Apply fallback prompts, reset benchmarks, and shard large pending inputs

cot_trajectory_generation/glm5.2/MMLU-Pro/SOURCE.md

@@ -4,10 +4,10 @@
 - Source: https://github.com/TIGER-AI-Lab/MMLU-Pro
 - Official API evaluator:
   https://github.com/TIGER-AI-Lab/MMLU-Pro/blob/main/evaluate_from_api.py
-- Official system prompt: none
-- Applied input: one user message containing the official category-specific
-  instruction, all official validation demonstrations for that category, and
-  the complete target question and options.
+- Official system prompt: none; the shared fallback system prompt is used.
+- Applied user input: the official category-specific instruction, all official
+  validation demonstrations for that category, and the complete target
+  question and options.
 - `official_few_shot_validation.parquet` is retained because those rows supply
   the demonstrations; they are not pending evaluation examples.
-
+- All 12,032 test rows are pending; earlier trajectories were discarded.

cot_trajectory_generation/glm5.2/MMLU-Pro/pending.jsonl.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:07d0fa9d20f5ec039c4e1d00195d5c506e306217796393a01faf7cb769c7c610
-size 3971677
+oid sha256:271d5cc5f969399b14a36d92f2b3c6c9109165f4ab646d07fc6d2042b473adb0
+size 4031713

cot_trajectory_generation/glm5.2/Multi-subject-RLVR/SOURCE.md

@@ -6,3 +6,4 @@
   `official_dataset_system_prompt.txt` for provenance.
 - Applied input: the published system message followed by the original user
   question. No additional boxed-answer prompt is added.
+- Pending rows are split across `pending/part-*.jsonl.gz`.

cot_trajectory_generation/glm5.2/{MMLU-Pro/completed.jsonl.gz → Multi-subject-RLVR/pending/part-00000.jsonl.gz}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc22480df383831139df10b7329dd62197b1b7cb3791312f3e39b124ad58bd23
-size 54861
+oid sha256:b9bbc7aafec3fd1bacbebe327e0dae21d1b6db0f869ee5e2d9fde959bce64cbd
+size 5126572

cot_trajectory_generation/glm5.2/{SuperGPQA/completed.jsonl.gz → Multi-subject-RLVR/pending/part-00001.jsonl.gz}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1683e9e647da7fdd9d256114e253f7a071271fbbe0011b615462550417c0f56d
-size 78419
+oid sha256:5a5439c41f184dbb17c9fe45e95a8f5a321ee27c6370614722f891f6f6911dfd
+size 5131848

cot_trajectory_generation/glm5.2/Multi-subject-RLVR/{pending.jsonl.gz → pending/part-00002.jsonl.gz}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a679bd214b6d8d1a87a399a718bf0f3e9d131c9a166f589d67af2288fa6a5108
-size 24905646
+oid sha256:a98cf176a027408a73b92f18db39a3423111a39fed7f2118e00452f6470afafa
+size 5123817

cot_trajectory_generation/glm5.2/{WebInstruct-verified/pending.jsonl.gz → Multi-subject-RLVR/pending/part-00003.jsonl.gz}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d39ecf9d0ab12a3c41dcfa18066d65c6ba41b531d655dc38d0a19b3b2eb632f
-size 23363726
+oid sha256:1cf1e0e54cb08715e44e81d50b15d404a8099b763fbb89788caeb8a06f7eb0b3
+size 5141647

cot_trajectory_generation/glm5.2/Multi-subject-RLVR/pending/part-00004.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4039cfc7d24df268129409e2e678708af49c9eb2d65f7308c8cf394f7d18fc0
+size 4384413

cot_trajectory_generation/glm5.2/README.md

@@ -16,18 +16,21 @@ This directory contains only the seven datasets currently in scope.
 
