-
Notifications
You must be signed in to change notification settings - Fork 581
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Addition of types #493
Draft
huyinan
wants to merge
10
commits into
camel-ai:master
Choose a base branch
from
huyinan:master
base: master
Could not load branches
Branch not found: {{ refName }}
Could not load tags
Nothing to show
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Draft
Addition of types #493
Changes from all commits
Commits
Show all changes
10 commits
Select commit
Hold shift + click to select a range
7d9e175
Addition of types
huyinan 8d9af9d
Merge branch 'master' into master
huyinan b3b5eda
Mimic example of a react agent from deductive agent
huyinan bcdcf97
Update test_react_embodied_agent.py
huyinan 8eaf8e3
Merge branch 'master' into master
huyinan d6ebb2d
Update react_embodied_agent.py
huyinan 30c5056
Merge branch 'master' of https://github.com/huyinan/camel
huyinan 2174e56
Adopting LLF-BENCH AS EXAMPLES
huyinan 8a4ea2a
implentation of llf-bench as a prototype example for REACT in reinfor…
huyinan 6c8544b
test case
huyinan File filter
Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,263 @@ | ||
# =========== Copyright 2023 @ CAMEL-AI.org. All Rights Reserved. =========== | ||
# Licensed under the Apache License, Version 2.0 (the “License”); | ||
# you may not use this file except in compliance with the License. | ||
# You may obtain a copy of the License at | ||
# | ||
# http://www.apache.org/licenses/LICENSE-2.0 | ||
# | ||
# Unless required by applicable law or agreed to in writing, software | ||
# distributed under the License is distributed on an “AS IS” BASIS, | ||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
# See the License for the specific language governing permissions and | ||
# limitations under the License. | ||
# =========== Copyright 2023 @ CAMEL-AI.org. All Rights Reserved. =========== | ||
import re | ||
from typing import Dict, List, Optional, Union | ||
import json | ||
import sys | ||
import os | ||
import camel.llfbench as gym | ||
import random | ||
from camel.agents import ChatAgent | ||
from camel.configs import BaseConfig | ||
from camel.messages import BaseMessage | ||
from camel.prompts import TextPrompt | ||
from camel.types import ModelType, RoleType, ReasonType | ||
#from camel.functions.search_functions import * | ||
from camel.envs import bandits, optimization, poem | ||
|
||
|
||
class ReactAgent(ChatAgent): | ||
r"""An agent responsible for reactive reasoning. Model of reactive reasoning: | ||
- Thought. The agent's explanation on how to answer the question, illustrating what tools to implement. | ||
- Action. The calling of functions based on the thought. This step includes what the input of the function is and what functions to call. | ||
- Observation. Summarizing the output of the function. | ||
|
||
Args: | ||
model_type (ModelType, optional): The type of model to use for the | ||
agent. (default: :obj: `None`) | ||
model_config (BaseConfig, optional): The configuration for the model. | ||
(default: :obj:`None`) | ||
""" | ||
def __init__( | ||
self, | ||
model_type: Optional[ModelType] = None, | ||
model_config: Optional[BaseConfig] = None, | ||
) -> None: | ||
system_message = BaseMessage( | ||
role_name="React Agent", | ||
role_type=RoleType.ASSISTANT, | ||
meta_dict=None, | ||
content="You assign roles based on tasks.", | ||
# reason_type=ReasonType.REACT, | ||
) | ||
super().__init__(system_message, model_type, model_config) | ||
|
||
def re_act_reasoning( | ||
self, | ||
role_descriptions_dict: Optional[Dict[str, str]] = None, | ||
) -> Dict[str, Union[List[str], Dict[str, str]]]: | ||
r"""Derives the conditions and quality from the starting state and the | ||
