bed

README.md

@@ -11,5 +11,15 @@ pip uninstall -r uninstall.txt
 ```
 I had to uninstall some cuda stuff that torch installed to make this work.
 
+# References
+https://github.com/tloen/alpaca-lora
+https://huggingface.co/docs/transformers/main/en/model_doc/llama#llama
+https://huggingface.co/docs/transformers/index
+
+
+https://arxiv.org/pdf/1910.13461.pdf
+https://arxiv.org/pdf/1808.06226.pdf
+
+
 
 

llama_test.ipynb

@@ -18,12 +18,13 @@
    ],
    "source": [
     "import torch\n",
+    "import torch.nn as nn\n",
     "torch.cuda.is_available()"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 2,
    "metadata": {},
    "outputs": [
     {
@@ -38,7 +39,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "b687cc7da1a74a058775e5db887f0634",
+       "model_id": "37df56d6ddb747f3a91bc9100b33f47b",
        "version_major": 2,
        "version_minor": 0
       },
@@ -52,15 +53,50 @@
    ],
    "source": [
     "from transformers import LlamaTokenizer, LlamaForCausalLM\n",
-    "tokenizer = LlamaTokenizer.from_pretrained(\"decapoda-research/llama-7b-hf\", add_eos_token=True)\n",
-    "\n",
+    "tokenizer = LlamaTokenizer.from_pretrained(\n",
+    "    \"decapoda-research/llama-7b-hf\")\n",
+    "   \n",
+    "#tokenizer.pad_token_id = (0)\n",
+    "#tokenizer.padding_side = 'left'\n",
     "model = LlamaForCausalLM.from_pretrained(\n",
     "    \"decapoda-research/llama-7b-hf\",\n",
-    "    load_in_8bit=True,\n",
     "    device_map=\"auto\",\n",
     "    torch_dtype=torch.float16\n",
     ")\n"
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      " ⁇  hey dude, talk to me.\n",
+      "I'm a 20 year old guy from the UK. I'm a bit of a nerd, I like to read, I like to write, I like to play video games, I like to watch movies, I like to listen\n",
+      " ⁇  whats the capital of georgia?\n",
+      "What is the capital of Georgia?\n",
+      "The capital of Georgia is Atlanta.\n",
+      "What is the capital of Georgia?\n",
+      "The capital of Georgia is Atlanta. The capital of Georgia is Atlanta. The capital of Georgia is Atlanta. The capital of Georgia is Atlanta. The\n"
+     ]
+    }
+   ],
+   "source": [
+    "def ask(q,l=64):\n",
+    "    toks = tokenizer(q , return_tensors='pt')\n",
+    "    ctoks = toks.input_ids.to('cuda')\n",
+    "    gen = model.generate(ctoks, max_length=64)\n",
+    "    return tokenizer.decode(gen[0])\n",
+    "\n",
+    "r = ask('hey dude, talk to me')\n",
+    "print(r)\n",
+    "\n",
+    "r = ask('whats the capital of georgia?')\n",
+    "print(r)\n"
+   ]
   }
  ],
  "metadata": {