builds/ling3-moe: Ling 3's markup and vocabulary (BailingMoeV3, bailingmoe3)

Recipe ling3-moe (MoE shares) rendered through inclusionAI/Ling-3.0-flash's own template (auto_fmt); vocab sweep over its 157184-row tokenizer (100% reachable). For Ling-3.0-flash-Sante (same tokenizer/template). calib_train 4.69M tokens, calib_longctx 0.78M.

builds/ling3-moe/calib_longctx.manifest.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19fb375a62a9121bd6860476e5e70ad4550de0526c1b478e613c65c3a9d9f20d
+size 7145

builds/ling3-moe/calib_longctx.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23ba0aab0b6ce38c3758665c85787282fab8542a04380355b83876ba4dc30f8f
+size 2662969

builds/ling3-moe/calib_train.manifest.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2269929a8911be17d35c17849852cf07728a0730498220fb01f49688cf872a5
+size 986536

builds/ling3-moe/calib_train.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5becda335fd68da747199170af7e6d9186988074b2920e5ba1b796e9f2ffcb65
+size 15207782

builds/ling3-moe/manifest.json

@@ -0,0 +1,273 @@
+{
+  "build": "ling3-moe",
+  "built_at": null,
+  "build_tool_version": "build.py/1.0",
+  "recipe": {
+    "path": "recipes/ling3-moe.yaml",
+    "sha256": "93a29eec37d9e63484ab121f011b28ced8302afcc07074afb2b993cd1a91d884",
+    "seed": 20261013
+  },
+  "model": {
+    "repo": "inclusionAI/Ling-3.0-flash",
+    "tokenizer": "/src/tokenizer.json",
+    "gguf": "/gguf/ling3-bf16.gguf",
+    "vocab_size": 157184,
+    "layers": 72,
+    "sliding_window": null,
+    "context_length": 262144,
+    "llama_cpp_commit": "097cf33d9499"
+  },
+  "tokenizer": {
+    "path": "/src/tokenizer.json",
+    "sha256": "40fb9d7d7795b8bd305aeff39ce9963f3f450915b9553f2938e009be9a1fed60",
+    "rows": 157184,
+    "pre_tokenizer": "llama4"
+  },
+  "llama_cpp": {
+    "commit": "097cf33d9499",
+    "note": "llama-imatrix must be given --parse-special or the chat markup in calib_train.txt is tokenised as literal text (parse_special defaults to false in that tool)"
+  },
+  "calib_train": {
+    "path": "calib_train.txt",
+    "sha256": "5becda335fd68da747199170af7e6d9186988074b2920e5ba1b796e9f2ffcb65",
+    "bytes": 15207782,
+    "documents": 2726,
+    "tokens": 4685467,
+    "target_tokens": 5000000,
+    "document_separator": "\n\n",
+    "shares_requested_percent": {
+      "agentic": 22.0,
+      "code": 16.0,
+      "graphics": 4.0,
+      "longctx": 10.0,
+      "multilingual": 18.0,
+      "reasoning": 12.0,
+      "structured": 6.0,
+      "vocab_sweep": 12.0
+    },
+    "shares_actual_percent": {
+      "agentic": 23.585,
+      "code": 17.097,
+      "graphics": 4.272,
+      "longctx": 11.443,
+      "multilingual": 19.421,
+      "reasoning": 12.817,
+      "structured": 4.626,
+      "vocab_sweep": 6.739
+    },
+    "documents_by_category": {
+      "agentic": 377,
+      "code": 486,
+      "graphics": 120,
+      "longctx": 17,
+      "multilingual": 171,
+      "reasoning": 972,
+      "structured": 451,
+      "vocab_sweep": 132
+    },
+    "tokens_by_category": {
+      "agentic": 1105082,
+      "code": 801088,
+      "graphics": 200155,
+      "longctx": 536153,
+      "multilingual": 909943,
+      "reasoning": 600515,
+      "structured": 216772,
+      "vocab_sweep": 315759
+    },
+    "synthetic_tokens": 2089024,
+    "synthetic_percent": 44.585,
+    "document_tokens": {
+      "p50": 701,
+      "p90": 3829,
+      "p95": 5871,