target state based on the model of the deductive reasoning and the | ||
knowledge base. It can optionally consider the roles involved in the | ||
scenario, which allows tailoring the output more closely to the AI | ||
agent's environment. | ||
|
||
Args: | ||
role_descriptions_dict (Optional[Dict[str, str]], optional): The | ||
descriptions of the roles. (default: :obj:`None`) | ||
role_descriptions_dict (Optional[Dict[str, str]], optional): A | ||
Comment on lines
+67
to
+69
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. repetitive docstring here |
||
dictionary describing the roles involved in the scenario. This | ||
is optional and can be used to provide a context for the | ||
CAMEL's role-playing, enabling the generation of more relevant | ||
and tailored conditions and quality assessments. This could be | ||
generated using a `RoleAssignmentAgent()` or defined manually | ||
by the user. | ||
|
||
Returns: | ||
Dict[str, Union[List[str], Dict[str, str]]]: A dictionary with the | ||
extracted data from the message. The dictionary contains three | ||
keys: | ||
- 'Thought': A list where each key is a condition ID and | ||
each value is the corresponding condition text. | ||
- 'Action': A list of label strings extracted from the message. | ||
- 'Observation': A string of quality assessment strings extracted | ||
from the message. | ||
""" | ||
self.reset() | ||
|
||
react_instructions = """You are an reactive reasoner. You solve a question answering task with interleaving Thought, Action, Observation steps. The Thought part can reason about the current situation, explain the thinking process, and deciding which tools to implement. | ||
The Action part implements the tool. The observation part collects and summarizes the output of the function called in the Action part. In summary, the response contains three parts | ||
(1) Thought:\n <BLANK>. Because our thoughts contain {keywords}, we decide to implement the {entity}. | ||
(2) Action: \n We implement the {entity}. We summarize the output of {entity}. | ||
(3) Observation: <BLANK> | ||
|
||
Here are a few examples regarding the tool of {entity}: | ||
{role_with_description_prompt} | ||
""" # noqa: E501 | ||
|
||
# write a wrapper beyond the current existing wrappers. | ||
# After deciding the factor. First present corresponding prompts for llm to brush up. | ||
react_prompt = react_instructions | ||
# add the prompts stored in | ||
# idx_dict = ['reco', 'poem', 'optimization', 'bandits'] | ||
# for i in range(4): | ||
# prompt_path = os.path.join('./camel/envs/', idx_dict[i], 'prompts.py') | ||
# f = open(prompt_path) | ||
# examples = f.readlines() | ||
# react_examples = TextPrompt(examples) | ||
# react_prompt = react_prompt + react_examples | ||
|
||
# Now create some examples for each task and put them into the prompt | ||
# For the moment, we only support the tool agents included in llf-bench. | ||
# We implement a two-step language model, where the first steps decide what tool agent in llf-bench to use. The second | ||
# step implements the tool agent in llf-bench, noticing that it is a language-based reinforcement learning model. | ||
react_optimization_example = """ | ||
========Optimization Tool Agent =============== | ||
Question: I would like to minimize the {func_name} function. | ||
Thought: Based on the keyword optimization, we could implement the optimization tool agent concerning {func_name} with limited information. | ||
Question: I would like to minimize a Booth function without knowing its the exact expression. | ||
Thought: Based on the keyword optimization, we could implement the optimization tool agent concerning Booth with limited information. | ||
