sima-laya / models.json
TDoSiMa's picture
Update the model list
ca1277e verified
Raw History Blame Contribute Delete
13.5 kB
{
"version": 1,
"models": {
"general": {
"title": "Laya",
"about": "The general question model: routing, triage, moderation and yes-or-no questions about a piece of English text.",
"card": {
"encoder": "ModernBERT-large, 421M parameters",
"languages": "English",
"upstream": "convaiinnovations/laya",
"license": "Apache-2.0"
},
"path": "general",
"checkpoint": "laya",
"precision": "BF16",
"graphs": {
"128": "laya_s128_stage1_mla.elf",
"256": "laya_s256_stage1_mla.elf",
"512": "laya_s512_stage1_mla.elf"
},
"latency_ms": {
"128": 19.5,
"256": 35.2,
"512": 87.5
},
"agreement": "100 / 100",
"files": [
{
"name": "act_tail.f32",
"bytes": 6152,
"sha256": "d8d35135d97cac56479df92a7fe4a330dd46ed13e10dd7b8cca04963f23da3c6"
},
{
"name": "laya_config.json",
"bytes": 1040,
"sha256": "6335dc8584e0febe4919b75b00b4b7f63762d65df12f96343e00ee485f665862"
},
{
"name": "laya_s128_stage1_mla.elf",
"bytes": 880747528,
"sha256": "4f0c747f0aab3b24a9bf3311880c8c0c970464fdc60ee0783129c1f1fb80b997"
},
{
"name": "laya_s256_stage1_mla.elf",
"bytes": 992027048,
"sha256": "bf9136106507d490249f5476c1b186a769d271f053099832b990c4638913f426"
},
{
"name": "laya_s512_stage1_mla.elf",
"bytes": 1293300552,
"sha256": "df02ae5dacb3723d03b205354028f1babe26c06e5e1e262d959d5b2a6ddfd463"
},
{
"name": "token_embeddings.bf16",
"bytes": 103153664,
"sha256": "ccdb9412fba7fb7f0e62a3123e5821477d4d359670545efd27807d9727236212"
},
{
"name": "tokenizer.json",
"bytes": 3583228,
"sha256": "6c8aaa9a542084f2457eab775d4eeb51f92a70c0fd9de28d5edb0ddec3c08d30"
}
],
"kind": "laya"
},
"general-int8": {
"title": "Laya INT8",
"about": "The general model with its weights stored as 8-bit integers: 28% faster and a third smaller, slightly less exact.",
"card": {
"encoder": "ModernBERT-large, 421M parameters",
"languages": "English",
"upstream": "convaiinnovations/laya",
"license": "Apache-2.0"
},
"path": "general-int8",
"checkpoint": "laya",
"precision": "A_BF16_W_INT8",
"graphs": {
"128": "laya_s128_stage1_mla.elf"
},
"latency_ms": {
"128": 14.0
},
"agreement": "98 / 100",
"files": [
{
"name": "act_tail.f32",
"bytes": 6152,
"sha256": "d8d35135d97cac56479df92a7fe4a330dd46ed13e10dd7b8cca04963f23da3c6"
},
{
"name": "laya_config.json",
"bytes": 963,
"sha256": "1e17c230076035cdd57bc590de0dca4b739c57bd4f6f7265798f284186def266"
},
{
"name": "laya_s128_stage1_mla.elf",
"bytes": 561316472,
"sha256": "94a75dbf58edb68f9ee80a306db9c1d7bb4aed1bcd15b515bdcf11919d2e22c3"
},
{
"name": "token_embeddings.bf16",
"bytes": 103153664,
"sha256": "ccdb9412fba7fb7f0e62a3123e5821477d4d359670545efd27807d9727236212"
},
{
"name": "tokenizer.json",
"bytes": 3583228,
"sha256": "6c8aaa9a542084f2457eab775d4eeb51f92a70c0fd9de28d5edb0ddec3c08d30"
}
],
"kind": "laya"
},
"typed-decisions": {
"title": "Laya typed-decisions",
"about": "Fine-tuned by upstream on typed-decision workflows, with graphs for inputs of up to 1024 tokens.",