+      "p99": 19278,
+      "max": 54143,
+      "min": 61
+    }
+  },
+  "pool_sources": {
+    "pool/_quarantine/eval-residue.jsonl": {
+      "sha256": "9e7703a3c205b425f83f17110cf7a00b7558f8ed501b45f1fcb0f4145f5f2579",
+      "bytes": 1635378,
+      "documents": 64
+    },
+    "pool/_quarantine/removed.jsonl": {
+      "sha256": "28615d8080e4b3b95a29447402e8163786c42e23e6c5175a18783a591bde13d0",
+      "bytes": 8600246,
+      "documents": 1056
+    },
+    "pool/_quarantine/untokenizable.jsonl": {
+      "sha256": "8226d1b281ce13ceab909c06383d78176b3cc5c2d3f1a49094ef808e3f220335",
+      "bytes": 88848,
+      "documents": 2
+    },
+    "pool/agentic/synthetic/conversations.jsonl": {
+      "sha256": "40478e56fdee9bf965b0e56800e3d940e772db11e856f0170d92043d86861973",
+      "bytes": 5641610,
+      "documents": 531
+    },
+    "pool/agentic/synthetic/legacy-dsv4-rendered.jsonl": {
+      "sha256": "24e99d36930e6547e90058654b712ae50a1294a1cdc4da7b38ae688a97537dca",
+      "bytes": 948845,
+      "documents": 50
+    },
+    "pool/code/legacy-dsv4.jsonl": {
+      "sha256": "28076d09ec882624d77b30289094ab65ae8c1fb57d3dceb55011b5e81fa99916",
+      "bytes": 471958,
+      "documents": 119
+    },
+    "pool/code/repos.jsonl": {
+      "sha256": "549d67465b052d78a4277da12033840309d767cd897f7d1280df408b43bbc916",
+      "bytes": 9133710,
+      "documents": 831
+    },
+    "pool/german/agentic_de.jsonl": {
+      "sha256": "f7afba5c9872c9e69f6fc9b0e4eb72dc1e80d8eaef1c92c736636b924eb00609",
+      "bytes": 1230904,
+      "documents": 735
+    },
+    "pool/german/germanrag.jsonl": {
+      "sha256": "23a5c7e8d2cea57ab8e01d5c8b953b81e6eb7811f12dae4b59b8396ceebf27ba",
+      "bytes": 1029498,
+      "documents": 265
+    },
+    "pool/german/longctx.jsonl": {
+      "sha256": "26d9b0f152105337fad88e098dcffa7dbcae8fe3eccd8c0a35850603b485ea60",
+      "bytes": 2488298,
+      "documents": 24
+    },
+    "pool/german/oasst2.jsonl": {
+      "sha256": "8ca93fad0812efdc21ed2ce06123627e97f41345eda8939419bd5af9b131ac01",
+      "bytes": 888635,
+      "documents": 410
+    },
+    "pool/german/ultrachat_de.jsonl": {
+      "sha256": "d42ccac07f7187c1f55f4de99335be4164f3b663d4319c2c29c31f234019d280",
+      "bytes": 1698445,
+      "documents": 488
+    },
+    "pool/german/wikipedia.jsonl": {
+      "sha256": "f0771f38f5f42aee0ff88741b7f6188b8b3bad3f0927181f1038716374285494",
+      "bytes": 2291623,
+      "documents": 443
+    },
+    "pool/graphics/legacy-dsv4.jsonl": {
+      "sha256": "09e62f79a8859e2c0f275bc38b93dde63f5751660eb3e9fc8636782cb47d3689",
+      "bytes": 2068241,
+      "documents": 335
+    },
+    "pool/graphics/repos.jsonl": {
+      "sha256": "862600d1842bd530e65884d8bc4bfcfa8892ddd1e430a927f30d35651539b7d1",
+      "bytes": 6231338,
+      "documents": 827
+    },
+    "pool/longctx/legacy-dsv4.jsonl": {
+      "sha256": "1f17bf4553c25558953fd746ce3bce278e932690e02d926395f8ecf21dacc100",
+      "bytes": 591648,
+      "documents": 10
+    },
+    "pool/longctx/module-bundles.jsonl": {
+      "sha256": "767b0cb154af75a8e5023e948f39fb3cbef424c99ba643779c5bd3e289945bb5",
+      "bytes": 9837165,
+      "documents": 83
+    },
+    "pool/multilingual/legacy-dsv4-wikipedia.jsonl": {
+      "sha256": "aa0344e84c60e1c423ac543d205bad77e3c0cc475799ddcb76d4e338cbdfb266",
+      "bytes": 420903,
+      "documents": 14
+    },
+    "pool/multilingual/wikipedia.jsonl": {
