1 write to _hiddenSize
Microsoft.ML.GenAI.Core (1)
Module\Attention.cs (1)
104this._hiddenSize = hiddenSize;
8 references to _hiddenSize
Microsoft.ML.GenAI.Core (8)
Module\Attention.cs (8)
112Contract.Assert(this._hiddenSize % (this._headDim * this._numHeads) == 0, "hidden_size must be divisible by num_heads"); 114this.o_proj = new QuantizedLinear(this._hiddenSize, this._hiddenSize, hasBias: attentionBias, dtype: dtype); 118this.qkv_proj = new QuantizedLinear(this._hiddenSize, opSize, hasBias: attentionBias, dtype: dtype); 122this.q_proj = new QuantizedLinear(this._hiddenSize, this._numHeads * this._headDim, hasBias: attentionBias, dtype: dtype); 123this.k_proj = new QuantizedLinear(this._hiddenSize, this._numKeyValueHeads * this._headDim, hasBias: attentionBias, dtype: dtype); 124this.v_proj = new QuantizedLinear(this._hiddenSize, this._numKeyValueHeads * this._headDim, hasBias: attentionBias, dtype: dtype); 212attnOutput = attnOutput.reshape(bsz, qLen, this._hiddenSize);