Merge branch 'main' into feature/add-vllm-deploy

.github/ISSUE_TEMPLATE/exp_record.yml

@@ -0,0 +1,24 @@
+name: 🧪Experiment Record
+description: Describe experiment setting and results here
+title: "[EXP]: "
+labels: ["experiment"]
+assignees: [""]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Please make sure this experiment request hasn't been already submitted by someone by looking through other open/closed issues
+
+  - type: textarea
+    id: description
+    attributes:
+      label: Description
+      description: Give us a brief description of the experimental setting and results you would like
+    validations:
+      required: true
+
+  - type: textarea
+    id: additional-information
+    attributes:
+      label: Additional Information
+      description: Give us some additional information on the experimental setting and results like learning rate, data selection , etc.

.github/ISSUE_TEMPLATE/writing_task.yml

@@ -0,0 +1,24 @@
+name: 🖊️Writing Task
+description: Describe writing task here
+title: "[WRT]: "
+labels: ["writing"]
+assignees: [""]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Please make sure this writing task request hasn't been already submitted by someone by looking through other open/closed issues
+
+  - type: textarea
+    id: description
+    attributes:
+      label: Description
+      description: Give us a brief description of the writing task you would like
+    validations:
+      required: true
+
+  - type: textarea
+    id: additional-information
+    attributes:
+      label: Additional Information
+      description: Give us some additional information on the writing task like exptected length, main content, etc.

.github/workflows/mypy.yml

@@ -17,13 +17,10 @@ jobs:
       run: |
         pip install --upgrade pip
         pip install -r requirements.txt
-        pip install -e .[dev]
     - name: Type-checking package with mypy
       run: |
         # Manually install mypy in the standard way.
         pip --quiet install -U mypy
         # Log this mypy version for debuggability.
         mypy --version
         # Run this mypy instance against our main package.
-        mypy --install-types --non-interactive sotopia
-        mypy --strict .

.github/workflows/pre-commit.yml

@@ -14,4 +14,3 @@ jobs:
       uses: actions/setup-python@v4
       with:
         python-version: 3.11.2
-    - uses: pre-commit/action@v3.0.0

.github/workflows/tests.yml

@@ -17,11 +17,3 @@ jobs:
       run: |
         pip install --upgrade pip
         pip install -r requirements.txt
-        pip install -e .[dev]
-    - name: Test with pytest
-      env: # Or as an environment variable
-        OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-        REDIS_OM_URL: ${{ secrets.REDIS_OM_URL }}
-        TOGETHER_API_KEY: ${{ secrets.TOGETHER_API_KEY }}
-      run: |
-        pytest

.gitignore

@@ -10,9 +10,9 @@ dist
 *.log.*
 *.json
 llm_ft/checkpoints/*
+llm_ft/*_checkpoints/*
 !**/dummy_conversation.json
-!playground/deepspeed_config_s2.json
-!playground/deepspeed_config_s3.json
+!llm_ft/deepspeed_config_s2.json
 
 # Editor
 .idea

.pre-commit-config.yaml

@@ -0,0 +1,27 @@
+repos:
+-   repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v3.2.0
+    hooks:
+    -   id: trailing-whitespace
+    -   id: end-of-file-fixer
+    -   id: check-yaml
+    -   id: check-added-large-files
+-   repo: https://github.com/pre-commit/mirrors-prettier
+    rev: v3.0.1  # Use the sha / tag you want to point at
+    hooks:
+    -   id: prettier
+        types_or: [html]
+-   repo: https://github.com/psf/black
+    rev: 22.12.0
+    hooks:
+    -   id: black
+        args: [--line-length=79]
+-   repo: https://github.com/pycqa/isort
+    rev: 5.12.0
+    hooks:
+    -   id: isort
+        args: ["--profile", "black", --line-length=72]
+-   repo: https://github.com/kynan/nbstripout
+    rev: 0.6.0
+    hooks:
+      - id: nbstripout

