update

Author: LittleLittleCloud

Extension.cs

@@ -1,25 +1,30 @@
+using Phi.Pipeline;
 using System.Text;
 using TorchSharp;
 using static TorchSharp.torch;
 
 public static class Extension
 {
     public static string Generate(
-        this Phi2ForCasualLM phi,
+        this CasualLMPipeline pipeline,
         string prompt,
         int maxLen = 128,
         float temperature = 0.7f,
         float topP = 0.9f,
-        string[]? stopSequences = null)
+        string[]? stopSequences = null,
+        string device = "cpu",
+        bool bos = true,
+        bool eos = false,
+        bool echo = false)
     {
-        var inputIds = phi.Tokenizer.Encode(prompt);
-        var inputTensor = torch.tensor(inputIds.ToArray(), dtype: ScalarType.Int64, device: phi.Device).unsqueeze(0);
+        var inputIds = pipeline.Tokenizer.Encode(prompt, bos, eos);
+        var inputTensor = torch.tensor(inputIds.ToArray(), dtype: ScalarType.Int64, device: device).unsqueeze(0);
         var attentionMask = torch.ones_like(inputTensor);
-        var stopTokenIds = stopSequences == null ? [[ phi.Tokenizer.EosId ]] : stopSequences.Select(x => phi.Tokenizer.Encode(x)).ToArray();
-        (var token, var _) = phi.Generate(inputTensor, attentionMask, temperature: temperature, maxLen: maxLen, topP: topP, stopTokenSequence: stopTokenIds);
+        var stopTokenIds = stopSequences == null ? [[ pipeline.Tokenizer.EosId ]] : stopSequences.Select(x => pipeline.Tokenizer.Encode(x, bos, eos)).ToArray();
+        (var token, var _) = pipeline.Generate(inputTensor, attentionMask, temperature: temperature, maxLen: maxLen, topP: topP, stopTokenSequence: stopTokenIds, echo: echo);
 
         var tokenIds = token[0].to_type(ScalarType.Int32).data<int>().ToArray();
-        var output = phi.Tokenizer.Decode(tokenIds);
+        var output = pipeline.Tokenizer.Decode(tokenIds);
         return output;
 
     }

Module/Phi3MLP.cs

@@ -18,7 +18,7 @@ public Phi3MLP(Phi3Config config)
     {
         this.gate_up_proj = torch.nn.Linear(config.HiddenSize, 2 * config.IntermediateSize, hasBias: false, dtype: config.DType);
         this.down_proj = torch.nn.Linear(config.IntermediateSize, config.HiddenSize, hasBias: false, dtype: config.DType);
-        this.activation_fn = new NewGELUActivation();
+        this.activation_fn = Utils.GetActivation(config.HiddenAct);
     }
     public override Tensor forward(Tensor input)
     {

Module/Phi3Model.cs

@@ -55,12 +55,12 @@ public CasualLMModelOutput(
     {
         this.last_hidden_state = last_hidden_state;
         this.hidden_states = hidden_states;
-        this.legits = legits;
+        this.logits = legits;
         this.attentions = attentions;
         this.past_key_values = past_key_values;
     }
 
-    public Tensor legits { get; set; }
+    public Tensor logits { get; set; }
 
     public Tensor last_hidden_state { get; set; }