 ## Input alignment
 
-- Training datasets reproduce their original open-domain SFT query setting.
-- Multi-subject-RLVR retains its published system message.
-- ARC, TriviaQA, and WebInstruct-verified have no source system message in the
-  local source records, so their queries contain only the complete user input.
-- No training dataset is given an additional `\boxed{}` instruction.
+- A dataset's original system prompt is retained when one is published.
+- When no original system prompt exists, `fallback_system_prompt.txt` is used.
+- Multi-subject-RLVR publishes the same text as the selected fallback.
+- No dataset is given an additional `\boxed{}` instruction.
 - MMLU-Pro reproduces the official category-specific few-shot user prompt.
 - SuperGPQA reproduces the official reasoning-model zero-shot user prompt.
-- Neither validation benchmark is given an invented system role.
+- MMLU-Pro and SuperGPQA have no official system role, so the shared fallback is
+  prepended while their official user-level benchmark prompts remain unchanged.
 
 The current Qwen branch of `src/stage1/data.py` still hard-codes a boxed system
-prompt. It must be updated before training so that the runtime input matches
-these dataset-native queries.
+prompt. It must be updated before training so runtime input uses the system
+message stored in each row's `query`.
 
-Each dataset directory contains `pending.jsonl.gz`. Existing validation
-trajectories are retained separately in `completed.jsonl.gz`.
+Each dataset directory contains `pending.jsonl.gz`; the two largest training
+sets use equivalent `pending/part-*.jsonl.gz` shards for reliable transfer and
+parallel processing. All earlier MMLU-Pro and SuperGPQA trajectories were reset
+because they were generated under a different request/output protocol; both
+benchmarks must be regenerated uniformly.

cot_trajectory_generation/glm5.2/SuperGPQA/SOURCE.md

@@ -6,7 +6,7 @@
   https://github.com/SuperGPQA/SuperGPQA/blob/main/infer/data_loader.py
 - Official zero-shot template:
   https://github.com/SuperGPQA/SuperGPQA/blob/main/config/prompt/zero-shot.yaml
-- Official system prompt: none
-- Applied input: one user message using the official zero-shot template,
-  complete question, and all options formatted as `A) ...`.
-
+- Official system prompt: none; the shared fallback system prompt is used.
+- Applied user input: the official zero-shot template, complete question, and
+  all options formatted as `A) ...`.
+- All 26,529 rows are pending; earlier trajectories were discarded.

cot_trajectory_generation/glm5.2/SuperGPQA/pending.jsonl.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27602544b52370dd94cb01169ee2c8743bd5d4134a1ca9af23d937d6dbca2681
-size 8033709
+oid sha256:fed01a3f4f482e77c9e172a2b79f73ac46f87d7c99c9db3beffd8de5c9c7e0c3
+size 8197324

cot_trajectory_generation/glm5.2/WebInstruct-verified/SOURCE.md

@@ -3,4 +3,6 @@
 - Role: training
 - Source: https://huggingface.co/datasets/TIGER-Lab/WebInstruct-verified
 - Official dataset system prompt: none published
-- Applied input: one user message containing the original question.
+- Applied input: the shared fallback system prompt followed by the original
+  question.
+- Pending rows are split across `pending/part-*.jsonl.gz`.

cot_trajectory_generation/glm5.2/WebInstruct-verified/pending/part-00000.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06a48721c6f3ff8ec085d2ff43f3dba08e024cbdddfb2aafa8338be185db6b04
+size 7287327

cot_trajectory_generation/glm5.2/WebInstruct-verified/pending/part-00001.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41fb0b3080f2c1cc19a8dc703175d8231178f0970c2c93bf6466d3c173e2d432
+size 7297570

cot_trajectory_generation/glm5.2/WebInstruct-verified/pending/part-00002.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86ee79d3a9108c3dec919342d48e0d0b0bd133d914173998232ca88cbdfe7d19
+size 7278618

cot_trajectory_generation/glm5.2/WebInstruct-verified/pending/part-00003.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c43dab71575cb4bf0494d08066fec93b32dbdb02e9db2dc27579ab67a00f68e
+size 2449616

cot_trajectory_generation/glm5.2/ai2-arc-ARC-Challenge/SOURCE.md

@@ -4,4 +4,5 @@
 - Source: https://huggingface.co/datasets/allenai/ai2_arc
 - Configuration: `ARC-Challenge`
 - Official dataset system prompt: none published
-- Applied input: one user message containing the complete question and options.
+- Applied input: the shared fallback system prompt followed by one user message
+  containing the complete question and options.

