 0.10833333333333334,
+              "p10_overlap": 0.14075000000000001
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 27,
+            "rms_logit_error": 0.05867970362305641
+          },
+          "S7/rxrce": {
+            "action_nll": 0.3190687894821167,
+            "argmax_agreement": 0.9947643876075745,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 1.8125,
+            "max_relative_score_error": 0.01257475761680225,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9989583333333333,
+            "min_selected_overlap": 0.75,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.21531463057282782,
+              "mean_overlap": 0.3085381944444444,
+              "min_jaccard": 0.041666666666666664,
+              "min_overlap": 0.08,
+              "p10_overlap": 0.12575
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 2,
+            "rms_logit_error": 0.04470103979110718
+          },
+          "S8N/r2rce": {
+            "action_nll": 0.10524125397205353,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.75,
+            "max_relative_score_error": 0.01041384478690497,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9973958333333334,
+            "min_selected_overlap": 0.75,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.18311920992645483,
+              "mean_overlap": 0.25848958333333333,
+              "min_jaccard": 0.016088060965283656,
+              "min_overlap": 0.03166666666666667,
+              "p10_overlap": 0.07416666666666667
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 5,
+            "rms_logit_error": 0.049579259008169174
+          },
+          "S8N/rxrce": {
+            "action_nll": 0.3092760443687439,
+            "argmax_agreement": 0.9947643876075745,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.4921875,
+            "max_relative_score_error": 0.009368776194855065,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 1.0,
+            "min_selected_overlap": 1.0,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.19135533364457677,
+              "mean_overlap": 0.27209374999999997,
+              "min_jaccard": 0.030927835051546393,
+              "min_overlap": 0.06,
+              "p10_overlap": 0.10583333333333333
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 0,
+            "rms_logit_error": 0.03222713991999626
+          },
+          "S8W/r2rce": {
+            "action_nll": 0.1044732928276062,
+            "argmax_agreement": 0.9947643876075745,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.75,
+            "max_relative_score_error": 0.009059712789100963,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9994791666666667,
+            "min_selected_overlap": 0.75,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.19466206643848255,
+              "mean_overlap": 0.27606770833333333,
+              "min_jaccard": 0.03181427343078246,
+              "min_overlap": 0.06166666666666667,
+              "p10_overlap": 0.09991666666666668
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 1,
+            "rms_logit_error": 0.05015679448843002
+          },
+          "S8W/rxrce": {
+            "action_nll": 0.316342830657959,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.6875,
+            "max_relative_score_error": 0.014894609003178131,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9651041666666667,
+            "min_selected_overlap": 0.75,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.19849231977691748,
+              "mean_overlap": 0.28329861111111115,
+              "min_jaccard": 0.03181427343078246,
+              "min_overlap": 0.06166666666666667,
+              "p10_overlap": 0.1125
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 67,
+            "rms_logit_error": 0.03891213238239288
+          },
+          "S9/r2rce": {
+            "action_nll": 0.10586985945701599,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 0.75,
+            "max_relative_score_error": 0.10653744637966156,
+            "max_resident_kv": 2578,
+            "mean_selected_overlap": 0.9818697916666667,
+            "min_selected_overlap": 0.95,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.1273741370239432,
+              "mean_overlap": 0.1605763888888889,
+              "min_jaccard": 0.0,
+              "min_overlap": 0.0,
+              "p10_overlap": 0.0
+            },
+            "passed": true,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 448,
+            "rms_logit_error": 0.056713271886110306
+          },
+          "S9/rxrce": {
+            "action_nll": 0.32540637254714966,
+            "argmax_agreement": 1.0,
+            "decoded_actions_match": 1.0,
+            "kv_guard": 4784,
+            "max_logit_error": 1.22265625,
+            "max_relative_score_error": 0.10868020355701447,
+            "max_resident_kv": 2643,
+            "mean_selected_overlap": 0.9836302083333334,
+            "min_selected_overlap": 0.9608333333333333,
+            "overlap_vs_S0_legacy": {
+              "mean_jaccard": 0.14643553713511956,
+              "mean_overlap": 0.19402256944444443,
+              "min_jaccard": 0.00041684035014589413,
+              "min_overlap": 0.0008333333333333334,
+              "p10_overlap": 0.0075
+            },
+            "passed": false,
+            "plan_comparisons": 512,
+            "plan_id_mismatches": 447,
+            "rms_logit_error": 0.039100661873817444
+          }
+        },
+        "rms_failed": [],
+        "s0_inherited_allclose_failures": [
+          "rxrce"
+        ],
+        "serving_tolerance_failed": [
+          "S1",
+          "S2",
+          "S4",
+          "S5",
+          "S6",
+          "S7",
+          "S9"
+        ]
+      },
+      "utc": "2026-10-20T21:11:03Z"
+    }
+  },
+  "passed": true,
+  "source_commit": "dabf4ad085d20512640fe0ba3bda72ad4724dc76"
+}

