Update You_can_still_run_it_directly_on_your_own_machine_if_it's_too_small.ipynb

You_can_still_run_it_directly_on_your_own_machine_if_it's_too_small.ipynb

@@ -3,21 +3,15 @@
     {
       "cell_type": "code",
       "execution_count": null,
-      "metadata": {
-        "id": "OEKVU8aAXeQx"
-      },
+      "metadata": {"id": "OEKVU8aAXeQx"},
       "outputs": [],
-      "source": [
-        "REPO_ID = \"mondk/msh-tiny.For-Test-In-Note-Book-Google-Colab\"\n"
-      ],
+      "source": ["REPO_ID = \"mondk/msh-tiny.For-Test-In-Note-Book-Google-Colab\"\n"],
       "id": "OEKVU8aAXeQx"
     },
     {
       "cell_type": "code",
       "execution_count": null,
-      "metadata": {
-        "id": "Muc0ftk9XeQz"
-      },
+      "metadata": {"id": "Muc0ftk9XeQz"},
       "outputs": [],
       "source": [
         "!pip install -q -U huggingface_hub tokenizers\n",
@@ -30,9 +24,7 @@
     {
       "cell_type": "code",
       "execution_count": null,
-      "metadata": {
-        "id": "h5Lf-nDoXeQ0"
-      },
+      "metadata": {"id": "h5Lf-nDoXeQ0"},
       "outputs": [],
       "source": [
         "from huggingface_hub import hf_hub_download\n",
@@ -49,9 +41,7 @@
     {
       "cell_type": "code",
       "execution_count": null,
-      "metadata": {
-        "id": "rhTG3mtsXeQ0"
-      },
+      "metadata": {"id": "rhTG3mtsXeQ0"},
       "outputs": [],
       "source": [
         "import math\n",
@@ -123,7 +113,7 @@
         "        self.blocks = nn.ModuleList([Block(n_embd, n_head, block_size, dropout) for _ in range(n_layer)])\n",
         "        self.ln_f = nn.LayerNorm(n_embd)\n",
         "        self.head = nn.Linear(n_embd, vocab_size, bias=False)\n",
-        "        self.head.weight = self.tok_emb.weight  # weight tying\n",
+        "        self.head.weight = self.tok_emb.weight\n",
         "\n",
         "    def forward(self, idx, targets=None):\n",
         "        B, T = idx.shape\n",
@@ -158,9 +148,7 @@
     {
       "cell_type": "code",
       "execution_count": null,
-      "metadata": {
-        "id": "PP6KcuhQXeQ1"
-      },
+      "metadata": {"id": "PP6KcuhQXeQ1"},
       "outputs": [],
       "source": [
         "import json\n",
@@ -196,21 +184,13 @@
     {
       "cell_type": "code",
       "execution_count": null,
-      "metadata": {
-        "id": "fix1chatFn"
-      },
+      "metadata": {"id": "fix1chatFn"},
       "outputs": [],
       "source": [
-        "# --- FIX: ham chat() bi thieu, day la ham duoc them lai ---\n",
-        "# Encode/decode qua tokenizers backend, va tu sinh token cho den khi\n",
-        "# gap END_ID (thay vi dung model.generate() goc, vi ham do khong biet\n",
-        "# dung lai o END_TOK va se tra ve ca prompt lan phan sinh ra).\n",
-        "\n",
         "@torch.no_grad()\n",
         "def chat(user_input, max_new_tokens=200, temperature=0.8, top_k=40):\n",
         "    prompt = f\"{USER_TOK}{user_input}{ASSISTANT_TOK}\"\n",
         "    prompt_ids = tokenizer_backend.encode(prompt).ids\n",
-        "    # cat bot neu prompt dai hon block_size\n",
         "    prompt_ids = prompt_ids[-model.block_size:]\n",
         "    idx = torch.tensor([prompt_ids], dtype=torch.long, device=device)\n",
         "\n",
@@ -239,9 +219,7 @@
     {
       "cell_type": "code",
       "execution_count": null,
-      "metadata": {
-        "id": "u1CRQJL2XeQ2"
-      },
+      "metadata": {"id": "u1CRQJL2XeQ2"},
       "outputs": [],
       "source": [
         "print(\"Chat started. Type 'exit' to quit.\\n\")\n",
@@ -258,18 +236,10 @@
   ],
   "metadata": {
     "accelerator": "GPU",
-    "colab": {
-      "provenance": [],
-      "gpuType": "T4"
-    },
-    "kernelspec": {
-      "display_name": "Python 3",
-      "name": "python3"
-    },
-    "language_info": {
-      "name": "python"
-    }
+    "colab": {"provenance": [], "gpuType": "T4"},
+    "kernelspec": {"display_name": "Python 3", "name": "python3"},
+    "language_info": {"name": "python"}
   },
   "nbformat": 4,
   "nbformat_minor": 5
-}
+}