"card": {
"encoder": "ModernBERT-large, 421M parameters",
"languages": "English",
"upstream": "convaiinnovations/laya",
"license": "Apache-2.0"
},
"path": "typed-decisions",
"checkpoint": "laya-typed-decisions",
"precision": "BF16",
"graphs": {
"128": "laya_s128_stage1_mla.elf",
"512": "laya_s512_stage1_mla.elf",
"1024": "laya_s1024_stage1_mla.elf"
},
"latency_ms": {
"128": 19.4,
"512": 87.5,
"1024": 282.0
},
"agreement": "99 / 100",
"files": [
{
"name": "act_tail.f32",
"bytes": 6152,
"sha256": "d8d35135d97cac56479df92a7fe4a330dd46ed13e10dd7b8cca04963f23da3c6"
},
{
"name": "laya_config.json",
"bytes": 1060,
"sha256": "57dc71fe726964f653759fc86003191a1b63a05ab4e428cf2b685ae83992b639"
},
{
"name": "laya_s1024_stage1_mla.elf",
"bytes": 2069640872,
"sha256": "9dbd0a730f93b61a18e30c2df50ebcb48d03154554faa3faca5a20338cd53c8a"
},
{
"name": "laya_s128_stage1_mla.elf",
"bytes": 880747536,
"sha256": "4a77a7f77bd20d02acc974624bbdf3b88ebd1bb137d2f7fa379db4804178bfb2"
},
{
"name": "laya_s512_stage1_mla.elf",
"bytes": 1293300552,
"sha256": "ab70250960c42767b046d40f9eb742cf179caf58129f93206fc12828863912ae"
},
{
"name": "token_embeddings.bf16",
"bytes": 103153664,
"sha256": "e5fc1c0be462527c6d9745b545dd3c50a606f0c0dc5b0028ae76edabfde93024"
},
{
"name": "tokenizer.json",
"bytes": 3583228,
"sha256": "6c8aaa9a542084f2457eab775d4eeb51f92a70c0fd9de28d5edb0ddec3c08d30"
}
],
"kind": "laya"
},
"multilingual": {
"title": "Laya multilingual",
"about": "The same questions in more than 100 languages, on a smaller encoder that answers about twice as fast.",
"card": {
"encoder": "mmBERT-base, 322M parameters",
"languages": "100+ languages",
"upstream": "convaiinnovations/laya",
"license": "Apache-2.0"
},
"path": "multilingual",
"checkpoint": "laya-multilingual",
"precision": "BF16",
"graphs": {
"128": "laya_s128_stage1_mla.elf",
"256": "laya_s256_stage1_mla.elf",
"512": "laya_s512_stage1_mla.elf",
"1024": "laya_s1024_stage1_mla.elf"
},
"latency_ms": {
"128": 8.1,
"256": 15.3,
"512": 33.9,
"1024": 116.5
},
"agreement": "100 / 100",
"files": [
{
"name": "act_tail.f32",
"bytes": 6152,
"sha256": "d5f5e87fa3d09251f018611c50930921725e005621dfb28120bd1d697cd3a239"
},
{
"name": "laya_config.json",
"bytes": 789,
"sha256": "2072042d5272df9655f7ac4580f0549347510534b864d8e7b3f092b1e75042ea"
},
{
"name": "laya_s1024_stage1_mla.elf",
"bytes": 926862376,
"sha256": "c6b1946d5caba586766e251a9d8077b5caebedea97dd04c8fc39da91c5092151"
},
{
"name": "laya_s128_stage1_mla.elf",
"bytes": 346199152,
"sha256": "1b23b62f06e7e4b43e5e900e61f675caad909cea1df28f8b70d6b3d412b64a30"
},
{
"name": "laya_s256_stage1_mla.elf",
"bytes": 376300056,
"sha256": "42cbc41a9b937768d30f2b74c48366d3f0fb6384300a85addbdcfaf77a06ea20"
},
{
"name": "laya_s512_stage1_mla.elf",
"bytes": 529239584,
"sha256": "f46adba6d5dc9a165911f8a6c30c6306d06440037e9a8d4cebb2b40e70723070"
},
{
"name": "token_embeddings.bf16",
"bytes": 393216000,
