Model and Inference script

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+model_Uint8.sentis filter=lfs diff=lfs merge=lfs -text

Phi3Claude.cs

@@ -0,0 +1,134 @@
+using UnityEngine;
+using Microsoft.ML.Tokenizers;
+using Unity.Sentis;
+using System.IO;
+using System.Linq;
+using System.Collections.Generic;
+using System.Collections;
+
+public class Phi3Claude : MonoBehaviour
+{
+    IWorker worker;
+    LlamaTokenizer tokenizer;
+
+    List<int> tokens = new();
+    TensorInt inputTensor, attentionMaskTensor, positionIdsTensor;
+    TensorFloat outputLogits;
+
+    int maxTokens = 100; // Maximum number of tokens to generate
+    List<int> eosTokens; // End of sequence tokens
+
+    private IBackend backend;
+
+    private void Start()
+    {
+        var tokenizerModelPath = Path.Combine(Application.streamingAssetsPath, "Phi35/tokenizer.model");
+        var sentisModelPath = Path.Combine(Application.streamingAssetsPath, "Phi35/model_Uint8.sentis");
+        var configPath = Path.Combine(Application.streamingAssetsPath, "Phi35/generation_config.json");
+        
+        var model = ModelLoader.Load(sentisModelPath);
+        
+        worker = WorkerFactory.CreateWorker(BackendType.GPUCompute, model);
+        Dictionary<string, int> specialTokens = TokenizerUtils.LoadSpecialTokens(Path.Combine(Application.streamingAssetsPath, "Phi35/added_tokens.json"));
+
+        using (Stream tokenizerModelStream = new FileStream(tokenizerModelPath, FileMode.Open, FileAccess.Read))
+        {
+            tokenizer = LlamaTokenizer.Create(
+                tokenizerModelStream,
+                addBeginOfSentence: true,
+                addEndOfSentence: false,
+                specialTokens: specialTokens
+            );
+        }
+
+        eosTokens = TokenizerUtils.IdentifyEOSTokens(configPath);
+        backend = WorkerFactory.CreateBackend(BackendType.GPUCompute);
+
+        Generate("Hello, how is your day?");
+    }
+
+    public void Generate(string userPrompt, string systemPrompt = "You are a helpful assistant.") 
+    {
+        string completePrompt = Phi3InputFormatter.FormatChatInput(systemPrompt, userPrompt);
+        Debug.Log("Complete prompt : " + completePrompt);
+
+        int[] inputIds = tokenizer.EncodeToIds(completePrompt).ToArray();
+        Debug.Log($"Tokenized input: [{string.Join(", ", inputIds)}]");
+        Debug.Log($"Decoded tokens: [{string.Join(", ", tokenizer.Decode(inputIds, true))}]");
+
+        tokens.Clear();
+        tokens.AddRange(inputIds);
+
+        StartCoroutine(GenerateSequence());
+    }
+    
+    private IEnumerator GenerateSequence()
+    {
+        for (int i = 0; i < maxTokens; i++)
+        {
+            RefreshTensors(tokens.ToArray());
+
+            worker.Execute(new Dictionary<string, Tensor>()
+            {
+                {"input_ids", inputTensor},
+                {"attention_mask", attentionMaskTensor},
+                {"position_ids", positionIdsTensor}
+            }); // > 15ms (/!\ should be async)
+
+            outputLogits = worker.PeekOutput("logits") as TensorFloat; // Async
+            outputLogits.ReadbackRequest(); // Async
+
+            yield return outputLogits.IsReadbackRequestDone(); // 236 ms
+
+            tokens.Add(ProcessLogits()); // > 200ms
+
+            int nextToken = tokens[tokens.Count - 1];
+
+            CleanupTensors();
+
+            if (eosTokens.Contains(nextToken))
+                break;
+        }
+
+        string generatedText = tokenizer.Decode(tokens.ToArray(), true); // 0 ms
+        Debug.Log($"Generated sequence: {generatedText}");
+    }
+
+
+    private int ProcessLogits()
+    {
+        // Greedy sampling for simplicity
+        using var argMaxTensor = TensorInt.AllocNoData(new TensorShape(1, outputLogits.shape[1]));
+        backend.ArgMax(outputLogits, argMaxTensor, axis: 2, selectLastIndex: false);
+
+        var argMaxTensorArray = argMaxTensor.ToReadOnlyArray(); // TODO : investigate on why it's long to process
+        int nextToken = argMaxTensorArray[outputLogits.shape[1] - 1];
+
+        Debug.Log($"<color=orange>Next token: [ID = {nextToken}, STR = \"{tokenizer.Decode(new[] { nextToken }, true)}\"]</color>");
+
+        return nextToken;
+    }
+
+    private void RefreshTensors(int[] ids) 
+    {
+        // Update input tensors with the full context
+        inputTensor = new TensorInt(new TensorShape(1, ids.Length), ids);
+        attentionMaskTensor = new TensorInt(new TensorShape(1, ids.Length), Enumerable.Repeat(1, ids.Length).ToArray());
+        positionIdsTensor = new TensorInt(new TensorShape(1, ids.Length), Enumerable.Range(0, ids.Length).ToArray());
+    }
+
+    private void CleanupTensors()
+    {
+        inputTensor?.Dispose();
+        attentionMaskTensor?.Dispose();
+        positionIdsTensor?.Dispose();
+        outputLogits?.Dispose();
+    }
+
+    private void OnDestroy() {
+        CleanupTensors();
+        
+        worker?.Dispose();
+        backend?.Dispose();
+    }
+}