+      "sha256": "0cd11f4e6a768439f37aa354c422eb980d161e9ac2a8dde856a2f61a362480d0",
+      "bytes": 12207837,
+      "documents": 687
+    },
+    "pool/reasoning/synthetic/conversations.jsonl": {
+      "sha256": "9e890645baaad2045ac87ce9a9649082ccc37bec1c3d72b1da2a5b6c92af3e2f",
+      "bytes": 6724419,
+      "documents": 3015
+    },
+    "pool/reasoning/synthetic/legacy-dsv4-rendered.jsonl": {
+      "sha256": "e690d0acb22a427e0b17f57ee8f255a36c5550de23775a52c9117198b02e8f95",
+      "bytes": 440412,
+      "documents": 153
+    },
+    "pool/structured/legacy-dsv4.jsonl": {
+      "sha256": "e0a6537986934b5418b6d711c2712728979e1019eb6ca40885429fc068e8c6d4",
+      "bytes": 108168,
+      "documents": 19
+    },
+    "pool/structured/repos.jsonl": {
+      "sha256": "631c8fd66fa7fc1d60fe50e257385512c77e0c1d0e1a86f88a482c29e0b2f83c",
+      "bytes": 1226353,
+      "documents": 227
+    },
+    "pool/structured/synthetic/structured-in-chat.jsonl": {
+      "sha256": "435a1be5a7b8bd230105b85fbe708649fb2ddef5a65198bdd808bcff1b3d1b64",
+      "bytes": 395812,
+      "documents": 207
+    },
+    "pool/vocab_sweep/synthetic/kolibri-1-moe.jsonl": {
+      "sha256": "3d165633bb5201090c11a63671e4f993e054cc342881007eb03b29e03d89cf27",
+      "bytes": 1323571,
+      "documents": 110
+    },
+    "pool/vocab_sweep/synthetic/legacy-dsv4-rendered.jsonl": {
+      "sha256": "e30414d83fc63d1a5bb293a50d174eb65385efa237111f7ae001f923a9498424",
+      "bytes": 745316,
+      "documents": 65
+    },
+    "pool/vocab_sweep/synthetic/ling3-moe.jsonl": {
+      "sha256": "3ea94880f0d64aad55055664957266d3b124249c20a2d5a09e27854d7acf35e6",
+      "bytes": 1459777,
+      "documents": 132
+    },
+    "pool/vocab_sweep/synthetic/minicpm-v-4.6-moe.jsonl": {
+      "sha256": "c8bcca4781cb4b842f42e3c4c5c699effde6aca401e85f656ace5fd4946db2b4",
+      "bytes": 2509124,
+      "documents": 208
+    },
+    "pool/vocab_sweep/synthetic/minicpm-v-4.6.jsonl": {
+      "sha256": "c8bcca4781cb4b842f42e3c4c5c699effde6aca401e85f656ace5fd4946db2b4",
+      "bytes": 2509124,
+      "documents": 208
+    },
+    "pool/vocab_sweep/synthetic/minicpm-v-4.7-moe.jsonl": {
+      "sha256": "e2c3b5b513d5c43e4f57d04f2094b8d62ba775bf8559b2f116327bbc36321417",
+      "bytes": 2746220,
+      "documents": 216
+    },
+    "pool/vocab_sweep/synthetic/minicpm-v-4.7.jsonl": {
+      "sha256": "2dfd825f85e2c9f31a28e3d1dc1a1fe25b3fa06fd0b3ac94630d21130a3c0cec",
+      "bytes": 2512036,
+      "documents": 208
+    },
+    "pool/vocab_sweep/synthetic/muse-glimmer-30b.jsonl": {
+      "sha256": "4eb612024896df3b979fa74b92cc27ddbce56af100a14611bcfccd4ea9ce92c5",
+      "bytes": 2068343,
+      "documents": 172
+    },
+    "pool/vocab_sweep/synthetic/nemotron-3.5-lightning.jsonl": {
+      "sha256": "cd73220ef1ac82132a4c2b217f77af7b2eaea2c06939394fe36650221cc22a1d",
+      "bytes": 1269096,
+      "documents": 113
+    },
+    "pool/vocab_sweep/synthetic/qwen3.8-flash-next-moe.jsonl": {
+      "sha256": "ee4657dd2394309d06d6e6a978e020d535e0f899c0b80ca2067672889a276913",
+      "bytes": 2508961,
+      "documents": 208
+    }
+  },
+  "calib_longctx": {
+    "path": "calib_longctx.txt",
+    "sha256": "23ba0aab0b6ce38c3758665c85787282fab8542a04380355b83876ba4dc30f8f",
+    "documents": 28,
+    "tokens": 781245,
+    "document_tokens": {
+      "min": 17304,
+      "p50": 32742,
+      "max": 32767
+    },
+    "contiguous": true
+  }
+}

