sample prompt

FastChat/run.sh

@@ -1,4 +1,8 @@
-cd /data1/lzengaf/cs285/proj/minigrid/FastChat
 conda activate fastchat
-export CUDA_VISIBLE_DEVICES=0
-python3 -m fastchat.serve.cli --model-path lmsys/vicuna-7b-v1.5
+cd /data1/lzengaf/cs285/proj/minigrid/FastChat
+export CUDA_VISIBLE_DEVICES=
+
+python3 -m fastchat.serve.cli --model-path lmsys/vicuna-7b-v1.5
+
+
+"You are an agent in a Minigrid environment. Your mission is go to a green key in as few steps aspossible. Your agent's direction is currently right. Your agent can only see in front of itself. It cannot see blocked objects. Here is the vision of youagent. Assume your agent is at the center of the last row. The first row is the furthest from yourfront: unseen, unseen, unseen, unseen, unseen, unseen, unseenunseen, unseen, unseen, unseen, unseen, unseen, unseenunseen, wall, wall, wall, wall, wall, wall. unseen, yellow closed door, empty, empty, empty, empty, empty,unseen, wall, empty, empty, empty, empty, empty,unseen, wall, empty, empty, empty, grey box, emptyunseen, wall, empty, empty, empty, empty, empty. You have a set of skills at your disposal. They are listed in the following: Skill 1: Go to Obiect (in the same room) Skill 2: Open door (in the same room) Skill 3: Pickup an item (in the same room) Skill 4. Put an item next to an item (in the same room) Skill 5. Unlock a door (in the same room) Skill 6: Find an object (in a random room) Skill 7: Go to the green object (in a random n). Generate a probability vector for using each of the skills given the circumstance, in a comma:"

llama/dry_run.py

@@ -0,0 +1,85 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# This software may be used and distributed according to the terms of the Llama 2 Community License Agreement.
+
+from typing import List, Optional
+
+import fire
+
+from llama import Llama, Dialog
+
+
+def main(
+    ckpt_dir: str,
+    tokenizer_path: str,
+    temperature: float = 0.6,
+    top_p: float = 0.9,
+    max_seq_len: int = 512,
+    max_batch_size: int = 8,
+    max_gen_len: Optional[int] = None,
+):
+    """
+    Entry point of the program for generating text using a pretrained model.
+
+    Args:
+        ckpt_dir (str): The directory containing checkpoint files for the pretrained model.
+        tokenizer_path (str): The path to the tokenizer model used for text encoding/decoding.
+        temperature (float, optional): The temperature value for controlling randomness in generation.
+            Defaults to 0.6.
+        top_p (float, optional): The top-p sampling parameter for controlling diversity in generation.
+            Defaults to 0.9.
+        max_seq_len (int, optional): The maximum sequence length for input prompts. Defaults to 512.
+        max_batch_size (int, optional): The maximum batch size for generating sequences. Defaults to 8.
+        max_gen_len (int, optional): The maximum length of generated sequences. If None, it will be
+            set to the model's max sequence length. Defaults to None.
+    """
+    generator = Llama.build(
+        ckpt_dir=ckpt_dir,
+        tokenizer_path=tokenizer_path,
+        max_seq_len=max_seq_len,
+        max_batch_size=max_batch_size,
+    )
+
+    dialogs: List[Dialog] = [
+        [
+            # {
+            #     "role": "system",
+            #     "content": "Always answer with emojis",
+            # },
+            {
+                "role":
+                    "user",
+                "content":
+                    """You are an agent in a Minigrid environment. Your mission is go to a green key in as few steps aspossible.
+Your agent's direction is currently right.
+Your agent can only see in front of itself. It cannot see blocked objects. Here is the vision of youagent. Assume your agent is at the center of the last row. The first row is the furthest from yourfront:
+unseen, unseen, unseen, unseen, unseen, unseen, unseenunseen, unseen, unseen, unseen, unseen, unseen, unseenunseen, wall, wall, wall, wall, wall, wall.
+unseen, yellow closed door, empty, empty, empty, empty, empty,unseen, wall, empty, empty, empty, empty, empty,unseen, wall, empty, empty, empty, grey box, emptyunseen, wall, empty, empty, empty, empty, empty.
+You have a set of skills at your disposal. They are listed in the following:
+Skill 1: Go to Obiect (in the same room)
+Skill 2: Open door (in the same room)
+Skill 3: Pickup an item (in the same room)
+Skill 4. Put an item next to an item (in the same room)
+Skill 5. Unlock a door (in the same room)
+Skill 6: Find an object (in a random room)
+Skill 7: Go to the green object (in a random n).
+Generate a probability vector for using each of the skills given the circumstance, in a comma:
+             """
+            },
+        ],
+    ]
+    results = generator.chat_completion(
+        dialogs,  # type: ignore
+        max_gen_len=max_gen_len,
+        temperature=temperature,
+        top_p=top_p,
+    )
+
+    for dialog, result in zip(dialogs, results):
+        for msg in dialog:
+            print(f"{msg['role'].capitalize()}: {msg['content']}\n")
+        print(f"> {result['generation']['role'].capitalize()}: {result['generation']['content']}")
+        print("\n==================================\n")
+
+
+if __name__ == "__main__":
+    fire.Fire(main)

llama/run.sh

@@ -1,6 +1,13 @@
 # find -s somedir -type f -exec md5sum {} \; | md5sum
 
-cd /data1/lzengaf/cs285/proj/llama
-export CUDA_VISIBLE_DEVICES=0
+conda activate llama
+cd /data1/lzengaf/cs285/proj/minigrid/llama
+export CUDA_VISIBLE_DEVICES=
 
-torchrun --nproc_per_node 1 --master_port=29501 example_chat_completion.py    --ckpt_dir llama-2-7b-chat/     --tokenizer_path tokenizer.model     --max_seq_len 512 --max_batch_size 6
+# reproduce
+torchrun --nproc_per_node 1 --master_port=29501 example_chat_completion.py    --ckpt_dir llama-2-7b-chat/     --tokenizer_path tokenizer.model     --max_seq_len 512 --max_batch_size 6
+
+# dry run
+torchrun --nproc_per_node 1 --master_port=29501 dry_run.py    --ckpt_dir llama-2-7b-chat/     --tokenizer_path tokenizer.model     --max_seq_len 512 --max_batch_size 6
+
+torchrun --nproc_per_node 2 --master_port=29501 dry_run.py    --ckpt_dir llama-2-13b-chat/     --tokenizer_path tokenizer.model     --max_seq_len 512 --max_batch_size 6