Add certified 32K Card Corpus V2 training shards

Publish assistant-only 32K tokenized train, eval, and test shards for all 188,662 Card Corpus V2 examples. Include the exact tokenizer, chat contract, indexes, loss-mask labels, and portable manifests while preserving the certified source Parquet splits.

.gitattributes

@@ -1 +1,8 @@
 *.parquet filter=lfs diff=lfs merge=lfs -text
+tokenized/32k-v2/eval/input_ids.bin filter=lfs diff=lfs merge=lfs -text
+tokenized/32k-v2/eval/labels.bin filter=lfs diff=lfs merge=lfs -text
+tokenized/32k-v2/test/input_ids.bin filter=lfs diff=lfs merge=lfs -text
+tokenized/32k-v2/test/labels.bin filter=lfs diff=lfs merge=lfs -text
+tokenized/32k-v2/train/examples.jsonl filter=lfs diff=lfs merge=lfs -text
+tokenized/32k-v2/train/input_ids.bin filter=lfs diff=lfs merge=lfs -text
+tokenized/32k-v2/train/labels.bin filter=lfs diff=lfs merge=lfs -text

README.md

@@ -88,6 +88,33 @@ The published build passed the complete V2 release contract:
 Machine-readable evidence is included in `metadata/manifest.json`,
 `metadata/audit.json`, and `metadata/roadmap.json`.
 
+## Tokenized 32K training shards
+
+Framework-ready assistant-only SFT shards are published under
+`tokenized/32k-v2/`. They were generated only after the source release passed
+every quality gate.
+
+| Partition | Examples | Tokens | Supervised assistant tokens |
+|---|---:|---:|---:|
+| `train` | 184,801 | 24,175,171 | 16,219,759 |
+| `eval` | 1,920 | 250,307 | 168,265 |
+| `test` | 1,941 | 255,390 | 171,607 |
+| **Total** | **188,662** | **24,680,868** | **16,559,631** |
+
+The binary layout is documented by each partition's `sft.idx.json`:
+
+- `input_ids.bin`: little-endian unsigned 32-bit token IDs;
+- `labels.bin`: little-endian signed 32-bit causal labels;
+- prompt/system positions use the `-100` ignore index;
+- assistant tokens and EOS are supervised;
+- vocabulary size: 32,000;
+- chat contract: `complexity-chat-v2`;
+- tokenizer SHA-256: `852759014538299ed8e941a83fa1f254fbbaff7824189a4c7455f038cfede3f2`.
+
+The exact tokenizer files used for the build are included in
+`tokenized/32k-v2/tokenizer/`. The source Parquet splits remain available for
+inspection and alternative preprocessing.
+
 ## Source and license
 
 Generated by the open-source

metadata/hf_release.json

@@ -31,5 +31,30 @@
     "test": 1941,
     "train": 184801,
     "validation": 1920
+  },
+  "tokenized_32k": {
+    "examples": 188662,
+    "manifest_sha256": "7e32d6f8f312dcd76e663fb2a3856fb7d52727c37da0ede369012897d4a5e92b",
+    "partitions": {
+      "eval": {
+        "examples": 1920,
+        "supervised_tokens": 168265,
+        "tokens": 250307
+      },
+      "test": {
+        "examples": 1941,
+        "supervised_tokens": 171607,
+        "tokens": 255390
+      },
+      "train": {
+        "examples": 184801,
+        "supervised_tokens": 16219759,
+        "tokens": 24175171
+      }
+    },
+    "path": "tokenized/32k-v2",
+    "supervised_tokens": 16559631,
+    "tokenizer_sha256": "852759014538299ed8e941a83fa1f254fbbaff7824189a4c7455f038cfede3f2",
+    "tokens": 24680868
   }
 }

tokenized/32k-v2/chat_template.json

@@ -0,0 +1,20 @@
+{
+  "assistant_envelope": {
+    "final_end": "\n</final>",
+    "final_start": "\n<final>\n",
+    "scope": "reasoning_tasks",
+    "think_end": "\n</think>",
+    "think_start": "<think>\n",
+    "type": "optional_think_final"
+  },
+  "assistant_only_loss": true,
+  "assistant_prefix": "Assistant:\n",
+  "eos_token": "<|endoftext|>",
+  "id": "complexity-chat-v2",
+  "system_format": "System:\n{content}\n\n",
+  "system_prompt": "",
+  "training_projection": "card_corpus_v2_direct",
+  "turn_separator": "\n\n",
+  "user_format": "User:\n{content}\n\n",
+  "version": 2
+}

tokenized/32k-v2/eval/input_ids.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd07dc7850a3987baafa05039fd764c3c9581089c58c028f3d010f3152881aff
+size 1001228

tokenized/32k-v2/eval/labels.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9dc88bffcc619295b3effdff9a24c1fdf54093980771f121700b8e976fc8f2da
+size 1001228