recipes/ling3-moe.yaml

@@ -0,0 +1,77 @@
+# Calibration recipe with the MoE shares, Ling 3's markup and vocabulary (BailingMoeV3, bailingmoe3).
+# Written from minicpm-v-4.6-moe.yaml. Renders through inclusionAI/Ling-3.0-flash's own template (auto_fmt)
+# and sweeps its 157184-row vocabulary; every Ling 3 model with the same tokenizer.json and template
+# (sha-checked by the build node) takes this build. The August Ling 3.0 imatrix had none of this: a flat
+# text corpus, tokenized without --parse-special.
+#
+# The template: <role>SYSTEM|HUMAN|ASSISTANT|OBSERVATION</role> ... <|role_end|>; enable_thinking (default on)
+# writes "detailed thinking on|off" into the system turn and opens the answer with "\n<think>" (on) or
+# "\n<think></think>" (off); auto_fmt passes enable_thinking per document, so the thinking split of the
+# recipe reaches it. Tool calls are XML (<tool_call>name<arg_key>..</arg_key><arg_value>..</arg_value>),
+# results come back in <tool_response>. No reasoning effort.
+#
+# Architecture facts that matter here (the large BailingMoeV3):
+#   hidden_size = 4096 (%256 = 0)
+#   moe_intermediate_size = 1280 (%256 = 0), shared expert 1280; the 4 leading dense blocks 10240
+#   512 routed experts, 8 active: ~5M * 8 / 512 = ~78k tokens per expert
+#   63 KDA (linear attention) and 9 MLA blocks; untied embeddings
+#
+# Shares (they sum to 100, so they are percentages): the MoE ones of minicpm-v-4.6-moe.
+#   agentic 22      tool markup is a dense token set natural text never emits
+#   code 16
+#   reasoning 12    on MoE the slack goes to multilingual and vocab, which reach rarely routed experts
+#   multilingual 18 non-Latin text routes to experts English never touches
+#   longctx 10      no sliding window; long range also gets calib_longctx
+#   vocab_sweep 12  regenerated against THIS tokenizer; every id routes somewhere
+#   structured 6    JSON, YAML, TOML, SQL, as files and inside assistant turns
+#   graphics 4
+
+name: ling3-moe
+
+model:
+  repo: inclusionAI/Ling-3.0-flash
+  tokenizer: /src/tokenizer.json
+  gguf: /gguf/ling3-bf16.gguf
+  vocab_size: 157184
+  layers: 72
+  sliding_window: null
+  context_length: 262144
+  llama_cpp_commit: '097cf33d9499'
+
+seed: 20261013
+
+chat:
+  format: auto
+  add_bos_per_document: false
+
+calib_train:
+  target_tokens: 5000000
+  max_doc_fraction:
+    default: 0.05
+    longctx: 0.12
+  shares:
+    agentic: 22
+    code: 16
+    reasoning: 12
+    multilingual: 18
+    longctx: 10
+    vocab_sweep: 12
+    structured: 6
+    graphics: 4
+
+calib_longctx:
+  target_tokens: 750000
+  doc_tokens_min: 16384
+  doc_tokens_max: 32768
+  sources: [longctx]
+
+exclude:
+  # Anything pre-rendered with another model's markup.
+  render: [dsv4, nemotron, muse-glimmer]
+  provenance_key: excluded_from_builds
+
+notes:
+  imatrix: >
+    llama-imatrix defaults to parse_special = false. Without --parse-special the
+    chat markup in calib_train.txt is tokenised as literal punctuation and the
+    agentic and reasoning slices calibrate on text the model never sees.