Prompt-tuning notebooks: suggest to use a smaller model for faster prototyping (#234)

README.md

@@ -61,8 +61,8 @@ You can also host [BLOOMZ](https://huggingface.co/bigscience/bloomz), a version
 Basic tutorials:
 
 - Getting started: [tutorial](https://colab.research.google.com/drive/1Ervk6HPNS6AYVr3xVdQnY5a-TjjmLCdQ?usp=sharing)
-- Fine-tune BLOOM to be a personified chatbot: [tutorial](https://colab.research.google.com/github/bigscience-workshop/petals/blob/main/examples/prompt-tuning-personachat.ipynb)
-- Fine-tune BLOOM for text semantic classification: [tutorial](https://colab.research.google.com/github/bigscience-workshop/petals/blob/main/examples/prompt-tuning-sst2.ipynb)
+- Prompt-tune BLOOM to create a personified chatbot: [tutorial](https://colab.research.google.com/github/bigscience-workshop/petals/blob/main/examples/prompt-tuning-personachat.ipynb)
+- Prompt-tune BLOOM for text semantic classification: [tutorial](https://colab.research.google.com/github/bigscience-workshop/petals/blob/main/examples/prompt-tuning-sst2.ipynb)
 
 Example apps built with Petals:
 

examples/prompt-tuning-personachat.ipynb

@@ -75,7 +75,18 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "MODEL_NAME = \"bigscience/bloom-petals\" # select model you like\n",
+    "# Choose a model you'd like to prompt-tune. We recommend starting with\n",
+    "# the smaller 7.1B version of BLOOM (bigscience/bloom-7b1-petals) for faster prototyping.\n",
+    "# Once your code is ready, you can switch to full-scale\n",
+    "# 176B-parameter BLOOM (bigscience/bloom-petals) or BLOOMZ (bigscience/bloomz-petals).\n",
+    "MODEL_NAME = \"bigscience/bloom-7b1-petals\"\n",
+    "\n",
+    "# Choose a prompt-tuning mode ('ptune' or 'deep_ptune').\n",
+    "# The latter fine-tunes separate prefixes for each transformer block,\n",
+    "# so prompt-tuning will take more time but yield better results.\n",
+    "# See this paper for details of how it works: https://arxiv.org/pdf/2110.07602.pdf\n",
+    "TUNING_MODE = 'ptune'\n",
+    "\n",
     "NUM_PREFIX_TOKENS = 16\n",
     "DEVICE = 'cuda'\n",
     "BATCH_SIZE = 8\n",
@@ -83,8 +94,7 @@
     "WEIGHT_DECAY = 0.0\n",
     "NUM_SAMPLES = 1000\n",
     "SEED = 42\n",
-    "MODEL_MAX_LENGTH = 256\n",
-    "TUNING_MODE = 'ptune' # choose between ['ptune', 'deep_ptune'] "
+    "MODEL_MAX_LENGTH = 256"
    ]
   },
   {

examples/prompt-tuning-sst2.ipynb

@@ -77,7 +77,18 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "MODEL_NAME = \"bigscience/bloom-petals\" # select model you like\n",
+    "# Choose a model you'd like to prompt-tune. We recommend starting with\n",
+    "# the smaller 7.1B version of BLOOM (bigscience/bloom-7b1-petals) for faster prototyping.\n",
+    "# Once your code is ready, you can switch to full-scale\n",
+    "# 176B-parameter BLOOM (bigscience/bloom-petals) or BLOOMZ (bigscience/bloomz-petals).\n",
+    "MODEL_NAME = \"bigscience/bloom-7b1-petals\"\n",
+    "\n",
+    "# Choose a prompt-tuning mode ('ptune' or 'deep_ptune').\n",
+    "# The latter fine-tunes separate prefixes for each transformer block,\n",
+    "# so prompt-tuning will take more time but yield better results.\n",
+    "# See this paper for details of how it works: https://arxiv.org/pdf/2110.07602.pdf\n",
+    "TUNING_MODE = 'ptune'\n",
+    "\n",
     "NUM_PREFIX_TOKENS = 16\n",
     "DEVICE = 'cuda'\n",
     "BATCH_SIZE = 16\n",
@@ -85,8 +96,7 @@
     "WEIGHT_DECAY = 0.0\n",
     "NUM_EPOCHS = 3\n",
     "SEED = 42\n",
-    "MODEL_MAX_LENGTH = 64\n",
-    "TUNING_MODE = 'ptune' # choose between ['ptune', 'deep_ptune'] "
+    "MODEL_MAX_LENGTH = 64"
    ]
   },
   {