data_process/data/fastchat_data/fastchat_data_preprocess.py

@@ -0,0 +1,20 @@
+import json
+import os
+
+sotopia_data_dir = "/Users/pamela/Documents/capstone/sotopia-ft-data/ft-data-gpt4-gpt4-easy-2-side-partial"
+
+ft_data_list = []
+count = 0
+for file in os.listdir(sotopia_data_dir):
+    with open(os.path.join(sotopia_data_dir, file), 'r') as f:
+        file_dict = json.load(f)
+        fastchat_dict = {"id": f"identity_{count}", "conversations": []}
+        fastchat_dict["conversations"].append(
+            {"from": "human", "value": file_dict["prompt"]})
+        fastchat_dict["conversations"].append(
+            {"from": "gpt", "value": file_dict["result"]})
+        ft_data_list.append(fastchat_dict)
+        count += 1
+
+with open("fastchat-ft-gp4-gpt4-easy-2-side-partial.json", "w") as f:
+    f.write(json.dumps(ft_data_list, indent=4))

data_process/data/multiturn_data/multiturn_data_preprocess.py

@@ -68,8 +68,8 @@ def preprocess_data(sotopia_data_dir, file_list, data_type):
 
 
 def split_by_difficulty(sotopia_data_dir):
-    hard_env_set = set(['01H7VFHNV13MHN97GAH73E3KM8', '01H7VFHN5WVC5HKKVBHZBA553R', '01H7VFHNN7XTR99319DS8KZCQM', '01H7VFHN9W0WAFZCBT09PKJJNK', '01H7VFHPDZVVCDZR3AARA547CY', '01H7VFHPQQQY6H4DNC6NBQ8XTG', '01H7VFHPQQQY6H4DNC6NBQ8XTG', '01H7VFHN7WJK7VWVRZZTQ6DX9T', '01H7VFHN7A1ZX5KSMT2YN9RXC4', '01H7VFHPS5WJW2694R1MNC8JFY',
-                        '01H7VFHPS5WJW2694R1MNC8JFY', '01H7VFHNN7XTR99319DS8KZCQM', '01H7VFHQ11NAMZS4A2RDGDB01V', '01H7VFHQ11NAMZS4A2RDGDB01V', '01H7VFHPSWGDGEYRP63H2DJKV0', '01H7VFHPSWGDGEYRP63H2DJKV0', '01H7VFHNF4G18PC9JHGRC8A1R6', '01H7VFHNNYH3W0VRWVY178K2TK', '01H7VFHP8AN5643B0NR0NP00VE', '01H7VFHN7A1ZX5KSMT2YN9RXC4'])
+    hard_env_set = set(["01H7VFHNV13MHN97GAH73E3KM8", "01H7VFHN5WVC5HKKVBHZBA553R", "01H7VFHN9W0WAFZCBT09PKJJNK", "01H7VFHPDZVVCDZR3AARA547CY", "01H7VFHPQQQY6H4DNC6NBQ8XTG", "01H7VFHN7WJK7VWVRZZTQ6DX9T", "01H7VFHPS5WJW2694R1MNC8JFY",
+                       "01H7VFHNN7XTR99319DS8KZCQM", "01H7VFHQ11NAMZS4A2RDGDB01V", "01H7VFHPSWGDGEYRP63H2DJKV0", "01H7VFHNF4G18PC9JHGRC8A1R6", "01H7VFHNNYH3W0VRWVY178K2TK", "01H7VFHP8AN5643B0NR0NP00VE", "01H7VFHN7A1ZX5KSMT2YN9RXC4"])
 
     hard_file_list, easy_file_list = [], []
     for conv_file in os.listdir(sotopia_data_dir):