added_tokens.json

@@ -0,0 +1,13 @@
+{
+  "<|assistant|>": 32001,
+  "<|endoftext|>": 32000,
+  "<|end|>": 32007,
+  "<|placeholder1|>": 32002,
+  "<|placeholder2|>": 32003,
+  "<|placeholder3|>": 32004,
+  "<|placeholder4|>": 32005,
+  "<|placeholder5|>": 32008,
+  "<|placeholder6|>": 32009,
+  "<|system|>": 32006,
+  "<|user|>": 32010
+}

config.json

@@ -0,0 +1,137 @@
+{
+  "_name_or_path": "microsoft/Phi-3.5-mini-instruct",
+  "architectures": [
+    "Phi3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "microsoft/Phi-3.5-mini-instruct--configuration_phi3.Phi3Config",
+    "AutoModelForCausalLM": "microsoft/Phi-3.5-mini-instruct--modeling_phi3.Phi3ForCausalLM"
+  },
+  "bos_token_id": 1,
+  "embd_pdrop": 0.0,
+  "eos_token_id": 32000,
+  "hidden_act": "silu",
+  "hidden_size": 3072,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 131072,
+  "model_type": "phi3",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "original_max_position_embeddings": 4096,
+  "pad_token_id": 32000,
+  "resid_pdrop": 0.0,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": {
+    "long_factor": [
+      1.0800000429153442,
+      1.1100000143051147,
+      1.1399999856948853,
+      1.340000033378601,
+      1.5899999141693115,
+      1.600000023841858,
+      1.6200000047683716,
+      2.620000123977661,
+      3.2300000190734863,
+      3.2300000190734863,
+      4.789999961853027,
+      7.400000095367432,
+      7.700000286102295,
+      9.09000015258789,
+      12.199999809265137,
+      17.670000076293945,
+      24.46000099182129,
+      28.57000160217285,
+      30.420001983642578,
+      30.840002059936523,
+      32.590003967285156,
+      32.93000411987305,
+      42.320003509521484,
+      44.96000289916992,
+      50.340003967285156,
+      50.45000457763672,
+      57.55000305175781,
+      57.93000411987305,
+      58.21000289916992,
+      60.1400032043457,
+      62.61000442504883,
+      62.62000274658203,
+      62.71000289916992,
+      63.1400032043457,
+      63.1400032043457,
+      63.77000427246094,
+      63.93000411987305,
+      63.96000289916992,
+      63.970001220703125,
+      64.02999877929688,
+      64.06999969482422,
+      64.08000183105469,
+      64.12000274658203,
+      64.41000366210938,
+      64.4800033569336,
+      64.51000213623047,
+      64.52999877929688,
+      64.83999633789062
+    ],
+    "short_factor": [
+      1.0,
+      1.0199999809265137,
+      1.0299999713897705,
+      1.0299999713897705,
+      1.0499999523162842,
+      1.0499999523162842,
+      1.0499999523162842,
+      1.0499999523162842,
+      1.0499999523162842,
+      1.0699999332427979,
+      1.0999999046325684,
+      1.1099998950958252,
+      1.1599998474121094,
+      1.1599998474121094,
+      1.1699998378753662,
+      1.2899998426437378,
+      1.339999794960022,
+      1.679999828338623,
+      1.7899998426437378,
+      1.8199998140335083,
+      1.8499997854232788,
+      1.8799997568130493,
+      1.9099997282028198,
+      1.9399996995925903,
+      1.9899996519088745,
+      2.0199997425079346,
+      2.0199997425079346,
+      2.0199997425079346,
+      2.0199997425079346,
+      2.0199997425079346,
+      2.0199997425079346,
+      2.0299997329711914,
+      2.0299997329711914,
+      2.0299997329711914,
+      2.0299997329711914,
+      2.0299997329711914,
+      2.0299997329711914,
+      2.0299997329711914,
+      2.0299997329711914,
+      2.0299997329711914,
+      2.0799996852874756,
+      2.0899996757507324,
+      2.189999580383301,
+      2.2199995517730713,
+      2.5899994373321533,
+      2.729999542236328,
+      2.749999523162842,
+      2.8399994373321533
+    ],
+    "type": "longrope"
+  },
+  "rope_theta": 10000.0,
+  "sliding_window": 262144,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.43.3",
+  "use_cache": false,
+  "vocab_size": 32064
+}

generation_config.json

@@ -0,0 +1,11 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": [
+    32007,
+    32001,
+    32000
+  ],
+  "pad_token_id": 32000,
+  "transformers_version": "4.43.3"
+}

model_Uint8.sentis

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7f68876f299ef76ecf78157b5c11410dfd637eb3b841b27106218292d204c6c
+size 3822109304

special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

tokenizer_config.json

@@ -0,0 +1,131 @@
+{
+  "add_bos_token": false,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": false
+    },
+    "32000": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32001": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32002": {
+      "content": "<|placeholder1|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32003": {
+      "content": "<|placeholder2|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32004": {
+      "content": "<|placeholder3|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32005": {
+      "content": "<|placeholder4|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32006": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32007": {
+      "content": "<|end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32008": {
+      "content": "<|placeholder5|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32009": {
+      "content": "<|placeholder6|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32010": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "chat_template": "{% for message in messages %}{% if message['role'] == 'system' and message['content'] %}{{'<|system|>\n' + message['content'] + '<|end|>\n'}}{% elif message['role'] == 'user' %}{{'<|user|>\n' + message['content'] + '<|end|>\n'}}{% elif message['role'] == 'assistant' %}{{'<|assistant|>\n' + message['content'] + '<|end|>\n'}}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<|assistant|>\n' }}{% else %}{{ eos_token }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "legacy": false,
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "left",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}