Question: I would like to find a minimizer of Booth function without knowing its the exact expression. | ||
Thought: Based on the keyword optimization, we could implement the optimization tool agent concerning Booth with limited information. | ||
""" | ||
react_bandit_example = "" | ||
#react_bandit_example = """ | ||
#========Bandit Problem Tool Agent =============== | ||
#Question: I would like to do a bandit problem. | ||
#Thought: Based on the keywords in your request, we could implement the bandit tool agent. | ||
#Question: I want to apply a multi-armed bandit problem. | ||
#Thought: Based on the keywords in your request, we could implement the bandit tool agent. | ||
#""" | ||
|
||
react_poem_example = "" | ||
|
||
#react_poem_example = """ | ||
#========Poem Composing Tool Agent =============== | ||
#Question: I would like to write a poem. | ||
#Thought: Based on the keywords in your request, we could implement the poem composing tool agent. | ||
#Question: I would like to write something that has rhymes. | ||
#Thought: Based on the keywords in your request, we could implement the poem composing tool agent. | ||
#""" | ||
|
||
react_prompt = react_prompt + react_poem_example + react_bandit_example + react_optimization_example | ||
if role_descriptions_dict is not None: | ||
role_names = role_descriptions_dict.keys() | ||
role_with_description_prompt = \ | ||
"===== ROLES WITH DESCRIPTIONS =====\n" + "\n".join( | ||
f"{role_name}:\n{role_descriptions_dict[role_name]}\n" | ||
for role_name in role_names) + "\n\n" | ||
else: | ||
role_with_description_prompt = "" | ||
|
||
react_reasoning = react_prompt.format( | ||
role_with_description_prompt=role_with_description_prompt) | ||
|
||
react_reasoning_msg = \ | ||
BaseMessage.make_user_message(role_name="Reactive Reasoner", | ||
content=react_reasoning) | ||
|
||
response = self.step( | ||
input_message=react_reasoning_msg) | ||
|
||
if response.terminated: | ||
raise RuntimeError("Reasoning failed. Error:\n" + | ||
f"{response.info}") | ||
thought_action: BaseMessage = response.msg | ||
print(f"Message content:\n{thought_action.content}") | ||
ENVIRONMENTS_OPTIMIZATION = ( | ||
'Booth', | ||
'McCormick', | ||
'Rosenbrock', | ||
'SixHumpCamel', | ||
) | ||
name_func = 'optimization' | ||
|
||
if 'Booth' in ENVIRONMENTS_OPTIMIZATION: | ||
name_attribute = 'Booth' | ||
elif 'Rosenbrock' in ENVIRONMENTS_OPTIMIZATION: | ||
name_attribute = 'Rosenbrock' | ||
elif 'McCormick' in ENVIRONMENTS_OPTIMIZATION: | ||
name_attribute = 'McCormick' | ||
elif 'SixHumpCamel' in ENVIRONMENTS_OPTIMIZATION: | ||
name_attribute = 'SixHumpCamel' | ||
|
||
|
||
# ENVIRONMENTS_BANDIT = ( | ||
# 'BanditTenArmedRandomFixed-v0', | ||
# 'BanditTenArmedRandomRandom-v0', | ||
# 'BanditTenArmedGaussian-v0', | ||
# 'BanditTenArmedUniformDistributedReward-v0', | ||
# 'BanditTwoArmedDeterministicFixed-v0', | ||
# 'BanditTwoArmedHighHighFixed-v0', | ||
# 'BanditTwoArmedHighLowFixed-v0', | ||
# 'BanditTwoArmedLowLowFixed-v0', | ||
# ) | ||
# | ||
# ENVIRONMENTS_POEM = ( | ||
# 'Haiku', | ||
# 'Tanka', | ||
# 'LineSyllableConstrainedPoem', | ||
# 'SyllableConstrainedPoem', | ||
# ) | ||
# if 'bandits' in thought_action.content: | ||
# name_func = 'bandits' | ||
# l = len(ENVIRONMENTS_BANDIT) | ||
# idx = random.randint(0, l) | ||
# name_attribute = ENVIRONMENTS_OPTIMIZATION[idx] | ||
# elif 'optimization' in thought_action.content: | ||
# name_func = 'optimization' | ||
# l = len(ENVIRONMENTS_OPTIMIZATION) | ||
# idx = random.randint(0, l) | ||
# name_attribute = ENVIRONMENTS_OPTIMIZATION[idx] | ||
# elif 'poem' in thought_action.content: | ||
# name_func = 'poem' | ||
# l = len(ENVIRONMENTS_POEM) | ||
# idx = random.randint(0, l) | ||
# name_attribute = ENVIRONMENTS_POEM[idx] | ||
# Write a wrapper | ||
environment_name = 'llf-' + name_func + name_attribute + '-v0' | ||