tokenized/32k-v2/eval/sft.idx.json

@@ -0,0 +1,20 @@
+{
+  "assistant_supervision": "all_assistant_turns",
+  "chat_template_id": "complexity-chat-v2",
+  "eos_token_id": 0,
+  "examples": 1920,
+  "examples_sha256": "13c36fa15b317ce16851a557f1f5f2fc5ccec705fda406a5aec3f8507018dbf6",
+  "format": "complexity-sft-token-shard-v2",
+  "ignore_index": -100,
+  "input_dtype": "<u4",
+  "input_ids_sha256": "fd07dc7850a3987baafa05039fd764c3c9581089c58c028f3d010f3152881aff",
+  "label_dtype": "<i4",
+  "labels_sha256": "9dc88bffcc619295b3effdff9a24c1fdf54093980771f121700b8e976fc8f2da",
+  "num_tokens": 250307,
+  "partition": "eval",
+  "source_sha256": "fe0a4931ab213012d9494777573f3ca0c028669a2db512d51ccb54001e6fd1a6",
+  "supervised_tokens": 168265,
+  "tokenizer": "hf:tokenizer",
+  "tokenizer_sha256": "852759014538299ed8e941a83fa1f254fbbaff7824189a4c7455f038cfede3f2",
+  "vocab_size": 32000
+}

tokenized/32k-v2/manifest.json

@@ -0,0 +1,71 @@
+{
+  "created_at": "2026-08-13T11:34:18.052651+00:00",
+  "format": "complexity-card-corpus-v2-tokenized-v1",
+  "partitions": {
+    "eval": {
+      "assistant_supervision": "all_assistant_turns",
+      "chat_template_id": "complexity-chat-v2",
+      "eos_token_id": 0,
+      "examples": 1920,
+      "examples_sha256": "13c36fa15b317ce16851a557f1f5f2fc5ccec705fda406a5aec3f8507018dbf6",
+      "format": "complexity-sft-token-shard-v2",
+      "ignore_index": -100,
+      "input_dtype": "<u4",
+      "input_ids_sha256": "fd07dc7850a3987baafa05039fd764c3c9581089c58c028f3d010f3152881aff",
+      "label_dtype": "<i4",
+      "labels_sha256": "9dc88bffcc619295b3effdff9a24c1fdf54093980771f121700b8e976fc8f2da",
+      "num_tokens": 250307,
+      "partition": "eval",
+      "source_sha256": "fe0a4931ab213012d9494777573f3ca0c028669a2db512d51ccb54001e6fd1a6",
+      "supervised_tokens": 168265,
+      "tokenizer": "hf:tokenizer",
+      "tokenizer_sha256": "852759014538299ed8e941a83fa1f254fbbaff7824189a4c7455f038cfede3f2",
+      "vocab_size": 32000
+    },
+    "test": {
+      "assistant_supervision": "all_assistant_turns",
+      "chat_template_id": "complexity-chat-v2",
+      "eos_token_id": 0,
+      "examples": 1941,
+      "examples_sha256": "5720a9eb9da1c33254ff66748b54e5d2a44571a91944bfcbb18ce99663206ac9",
+      "format": "complexity-sft-token-shard-v2",
+      "ignore_index": -100,
+      "input_dtype": "<u4",
+      "input_ids_sha256": "fc8bf69bd838ba2afc3743b096dba4600b6de4c73d8bd9b3d81856060aed56e8",
+      "label_dtype": "<i4",
+      "labels_sha256": "98bbc5efc42131070f9b01a9f663603a175bbcc81351fab866dafcb1e54d8e89",
+      "num_tokens": 255390,
+      "partition": "test",
+      "source_sha256": "fe0a4931ab213012d9494777573f3ca0c028669a2db512d51ccb54001e6fd1a6",
+      "supervised_tokens": 171607,
+      "tokenizer": "hf:tokenizer",
+      "tokenizer_sha256": "852759014538299ed8e941a83fa1f254fbbaff7824189a4c7455f038cfede3f2",
+      "vocab_size": 32000
+    },
+    "train": {
+      "assistant_supervision": "all_assistant_turns",
+      "chat_template_id": "complexity-chat-v2",
+      "eos_token_id": 0,
+      "examples": 184801,
+      "examples_sha256": "4b31e29753690808c716294614ec48550cc33ec054f2fef9406085f5053917c9",
+      "format": "complexity-sft-token-shard-v2",
+      "ignore_index": -100,
+      "input_dtype": "<u4",
+      "input_ids_sha256": "8f9368f82cebf6eccc2d13abd0c866433a7d925e3cd31a011c4494465640a0d7",
+      "label_dtype": "<i4",
+      "labels_sha256": "b0b33cbfbfcb304e34e5529c8fc4f6cb981ce4d371207602f7be02bdaf84558e",
+      "num_tokens": 24175171,
+      "partition": "train",
+      "source_sha256": "fe0a4931ab213012d9494777573f3ca0c028669a2db512d51ccb54001e6fd1a6",
+      "supervised_tokens": 16219759,
+      "tokenizer": "hf:tokenizer",
+      "tokenizer_sha256": "852759014538299ed8e941a83fa1f254fbbaff7824189a4c7455f038cfede3f2",
+      "vocab_size": 32000
+    }
+  },
+  "quality_status": "passed",
+  "source": "AETHORIA-AI/complexity-atlas-posttrain:data/*.parquet",
+  "total_examples": 188662,
+  "total_supervised_tokens": 16559631,
+  "total_tokens": 24680868
+}