S1/provenance/resolved_config.json

@@ -0,0 +1,88 @@
+{
+  "model": {
+    "backbone": "Qwen/Qwen3.5-4B",
+    "revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
+    "output_mode": "qwen_text"
+  },
+  "observations": {
+    "serializer_version": "vln_append_only_chat_v1",
+    "append_action_tokens": true,
+    "expected_rgb_shape": [
+      480,
+      640,
+      3
+    ],
+    "expected_visual_tokens": 300,
+    "max_prefix_tokens": 2048,
+    "max_step_group_tokens": 384
+  },
+  "generation": {
+    "max_response_tokens_including_eos": 16,
+    "fixed_separator": "\n"
+  },
+  "training": {
+    "use_cache": false,
+    "microbatch_episodes_per_rank": 1,
+    "nominal_episodes_per_update": 8,
+    "model_max_length": 262144,
+    "epochs": 1,
+    "backbone_lr": 5e-06,
+    "classifier_lr": 5e-06,
+    "weight_decay": 0.01,
+    "warmup_steps": 116,
+    "expected_total_steps": 3852,
+    "expected_warmup_steps": 116,
+    "seed": 429,
+    "checkpoint_decoder_layers": true,
+    "activation_offload_min_tokens": 65536,
+    "length_bucket_pool_episodes": 64,
+    "tail_policy": "accelerate_repeat",
+    "supervise_assistant_terminator": true,
+    "class_weighting": "none",
+    "save_strategy": "epoch",
+    "save_total_limit": null,
+    "activation_offload_backend": "pinned_arena_v1",
+    "cpu_optimizer_impl": "deepspeed_cpu_adam",
+    "dataloader_num_workers": 4,
+    "dataloader_persistent_workers": true,
+    "dataloader_pin_memory": false,
+    "dataloader_prefetch_factor": 1,
+    "gradient_accumulation_steps": 2,
+    "optimizer_group_chunks": 1
+  },
+  "runtime": {
+    "text_attention": "simplememvln_step_attention",
+    "vision_attention": "flash_attention_2",
+    "vision_microbatch_images": 8,
+    "max_logical_context_tokens": 262144,
+    "window_attention_batch_steps": 32,
+    "profile_components": false,
+    "action_diagnostics": false,
+    "episodic_diagnostics": false,
+    "malloc_trim_each_update": false,
+    "stage_pixels_bf16": true
+  },
+  "memory": {
+    "mode": "episodic_window_gdn_sweep_v1",
+    "recent_groups_including_current": 4,
+    "episodic_budget_tokens": 1200,
+    "selection_unit": "token",
+    "score_id": "S1",
+    "head_reduction": "mean",
+    "scorer_backend": "fp32_torch_unscaled_residual_block64_v1",
+    "score_segmentation": "prefix_obs_response_block64_v1",
+    "selector_seed": 429,
+    "evict_before_new_step": true,
+    "kv_guard_tokens": 4784,
+    "priority_timing": "birth_fixed",
+    "score_dtype": "float32",
+    "key_normalization": "native_additive_eps_1e-6",
+    "cosine_epsilon": 1e-06,
+    "zero_value_threshold": 1e-06,
+    "state_layout": "value_heads_key_by_value",
+    "attention_backend": "episodic_fa2_varlen_gather_fp32acc_v1",
+    "logical_position_policy": "preserve",
+    "candidate_policy": "resident_plus_newly_eligible",
+    "tie_break": "priority_desc_original_id_asc"
+  }
+}