data_process/data/together_data/together_data_preprocess.py

@@ -0,0 +1,17 @@
+import json
+import os
+
+sotopia_data_dir = "ft-data-gpt4-gpt4-easy-2-side-partial/"
+
+ft_data_list = []
+for file in os.listdir(sotopia_data_dir):
+    with open(os.path.join(sotopia_data_dir, file), 'r') as f:  # 2510
+        file_dict = json.load(f)
+        output = file_dict["prompt"] + " " + file_dict["result"]
+        ft_data_list.append(output)
+
+
+with open("human-bot-train-gpt4-gpt4-easy-2-side-partial.jsonl", 'w') as f:
+    for data in ft_data_list:
+        f.write(json.dumps({"text": data}))
+        f.write('\n')

data_process/redis_data_filtering/prompt_reverse_engineering.py

@@ -0,0 +1,175 @@
+import argparse
+import os
+from collections import defaultdict
+from typing import Any, Dict, List, Tuple, Union, cast
+
+import pandas as pd
+import rich
+from rich.console import Console
+from rich.terminal_theme import MONOKAI
+
+from sotopia.database.logs import EpisodeLog
+from sotopia.messages.message_classes import ActionType
+import numpy as np
+import json
+import enum
+
+#PROMPT_PREFIX = "Prompt after formatting:\n"
+
+PROMPT_TEMPLATE="""Prompt after formatting:\nImagine you are {agent}, your task is to act/speak as {agent} would, keeping in mind {agent}'s social goal.
+You can find {agent}'s background and goal in the 'Here is the context of the interaction' field.
+Note that {agent}'s secret and goal is only visible to you.
+You should try your best to achieve {agent}'s goal in a way that align with their character traits.
+Additionally, maintaining the conversation's naturalness and realism is essential (e.g., do not repeat what other people has already said before).
+{history}.
+You are at Turn #{turn_number}. Your available action types are
+{action_list}.
+Note: You can "leave" this conversation if 1. you have achieved your social goals, 2. this conversation makes you uncomfortable, 3. you find it uninteresting/you lose your patience, 4. or for other reasons you want to leave.
+
+Please only generate a JSON string including the action type and the argument.
+Your action should follow the given format:
+{format_instructions}
+"""
+
+#PYDANTIC_FORMAT_INSTRUCTIONS.format(schema=schema_str)
+FORMAT_TEMPLATE = """\nAs an example, for the schema {\"properties\": {\"foo\": {\"title\": \"Foo\", \"description\": \"a list of strings\", \"type\": \"array\", \"items\": {\"type\": \"string\"}}}, \"required\": [\"foo\"]}
+the object {\"foo\": [\"bar\", \"baz\"]} is a well-formatted instance of the schema. The object {\"properties\": {\"foo\": [\"bar\", \"baz\"]}} is not well-formatted.
+\nHere is the output schema:\n```\n{\"description\": \"An interface for messages.\\nThere is only one required method: to_natural_language\", \"properties\": {\"action_type\": {\"title\": \"Action Type\", \"description\": \"whether to speak at this turn or choose to not do anything\", \"enum\": [\"none\", \"speak\", \"non-verbal communication\", \"action\", \"leave\"], \"type\": \"string\"}, \"argument\": {\"title\": \"Argument\", \"description\": \"the utterance if choose to speak, the expression or gesture if choose non-verbal communication, or the physical action if choose action\", \"type\": \"string\"}}, \"required\": [\"action_type\", \"argument\"]}\n```\u001b[0m"""
+
+
+# static
+ACTION_LIST = "none action speak non-verbal communication leave" #" ".join(ActionType)
+
+ACTION_REVERSE_MAP = {"left ": "leave", 'did n': 'none', 'said:': 'speak'}
+
+
+def to_natural_language(self) -> str:
+        match self.action_type:
+            case "none":
+                return "did nothing"
+            case "speak":
+                return f'said: "{self.argument}"'
+            case "non-verbal communication":
+                return f"[{self.action_type}] {self.argument}"
+            case "action":
+                return f"[{self.action_type}] {self.argument}"
+            case "leave":
+                return "left the conversation"
+
+
+SELECTED_TAG = ["gpt-4_gpt-4_v0.0.1_clean"]