tokenized/32k-v2/test/input_ids.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc8bf69bd838ba2afc3743b096dba4600b6de4c73d8bd9b3d81856060aed56e8
+size 1021560

tokenized/32k-v2/test/labels.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98bbc5efc42131070f9b01a9f663603a175bbcc81351fab866dafcb1e54d8e89
+size 1021560

tokenized/32k-v2/test/sft.idx.json

@@ -0,0 +1,20 @@
+{
+  "assistant_supervision": "all_assistant_turns",
+  "chat_template_id": "complexity-chat-v2",
+  "eos_token_id": 0,
+  "examples": 1941,
+  "examples_sha256": "5720a9eb9da1c33254ff66748b54e5d2a44571a91944bfcbb18ce99663206ac9",
+  "format": "complexity-sft-token-shard-v2",
+  "ignore_index": -100,
+  "input_dtype": "<u4",
+  "input_ids_sha256": "fc8bf69bd838ba2afc3743b096dba4600b6de4c73d8bd9b3d81856060aed56e8",
+  "label_dtype": "<i4",
+  "labels_sha256": "98bbc5efc42131070f9b01a9f663603a175bbcc81351fab866dafcb1e54d8e89",
+  "num_tokens": 255390,
+  "partition": "test",
+  "source_sha256": "fe0a4931ab213012d9494777573f3ca0c028669a2db512d51ccb54001e6fd1a6",
+  "supervised_tokens": 171607,
+  "tokenizer": "hf:tokenizer",
+  "tokenizer_sha256": "852759014538299ed8e941a83fa1f254fbbaff7824189a4c7455f038cfede3f2",
+  "vocab_size": 32000
+}

tokenized/32k-v2/tokenizer/chat_template.jinja

@@ -0,0 +1 @@
+{{- "" -}}{%- for message in messages -%}{%- if message['role'] == 'system' -%}{{- "System:\n" + (message['content'] | trim) + "\n\n" -}}{%- elif message['role'] == 'user' -%}{{- "User:\n" + (message['content'] | trim) + "\n\n" -}}{%- elif message['role'] == 'assistant' -%}{{- "Assistant:\n" + (message['content'] | trim) + eos_token -}}{%- endif -%}{%- endfor -%}{%- if add_generation_prompt -%}{{- "Assistant:\n" -}}{%- endif -%}

tokenized/32k-v2/tokenizer/special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "<s>",
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

tokenized/32k-v2/tokenizer/tokenizer_config.json

@@ -0,0 +1,46 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "model_max_length": 2048,
+  "pad_token": "<pad>",
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "unk_token": "<unk>"
+}

tokenized/32k-v2/train/examples.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b31e29753690808c716294614ec48550cc33ec054f2fef9406085f5053917c9
+size 93872905

tokenized/32k-v2/train/input_ids.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f9368f82cebf6eccc2d13abd0c866433a7d925e3cd31a011c4494465640a0d7
+size 96700684

tokenized/32k-v2/train/labels.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0b33cbfbfcb304e34e5529c8fc4f6cb981ce4d371207602f7be02bdaf84558e
+size 96700684

tokenized/32k-v2/train/sft.idx.json

@@ -0,0 +1,20 @@
+{
+  "assistant_supervision": "all_assistant_turns",
+  "chat_template_id": "complexity-chat-v2",
+  "eos_token_id": 0,
+  "examples": 184801,
+  "examples_sha256": "4b31e29753690808c716294614ec48550cc33ec054f2fef9406085f5053917c9",
+  "format": "complexity-sft-token-shard-v2",
+  "ignore_index": -100,
+  "input_dtype": "<u4",
+  "input_ids_sha256": "8f9368f82cebf6eccc2d13abd0c866433a7d925e3cd31a011c4494465640a0d7",
+  "label_dtype": "<i4",
+  "labels_sha256": "b0b33cbfbfcb304e34e5529c8fc4f6cb981ce4d371207602f7be02bdaf84558e",
+  "num_tokens": 24175171,
+  "partition": "train",
+  "source_sha256": "fe0a4931ab213012d9494777573f3ca0c028669a2db512d51ccb54001e6fd1a6",
+  "supervised_tokens": 16219759,
+  "tokenizer": "hf:tokenizer",
+  "tokenizer_sha256": "852759014538299ed8e941a83fa1f254fbbaff7824189a4c7455f038cfede3f2",
+  "vocab_size": 32000
+}