S1/provenance/run_identity.json

@@ -0,0 +1,98 @@
+{
+  "arm": "S1",
+  "campaign": "m5_gdn_s_sweep_campaign_v1",
+  "config": {
+    "generation": {
+      "fixed_separator": "\n",
+      "max_response_tokens_including_eos": 16
+    },
+    "memory": {
+      "attention_backend": "episodic_fa2_varlen_gather_fp32acc_v1",
+      "candidate_policy": "resident_plus_newly_eligible",
+      "cosine_epsilon": 1e-06,
+      "episodic_budget_tokens": 1200,
+      "evict_before_new_step": true,
+      "head_reduction": "mean",
+      "key_normalization": "native_additive_eps_1e-6",
+      "kv_guard_tokens": 4784,
+      "logical_position_policy": "preserve",
+      "mode": "episodic_window_gdn_sweep_v1",
+      "priority_timing": "birth_fixed",
+      "recent_groups_including_current": 4,
+      "score_dtype": "float32",
+      "score_id": "S1",
+      "score_segmentation": "prefix_obs_response_block64_v1",
+      "scorer_backend": "fp32_torch_unscaled_residual_block64_v1",
+      "selection_unit": "token",
+      "selector_seed": 429,
+      "state_layout": "value_heads_key_by_value",
+      "tie_break": "priority_desc_original_id_asc",
+      "zero_value_threshold": 1e-06
+    },
+    "model": {
+      "backbone": "Qwen/Qwen3.5-4B",
+      "output_mode": "qwen_text",
+      "revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a"
+    },
+    "observations": {
+      "append_action_tokens": true,
+      "expected_rgb_shape": [
+        480,
+        640,
+        3
+      ],
+      "expected_visual_tokens": 300,
+      "max_prefix_tokens": 2048,
+      "max_step_group_tokens": 384,
+      "serializer_version": "vln_append_only_chat_v1"
+    },
+    "runtime": {
+      "action_diagnostics": false,
+      "episodic_diagnostics": false,
+      "malloc_trim_each_update": false,
+      "max_logical_context_tokens": 262144,
+      "profile_components": false,
+      "stage_pixels_bf16": true,
+      "text_attention": "simplememvln_step_attention",
+      "vision_attention": "flash_attention_2",
+      "vision_microbatch_images": 8,
+      "window_attention_batch_steps": 32
+    },
+    "training": {
+      "activation_offload_backend": "pinned_arena_v1",
+      "activation_offload_min_tokens": 65536,
+      "backbone_lr": 5e-06,
+      "checkpoint_decoder_layers": true,
+      "class_weighting": "none",
+      "classifier_lr": 5e-06,
+      "cpu_optimizer_impl": "deepspeed_cpu_adam",
+      "dataloader_num_workers": 4,
+      "dataloader_persistent_workers": true,
+      "dataloader_pin_memory": false,
+      "dataloader_prefetch_factor": 1,
+      "epochs": 1,
+      "expected_total_steps": 3852,
+      "expected_warmup_steps": 116,
+      "gradient_accumulation_steps": 2,
+      "length_bucket_pool_episodes": 64,
+      "microbatch_episodes_per_rank": 1,
+      "model_max_length": 262144,
+      "nominal_episodes_per_update": 8,
+      "optimizer_group_chunks": 1,
+      "save_strategy": "epoch",
+      "save_total_limit": null,
+      "seed": 429,
+      "supervise_assistant_terminator": true,
+      "tail_policy": "accelerate_repeat",
+      "use_cache": false,
+      "warmup_steps": 116,
+      "weight_decay": 0.01
+    }
+  },
+  "manifest_sha256": "5deb425d2594ce96931dd6ce12bd6084b066cd44f04808c1dd3ee2e4b3672933",
+  "memory_sha256": "aab3a2c6bba624a5172fa716a1ccabeb7d0509576c122cc8468a0a7ddcac1c7e",
+  "model": "/mnt/data/vmo-ai-task/anhdh35/cache/huggingface/hub/models--Qwen--Qwen3.5-4B/snapshots/851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
+  "qualification_sha256": "9469b364e3e652b9b2daa0d1ad47f710747f725d30112fcd8b93ba477fbd0522",
+  "runtime_sha256": "abc762cc00a8ada3f7d7c21dfafb5bee4e72474acaa272d0c0fd4ab2793cd101",
+  "source_commit": "dabf4ad085d20512640fe0ba3bda72ad4724dc76"
+}