"sha256": "f084b1c925462a078cb82657a90378c6f8186a6a1a04747dfa4e2fe2cf5cfb49"
},
{
"name": "tokenizer.json",
"bytes": 34363188,
"sha256": "609d8f4c067cd3950f88594c5a802616cea245823836ef5848ee4fc40aab5b6f"
}
],
"kind": "laya"
},
"dino": {
"title": "Laya-dino",
"about": "The English model with its decision head fine-tuned to play the Dino Arena game: run, jump or duck.",
"card": {
"encoder": "ModernBERT-large, 421M parameters",
"languages": "English (game states)",
"upstream": "convaiinnovations/laya",
"license": "Apache-2.0"
},
"path": "dino",
"checkpoint": "laya-dino",
"precision": "BF16",
"graphs": {
"64": "laya_s64_stage1_mla.elf"
},
"latency_ms": {
"64": 18.0
},
"agreement": "30 / 30 game states",
"files": [
{
"name": "act_tail.f32",
"bytes": 6152,
"sha256": "d8d35135d97cac56479df92a7fe4a330dd46ed13e10dd7b8cca04963f23da3c6"
},
{
"name": "laya_config.json",
"bytes": 662,
"sha256": "c07488a9b3e114262251302460170ab782c1a5e4456a981286587ceb06f4a358"
},
{
"name": "laya_s64_stage1_mla.elf",
"bytes": 840672736,
"sha256": "71a07aa6c02d945b3b6c27b87e6f704d795d7346ffef513394780333ae5d8f34"
},
{
"name": "token_embeddings.bf16",
"bytes": 103153664,
"sha256": "ccdb9412fba7fb7f0e62a3123e5821477d4d359670545efd27807d9727236212"
},
{
"name": "tokenizer.json",
"bytes": 3583228,
"sha256": "6c8aaa9a542084f2457eab775d4eeb51f92a70c0fd9de28d5edb0ddec3c08d30"
}
],
"kind": "laya"
}
},
"models_v2": {
"clm": {
"title": "CLM v0.1 8B",
"about": "A Contrastive Language Model: a frozen Qwen3-8B reads the state and each option, and two small heads compare them. The same three kinds of question as Laya on an encoder twenty times the size; a new question takes about 0.3 s.",
"card": {
"encoder": "Qwen3-8B, 8.2B parameters, as a chain of 18 graphs",
"languages": "English",
"upstream": "Contrastive-LM/CLM-v0.1-8B",
"license": "Apache-2.0"
},
"path": "clm",
"kind": "clm",
"checkpoint": "CLM-v0.1-8B",
"precision": "A_BF16_W_INT8",
"graphs": {
"128": [
"qwen_s128_l00n2_stage1_mla.elf",
"qwen_s128_l02n2_stage1_mla.elf",
"qwen_s128_l04n2_stage1_mla.elf",
"qwen_s128_l06n2_stage1_mla.elf",
"qwen_s128_l08n2_stage1_mla.elf",
"qwen_s128_l10n2_stage1_mla.elf",
"qwen_s128_l12n2_stage1_mla.elf",
"qwen_s128_l14n2_stage1_mla.elf",
"qwen_s128_l16n2_stage1_mla.elf",
"qwen_s128_l18n2_stage1_mla.elf",
"qwen_s128_l20n2_stage1_mla.elf",
"qwen_s128_l22n2_stage1_mla.elf",
"qwen_s128_l24n2_stage1_mla.elf",
"qwen_s128_l26n2_stage1_mla.elf",
"qwen_s128_l28n2_stage1_mla.elf",
"qwen_s128_l30n2_stage1_mla.elf",
"qwen_s128_l32n2_stage1_mla.elf",
"qwen_s128_l34n2_stage1_mla.elf"
]
},
"latency_ms": {
"128": 254.0
},
"agreement": "9 of 13 reference decisions (int8 weights; an approximation)",
"files": [
{
"name": "clm_config.json",
"bytes": 1175,
"sha256": "db79dccca7fe2e1337dd2760930422cbf5cb1265a94e81e52a7a3553b7e17071"
},
{
"name": "heads.f32",
"bytes": 75550720,
"sha256": "bd3d83bfbb5c117dfd2991317bd8c60aafe3c22e00298e18a119c32993f03dae"
},
{
"name": "heads.json",