env = gym.make(environment_name) | ||
|
||
# This part borrows the test program from the llf-bench repo | ||
done = False | ||
cumulative_reward = 0.0 | ||
observation, info = env.reset() | ||
if observation['observation'] == None: | ||
observation['observation'] = '' | ||
|
||
observation['feedback'] = '' | ||
|
||
while not done: | ||
# Observation is dict having 'observation', 'instruction', 'feedback' | ||
# Here we print the observation and ask the user for an action | ||
|
||
action = input(observation['observation'] + '\n' + | ||
observation['instruction'] + '\n' + | ||
observation['feedback'] + '\n' + | ||
'Action: ') | ||
|
||
# Gridworld has a text action space, so TextWrapper is not needed | ||
# to parse a valid action from the input string | ||
# pdb.set_trace() | ||
# action = int(action) | ||
observation, reward, terminated, truncated, info = env.step(action) | ||
|
||
if isinstance(observation['observation'], str) == False: | ||
observation['observation'] = '' | ||
if isinstance(observation['instruction'], str) == False: | ||
observation['instruction'] = '' | ||
if isinstance(observation['feedback'], str) == False: | ||
observation['feedback'] = '' | ||
# reward is never revealed to the agent; only used for evaluation | ||
|
||
cumulative_reward += reward | ||
|
||
# terminated and truncated follow the same semantics as in Gymnasium | ||
|
||
done = terminated or truncated | ||
|
||
print(f'Episode reward: {cumulative_reward}') | ||
# Leave the following part in test cases: extracting the conditions from the message and print. | ||
|
||
return observation |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
|
@@ -17,7 +17,7 @@ | |
from camel.agents import ChatAgent | ||
from camel.messages import BaseMessage | ||
from camel.prompts import TextPrompt | ||
from camel.types import ModelType, RoleType | ||
from camel.types import ModelType, RoleType, ReasonType | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. why added |
||
|
||
|
||
class RoleAssignmentAgent(ChatAgent): | ||
|
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,13 @@ | ||
import importlib | ||
from camel.envs import gridworld | ||
from camel.envs import bandits | ||
from camel.envs import optimization | ||
from camel.envs import reco | ||
from camel.envs import poem | ||
from camel.envs import highway | ||
|
||
if importlib.util.find_spec('metaworld'): | ||
from camel.envs import metaworld | ||
|
||
if importlib.util.find_spec('alfworld'): | ||
from camel.envs import alfworld |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,29 @@ | ||
from gymnasium.envs.registration import register | ||
from llfbench.envs.alfworld.alfworld import Alfworld | ||
from llfbench.envs.alfworld.wrapper import AlfworldWrapper | ||
|
||
|
||
|
||
ENVIRONMENTS = ( | ||
'alfworld-v0', | ||
) | ||
|
||
|
||
def make_env(env_name, | ||
instruction_type='b', | ||
feedback_type='r', | ||
): | ||
|
||
""" Make the original env and wrap it with the LLFWrapper. """ | ||
assert env_name.startswith("alfworld"), f"alfworld environment {env_name} must start with alfworld" | ||
env = Alfworld(instruction_type=instruction_type, feedback_type=feedback_type) | ||
return AlfworldWrapper(env, instruction_type=instruction_type, feedback_type=feedback_type) | ||
|
||
|
||
for env_name in ENVIRONMENTS: | ||
# default version (backwards compatibility) | ||
register( | ||
id=f"llf-{env_name}", | ||
entry_point='llfbench.envs.alfworld:make_env', | ||
kwargs=dict(env_name=env_name, feedback_type='a', instruction_type='b') | ||
) |
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
why added
ReasonType
here but not used?