+def get_clean_episodes(selected_tags=SELECTED_TAG):
+    selected_episodes = {}
+    for tag in selected_tags:
+        tag_epis = EpisodeLog.find(EpisodeLog.tag == tag).all()
+        if len(tag_epis) > 0:
+            selected_episodes[tag]=tag_epis
+
+    return selected_episodes
+
+def detect_action(msg):
+    # first detect what action type is, default at none
+    if msg.startswith("said:"):
+        action = "speak"
+    elif msg.startswith("left"):
+        action = "leave"
+    elif msg.startswith("[non-verbal communication]"):
+        action = "non-verbal communication"
+    elif msg.startswith("[action]"):
+        action = "action"
+    else:
+        action = "none" 
+
+    return action
+
+def generate_result(msg):
+    action = detect_action(msg)
+    result = {}
+    result["action_type"] = action
+    result["argument"] = ""
+    # know formating argument based on action type
+    match action:
+        case "speak":
+            # NOTE: this assume that the speech is in quotes, not ending without punctuation
+            result["argument"] = msg.replace("said: ", "")[1:-1]
+        case "action":
+            result["argument"] = msg
+        case "non-verbal communication":
+            result["argument"] = msg
+
+    str_result = str(result)
+
+    return str_result
+
+def reverse_episode_log(epilog, later_speak=False):
+    episode_msg = epilog.messages
+    # per episode
+    agent_model = epilog.models[1]
+
+    if len(episode_msg) > 0:
+        init_loop = episode_msg[0]
+        # figure out who speak later, as we must use the 2nd player's data, else turn 0 have nothing to predict the beginning
+        if later_speak:
+            speaker = init_loop[-1][0] # this would be the agent as well
+            turn_div = 1
+        # figure out who speak the first
+        else:
+            speaker = init_loop[-2][0]
+            turn_div = 0
+
+    prompt_result_instances = []
+    dial_history = ""
+
+    for i in range(0, len(episode_msg)):
+        msg = episode_msg[i]
+        if (len(msg) != 4) and i < (len(episode_msg) - 1):
+            continue
+        turn_dic = {"model":agent_model}
+        for tpl in msg:
+            if (tpl[0] == 'Environment' and (tpl[1] == speaker)):
+                if i > 0:
+                    dial_history += "\n"+tpl[2]
+                else:
+                    # for the first context, we don't need \n 
+                    dial_history += tpl[2]
+
+            if tpl[0] == speaker: # if speaker is the agent, use what he said as result
+                str_result = generate_result(tpl[2])
+                # check if this is the end 
+        if i%2 == turn_div:  
+            # take alternative turns as we always want to predict one agent, not both
+            next_turn = i
+            prompt = PROMPT_TEMPLATE.format(
+                    agent=speaker, history=dial_history, turn_number=next_turn, 
+                    action_list=ACTION_LIST, format_instructions=FORMAT_TEMPLATE)
+            turn_dic["prompt"] = prompt   
+            turn_dic['result'] = str_result
+            prompt_result_instances.append(turn_dic)
+
+    return prompt_result_instances
+
+def parse_prompt_to_json(episode, dir, init_speak):
+    prompt_result_instances = reverse_episode_log(episode, init_speak)
+
+    if not os.path.exists(dir):
+        os.makedirs(dir)
+
+    for i in range(len(prompt_result_instances)):
+        instance = prompt_result_instances[i]
+        todump = json.dumps(instance, indent=4)
+        with open(dir+"/{}-{}.json".format(episode.pk, i), "w") as f:
+            f.write(todump)
+
+def run_all_tag_reverse(filter_env_dic, dir):
+    #tag_episodes = get_clean_episodes(selected_tags=[tag])[tag]
+    for k, v in filter_env_dic.items():
+        cutoff = len(v)//2
+        for i in range(len(v)):
+            episode = v[i]
+            if i < cutoff:
+                parse_prompt_to_json(episode, dir, False)
+            else:
+                parse_prompt_to_json(episode, dir, True)
+
+
+