1 write to _numKeyValueHeads
Microsoft.ML.GenAI.Phi (1)
Module\Phi2Attention.cs (1)
64this._numKeyValueHeads = config.NumKeyValueHeads ?? throw new ArgumentException("num_key_value_heads must be specified");
7 references to _numKeyValueHeads
Microsoft.ML.GenAI.Phi (7)
Module\Phi2Attention.cs (7)
65this._numKeyValueGroups = this._numAttentionHeads / this._numKeyValueHeads; 72this.k_proj = new GenAILinear(this._hiddenSize, this._numKeyValueHeads * this._headDim, hasBias: true, dtype: config.Dtype); 73this.v_proj = new GenAILinear(this._hiddenSize, this._numKeyValueHeads * this._headDim, hasBias: true, dtype: config.Dtype); 88this.cache_k = torch.zeros(maxBatch, this._numKeyValueHeads, maxLength, this._headDim, dtype: config.Dtype); 89this.cache_v = torch.zeros(maxBatch, this._numKeyValueHeads, maxLength, this._headDim, dtype: config.Dtype); 122keyStates = keyStates.view(batchSize, seqLen, this._numKeyValueHeads, this._headDim).transpose_(1, 2); 123valueStates = valueStates.view(batchSize, seqLen, this._numKeyValueHeads, this._headDim).transpose_(1, 2);