{ "name": "ChatGBA", "architecture": "chatgba", "vocabSize": 256, "featureDim": 192, "numHeads": 6, "numKVHeads": 2, "headDim": 32, "ropeBase": 10000, "numBlocks": 10, "mlpHiddenDim": 512, "convKernel": 4, "contextLength": 512, "attentionWindows": [ 64, 64, 64, 64, 0, 64, 64, 64, 64, 0 ], "derived": { "parameterCount": 4003264, "queryHeadsPerKVHead": 3, "kvCacheBytesInt8": "2 global blocks x 512 positions x 128 + 8 windowed blocks x 64 positions x 128 = 196,608" }, "chat": { "user": "<|u|>", "assistant": "<|a|>", "end": "<|e|>", "endOfText": "<|endoftext|>", "encoding": "markers are plain UTF-8 text spelled out byte by byte; no reserved ids", "template": "<|u|>{user}<|e|><|a|>{assistant}<|e|>, repeated for each turn pair", "structure": "one or more user/assistant turn pairs, starting with the user; no system prompt", "generationPrompt": "the conversation so far followed by <|a|>; the reply ends at <|e|>" }, "weights": { "format": "self-describing GAB_ file: 12-byte preamble (magic GAB_, version, header length), JSON header, then little-endian float32 weights in this order", "matrixLayout": "row-major [in, out]; activations are multiplied as x @ W", "order": [ "tokenEmbeddings [vocabSize, featureDim] (tied: also the output projection, transposed)", "for each block: attentionNorm.gamma [1, featureDim], canonA.weight [convKernel, featureDim] (row j weighs the input j positions back)", " for each key/value head g: learnedK [featureDim, headDim], learnedV [featureDim, headDim], keyNorm.gamma [1, headDim]", " then for each of its numHeads / numKVHeads query heads: learnedQ [featureDim, headDim], learnedO [headDim, featureDim], queryNorm.gamma [1, headDim]", " mlpNorm.gamma [1, featureDim], canonC.weight [convKernel, featureDim], mlp.learnedGate [featureDim, mlpHiddenDim], mlp.learnedUp [featureDim, mlpHiddenDim], mlp.learnedDown [mlpHiddenDim, featureDim]", "finalNorm.gamma [1, featureDim]" ] } }