"bytes": 1387,
"sha256": "71bf7405610f7f1d990846b69a2c3162e6e6efa33c7e3bf66e5fc2e0caff35a6"
},
{
"name": "qwen_s128_l00n2_stage1_mla.elf",
"bytes": 419784304,
"sha256": "e63f23c745bcf493be1d6877eed1947daa157226989a255747b37a6ba972b7cd"
},
{
"name": "qwen_s128_l02n2_stage1_mla.elf",
"bytes": 398502840,
"sha256": "418ac276c2da2e83338c03a1ed9771657af70f882c8be3d1ee1f7d0122213ac7"
},
{
"name": "qwen_s128_l04n2_stage1_mla.elf",
"bytes": 432763576,
"sha256": "f8c1b07b2512a2092d632f96958da973dee5e6ab8a9c4cd367d2c2cc5012ad55"
},
{
"name": "qwen_s128_l06n2_stage1_mla.elf",
"bytes": 436585272,
"sha256": "cadc8a2cb77b973c3a98d1ef67f10e2fb0e56358a58030a14fb92516a28bcba4"
},
{
"name": "qwen_s128_l08n2_stage1_mla.elf",
"bytes": 436597992,
"sha256": "3f36a8031498b84c08f52d8e2dd4e57a6bfba1c249255c6881ba0841c71ce79b"
},
{
"name": "qwen_s128_l10n2_stage1_mla.elf",
"bytes": 436069896,
"sha256": "28fd3d49e817efc12a95df58f6f251f773c34131d8ef24506ea784db1b649f9a"
},
{
"name": "qwen_s128_l12n2_stage1_mla.elf",
"bytes": 434594824,
"sha256": "b46d516eb5c9a1a4c81a8fca699fc8d8ecee340edba732f3e608d2c2ee3c8e23"
},
{
"name": "qwen_s128_l14n2_stage1_mla.elf",
"bytes": 434365496,
"sha256": "34a58c88d6a33f9e2dc28651cba25e544230a6d77c8deef499ee176bbbe5edac"
},
{
"name": "qwen_s128_l16n2_stage1_mla.elf",
"bytes": 434663544,
"sha256": "845108ee3b20616a911a1b1e8f596542f7df4d1712f7b96995ca38d08f856f4e"
},
{
"name": "qwen_s128_l18n2_stage1_mla.elf",
"bytes": 433790200,
"sha256": "620e0011366756ba1b4121af890aa7e01c2a747ce763261733cec8604c9fcd27"
},
{
"name": "qwen_s128_l20n2_stage1_mla.elf",
"bytes": 434175896,
"sha256": "7c3cb32822095632ccfcfef58f7c8995dbb8ebf973d07991d70df659b14d8f6e"
},
{
"name": "qwen_s128_l22n2_stage1_mla.elf",
"bytes": 434193016,
"sha256": "85fa1ea6a318d085f0634d5452936e39faf7c5bce9ef17ea603def4c24e6c8eb"
},
{
"name": "qwen_s128_l24n2_stage1_mla.elf",
"bytes": 435984296,
"sha256": "eee531896cda0af5de1c8116fe1b5b4abc753d4f5d23c92c571221aff351b310"
},
{
"name": "qwen_s128_l26n2_stage1_mla.elf",
"bytes": 437245800,
"sha256": "aaa21d3e2cf3808338d88875750b2117f530c518703a368bde792970f8dfe939"
},
{
"name": "qwen_s128_l28n2_stage1_mla.elf",
"bytes": 437722296,
"sha256": "562f71307ba16d911bee0f04e4be22c9295d9a39a75c6225734bfe3a86073e90"
},
{
"name": "qwen_s128_l30n2_stage1_mla.elf",
"bytes": 438089400,
"sha256": "707c9f517325fc9b51e8c2394b5103d93cbee06d11bb3faf01c23d3082c30d04"
},
{
"name": "qwen_s128_l32n2_stage1_mla.elf",
"bytes": 437204936,
"sha256": "d222da9c0210cbf74128f0917c27dfc095c7cb5e35ff788070ad9f0997abd565"
},
{
"name": "qwen_s128_l34n2_stage1_mla.elf",
"bytes": 438373344,
"sha256": "b8bf4f53f9c5617291fc4849cd766007a1fa7035acebcfa141f78a8def9d5311"
},
{
"name": "token_embeddings.bf16",
"bytes": 1244659712,
"sha256": "458b4af1d22ed8d7d12235ed5249a4606c7606e35e775e7ea2244ba623b0312e"
},
{
"name": "tokenizer.json",
"bytes": 11422654,
"sha256": "aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4"
}
]
}
}
}