From 8083c9515ff68d5bc38d6a99c6435b25350f1ff4 Mon Sep 17 00:00:00 2001
From: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>
Date: Wed, 8 Feb 2023 18:54:46 +0530
Subject: [PATCH] update README and fix token_cls example

---
 README.md                                        | 10 ++++++++--
 .../peft_lora_token_cls.ipynb                    | 16 +++-------------
 2 files changed, 11 insertions(+), 15 deletions(-)

diff --git a/README.md b/README.md
index 290872d..765d2cc 100644
--- a/README.md
+++ b/README.md
@@ -127,6 +127,12 @@ Try out the 🤗 Gradio Space which should run seamlessly on a T4 instance:
 
 ### Parameter Efficient Tuning of LLMs for RLHF components such as Ranker and Policy [ToDo]
 
+### INT8 training of large models in Colab using PEFT LoRA and bits_and_bytes
+
+Here is now a demo on how to fine tune OPT-6.7b (14GB in fp16) in a Google colab: [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1jCkpikz0J2o20FBQmYmAGdiKmJGOMo-o?usp=sharing)
+
+Here is now a demo on how to fine tune wishper-large (1.5B params) (14GB in fp16) in a Google colab: [ToDo]
+
 ### Save compute and storage even for medium and small models
 
 Save storage by avoiding full finetuning of models on each of the downstream tasks/datasets,
@@ -307,12 +313,12 @@ any GPU memory savings. Please refer issue [[FSDP] FSDP with CPU offload consume
 
 2. When using `P_TUNING` or `PROMPT_TUNING` with `SEQ_2_SEQ` task, remember to remove the `num_virtual_token` virtual prompt predictions from the left side of the model outputs during evaluations. 
 
-3. `P_TUNING` or `PROMPT_TUNING` doesn't support `generate` functionality of transformers bcause `generate` strictly requires `input_ids`/`decoder_input_ids` but 
+3. For encoder-decoder models, `P_TUNING` or `PROMPT_TUNING` doesn't support `generate` functionality of transformers because `generate` strictly requires `decoder_input_ids` but 
 `P_TUNING`/`PROMPT_TUNING` appends soft prompt embeddings to `input_embeds` to create
 new `input_embeds` to be given to the model. Therefore, `generate` doesn't support this yet.
 
 ## Backlog:
-1. Explore and possibly integrate `(IA)^3` and `UniPELT`
+1. Explore and possibly integrate `(IA)^3`
 2. Add tests
 3. Add more use cases and examples
 
diff --git a/examples/token_classification/peft_lora_token_cls.ipynb b/examples/token_classification/peft_lora_token_cls.ipynb
index e070af2..8997649 100644
--- a/examples/token_classification/peft_lora_token_cls.ipynb
+++ b/examples/token_classification/peft_lora_token_cls.ipynb
@@ -827,7 +827,7 @@
     }
    ],
    "source": [
-    "from peft import get_peft_config, LoraModel, get_peft_model, LoraConfig, TaskType\n",
+    "from peft import get_peft_config, PeftModel, get_peft_model, LoraConfig, TaskType\n",
     "peft_config = LoraConfig(\n",
     "        task_type=TaskType.TOKEN_CLS,\n",
     "        inference_mode=False,\n",
@@ -1070,17 +1070,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from peft import get_peft_model_state_dict\n",
-    "to_return = get_peft_model_state_dict(model)\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 20,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "torch.save(to_return, \"layoutlm_funsd.pt\")"
+    "model.save_pretrained(\"peft_layoutlm\")\n"
    ]
   },
   {
@@ -1097,7 +1087,7 @@
     }
    ],
    "source": [
-    "!du -h \"layoutlm_funsd.pt\""
+    "!du -h \"peft_layoutlm/adapter_model.bin\""
    ]
   },
   {