cot_trajectory_generation/glm5.2/ai2-arc-ARC-Challenge/pending.jsonl.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:241b48a394ddbc1c6a7cf3eb146758cab92a84589a524c858cd5c6bbf60151bd
-size 123866
+oid sha256:61cedfc93ef04b594b360843f8144b2de68aaad4647d5542e6635c1b040050fa
+size 129500

cot_trajectory_generation/glm5.2/ai2-arc-ARC-Easy/SOURCE.md

@@ -4,4 +4,5 @@
 - Source: https://huggingface.co/datasets/allenai/ai2_arc
 - Configuration: `ARC-Easy`
 - Official dataset system prompt: none published
-- Applied input: one user message containing the complete question and options.
+- Applied input: the shared fallback system prompt followed by one user message
+  containing the complete question and options.

cot_trajectory_generation/glm5.2/ai2-arc-ARC-Easy/pending.jsonl.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3e81a5184b3959742df4c8b8d3566574888dcc7f2c0f2a1a988ad278c7b1e459
-size 235469
+oid sha256:f4b899ba9fcafdea01e003f6bd6c15a3f199c4ad7feae75ad34b54101689072f
+size 248182

cot_trajectory_generation/glm5.2/fallback_system_prompt.txt

@@ -0,0 +1 @@
+You are a chatbot who can solve problems. Please solve the following problem and give your thought process. Before giving the final result, you should output "Therefore, the answer is", and then give your final answer.

cot_trajectory_generation/glm5.2/manifest.json

@@ -35,27 +35,35 @@
     },
     "MMLU-Pro": {
       "total": 12032,
-      "pending": 11909,
-      "completed": 123
+      "pending": 12032,
+      "completed": 0
     },
     "SuperGPQA": {
       "total": 26529,
-      "pending": 26385,
-      "completed": 144
+      "pending": 26529,
+      "completed": 0
     }
   },
-  "training_input_policy": {
-    "policy": "follow each open-domain SFT dataset's original query setting",
-    "datasets_with_system_prompt": {
-      "Multi-subject-RLVR": "You are a chatbot who can solve problems. Please solve the following problem and give your thought process. Before giving the final result, you should output \"Therefore, the answer is\", and then give your final answer."
-    },
-    "datasets_without_system_prompt": [
+  "system_prompt_policy": {
+    "rule": "use the dataset original system prompt when present; otherwise use fallback_system_prompt.txt",
+    "fallback": "You are a chatbot who can solve problems. Please solve the following problem and give your thought process. Before giving the final result, you should output \"Therefore, the answer is\", and then give your final answer.",
+    "original_system_prompt_datasets": [
+      "Multi-subject-RLVR"
+    ],
+    "fallback_datasets": [
       "ai2-arc-ARC-Easy",
       "ai2-arc-ARC-Challenge",
       "trivia-qa",
-      "WebInstruct-verified"
-    ],
-    "boxed_prompt_added": false,
-    "training_code_note": "The current Qwen branch in src/stage1/data.py still hard-codes a boxed system prompt and must be changed before aligned training."
+      "WebInstruct-verified",
+      "MMLU-Pro",
+      "SuperGPQA"
+    ]
+  },
+  "file_layout": {
+    "default": "<dataset>/pending.jsonl.gz",
+    "sharded": {
+      "Multi-subject-RLVR": "Multi-subject-RLVR/pending/part-*.jsonl.gz",
+      "WebInstruct-verified": "WebInstruct-verified/pending/part-*.jsonl.gz"
+    }
   }
 }

cot_trajectory_generation/glm5.2/trivia-qa/SOURCE.md

@@ -4,4 +4,5 @@
 - Source: https://github.com/mandarjoshi90/triviaqa
 - Configuration used locally: `rc.nocontext`
 - Official dataset system prompt: none published
-- Applied input: one user message containing the original question.
+- Applied input: the shared fallback system prompt followed by the original
+  question.

cot_trajectory_generation/glm5.2/trivia-qa/pending.jsonl.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6bee048e9bf7f3e7adc13888464540ce545c669bfcfe5371b36aa338f2000b6f
-size 7124280
+oid sha256:6d12c3ef20ad036e2f3d5bccec24b10b66f86d7e2f4d695c1b28e9583bc242c5
+size 7735202