S1/provenance/run_timing.json

@@ -0,0 +1,11 @@
+{
+  "wall_seconds_including_load_and_save": 50704.07439165702,
+  "trainer_runtime_seconds": 50650.1947,
+  "profile_only": false,
+  "stopped_after_epochs": null,
+  "activation_arena": {
+    "longest_tokens": 199735,
+    "capacity_bytes": 32724615168
+  },
+  "update_timer_note": "Update callbacks exclude loader prefetch and model load/checkpoint save"
+}

S1/provenance/runtime.yaml

@@ -0,0 +1,17 @@
+# M5_E1 runtime (reports/vggt_five/M5_E1/operations/runtime.yaml) with loader 4 workers/prefetch 1:
+# qualified at 480 GiB and no measured loader starvation. Engineering only, never the objective.
+runtime:
+  episodic_diagnostics: false
+  malloc_trim_each_update: false
+  stage_pixels_bf16: true
+  vision_microbatch_images: 8
+  window_attention_batch_steps: 32
+training:
+  activation_offload_backend: pinned_arena_v1
+  cpu_optimizer_impl: deepspeed_cpu_adam
+  dataloader_num_workers: 4
+  dataloader_persistent_workers: true
+  dataloader_pin_memory: false
+  dataloader_prefetch_factor: 1
+  gradient_accumulation_steps: 2
+  optimizer_group_chunks: 1

S1/provenance/source_and_completion.json

@@ -0,0 +1,54 @@
+{
+  "arm": "S1",
+  "checkpoint": "/mnt/data/vmo-ai-task/anhdh35/m5-gdn-s-sweep-dabf4ad-20261020/runs/S1_seed429_e1/train/checkpoint-3852",
+  "completed_epoch": 1,
+  "controller_completion": {
+    "arm": "S1",
+    "lane": "A",
+    "pid": 1337773,
+    "pruned": [
+      "checkpoint-3700",
+      "checkpoint-3800"
+    ],
+    "status": "arm-complete",
+    "step": "5248.21",
+    "unix": 1792582295.9338107,
+    "utc": "2026-10-21T11:31:35Z"
+  },
+  "export_source_weights_sha256": "dd5876f50647fac8bfeb09bb4c7f6c8590dc796aa96db96642ec4b7af4e228df",
+  "export_weights_sha256": "dd5876f50647fac8bfeb09bb4c7f6c8590dc796aa96db96642ec4b7af4e228df",
+  "global_batch": 8,
+  "optimizer_updates": 3852,
+  "recovery_inventory": {
+    "epoch_boundary": true,
+    "files": {
+      "chat_template.jinja": 7756,
+      "config.json": 2903,
+      "global_step3852/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt": 12617266133,
+      "global_step3852/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt": 12617270933,
+      "global_step3852/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt": 12617271061,
+      "global_step3852/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt": 12617270229,
+      "global_step3852/mp_rank_00_model_states.pt": 9745951843,
+      "latest": 15,
+      "navigation.json": 5040,
+      "processor_config.json": 1191,
+      "pytorch_model.bin": 10350179723,
+      "resume_contract.json": 2015950,
+      "rng_state_0.pth": 15493,
+      "rng_state_1.pth": 15493,
+      "rng_state_2.pth": 15493,
+      "rng_state_3.pth": 15493,
+      "scheduler.pt": 1465,
+      "tokenizer.json": 19989325,
+      "tokenizer_config.json": 1164,
+      "trainer_state.json": 719101,
+      "training_args.bin": 6865,
+      "zero_to_fp32.py": 33272
+    },
+    "step": 3852,
+    "world_size": 4
+  },
+  "source_commit": "dabf4ad085d20512640fe0ba3bda72ad4724dc76",
+  "source_directory": "/mnt/data/vmo-ai-task/anhdh35/m5-gdn-s-sweep-source-dabf4ad",
+  "verified_utc": "2026-10-21T12:15:06.219501+00:00"
+}

S1/provenance/training_validation.json

@@ -0,0 +1,7 @@
+{
+  "exposures_identical_to_m5_e1": 30816,
+  "max_process_rss_gib": 59.17542266845703,
+  "passed": true,
+  "peak_reserved_gib": 68.515625,
+  "problems": []
+}

S1/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd5876f50647fac8bfeb09bb4c7f6c8590dc796aa96db96642ec4b7af4e228df
+size 10350179723

S1/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523
+size 19989325

S1/tokenizer_config.json

@@ -0,0 +1,33 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": true,
+  "local_files_only": true,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "processor_class": "Qwen3VLProcessor",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}