dialoGPT Persona (#185)

* Fix requirements.txt (#84) * fix itsdangerous requirements * pin itsdangerous requirements for all flask==1.1.1 servers * increase timeout to 5s * add logs * increase to 100 * add first working version gpt_persona * add sentence_ranking(not working) * fix wrong endpoint * create sentecnce ranker annotator * rewrite text generation logic * add comments to code * fix gpt_persona fallback * clean code, add train script, write tests * add get_intents, remove dataset, add hyperparams * add batch support * fix: move files * fix: merge * fix: codestyle * fix: remove sentence ranker * feat: new distribution and rename skill * feat: new annotator * feat: relative persona extractor * fix: codestyle * fix: proxy * fix: params * fix: volumes * fix: reqs * fix: tests * fix: tests relative sents extr * fix: batching * fix: codestyle * fix: persona extractor tests * fix: persona get * fix: tests * fix: imports * fix: logs * fix: docs * fix: add midas * fix: command * feat: add to main dist and tests * feat: remove infilling, add dialogpt persona based, docs * fix: gpus * fix: params * fix: param * fix: indent * fix: remove infilling from tests Co-authored-by: Andrii.Hura <54397922+AndriiHura@users.noreply.github.com> Co-authored-by: mtalimanchuk <mtalimanchuk@gmail.com> Co-authored-by: Dilyara Baymurzina <dilyara.rimovna@gmail.com>
deeppavlov · Oct 10, 2022 · c3bb406 · c3bb406
1 parent 9374614
commit c3bb406
Show file tree

Hide file tree

Showing 34 changed files with 2,575 additions and 49 deletions.
diff --git a/.env b/.env
@@ -34,3 +34,4 @@ DIALOGPT_SERVICE_URL=http://dialogpt:8091/respond
 DIALOGPT_CONTINUE_SERVICE_URL=http://dialogpt:8125/continue
 PROMPT_STORYGPT_SERVICE_URL=http://prompt-storygpt:8127/respond
 STORYGPT_SERVICE_URL=http://storygpt:8126/respond
+SENTENCE_RANKER_SERVICE_URL=http://sentence-ranker:8128/respond
diff --git a/README.md b/README.md
diff --git a/annotators/relative_persona_extractor/Dockerfile b/annotators/relative_persona_extractor/Dockerfile
@@ -0,0 +1,18 @@
+FROM python:3.8.4
+
+ARG SERVICE_PORT
+ENV SERVICE_PORT ${SERVICE_PORT}
+ARG N_SENTENCES_OT_RETURN
+ENV N_SENTENCES_OT_RETURN ${N_SENTENCES_OT_RETURN}
+
+RUN mkdir /src
+
+COPY ./annotators/relative_persona_extractor/ /src/
+COPY ./common/ /src/common/
+
+COPY annotators/relative_persona_extractor/requirements.txt /src/requirements.txt
+RUN pip install -r /src/requirements.txt
+
+WORKDIR /src
+
+CMD gunicorn --workers=1 server:app --bind 0.0.0.0:8000
diff --git a/annotators/relative_persona_extractor/README.txt b/annotators/relative_persona_extractor/README.txt
@@ -0,0 +1,5 @@
+# Relative Persona Extractor
+
+An annotator that utilizes Sentence Ranker to find the most relevant to the current context sentences from the bot's persona description.
+
+The number of returned sentences is given as an environmental variable using `N_SENTENCES_OT_RETURN` in `docker-compose.yml`.
diff --git a/annotators/relative_persona_extractor/requirements.txt b/annotators/relative_persona_extractor/requirements.txt
@@ -0,0 +1,9 @@
+flask==1.1.1
+itsdangerous==2.0.1
+gunicorn==20.0.4
+sentry-sdk==0.13.4
+requests==2.22.0
+click<=8.0.4
+jinja2<=3.0.3
+Werkzeug<=2.0.3
+numpy==1.17.2
diff --git a/annotators/relative_persona_extractor/server.py b/annotators/relative_persona_extractor/server.py
@@ -0,0 +1,73 @@
+import logging
+import requests
+import time
+from os import getenv
+
+import numpy as np
+import sentry_sdk
+from flask import Flask, request, jsonify
+
+
+sentry_sdk.init(getenv("SENTRY_DSN"))
+logging.basicConfig(format="%(asctime)s - %(name)s - %(levelname)s - %(message)s", level=logging.DEBUG)
+logger = logging.getLogger(__name__)
+app = Flask(__name__)
+
+SENTENCE_RANKER_SERVICE_URL = getenv("SENTENCE_RANKER_SERVICE_URL")
+N_SENTENCES_OT_RETURN = int(getenv("N_SENTENCES_OT_RETURN"))
+with open("common/persona_sentences.txt", "r") as f:
+    PERSONA_SENTENCES = f.read().splitlines()
+PERSONA_SENTENCES = [x.strip() for x in PERSONA_SENTENCES if len(x.strip())]
+
+
+def get_result(request):
+    st_time = time.time()
+    contexts = request.json["contexts"]
+    result = []
+    pairs = []
+    context_ids = []
+
+    for context_id, context in enumerate(contexts):
+        str_context = " ".join(context)
+        for sent in PERSONA_SENTENCES:
+            pairs += [[str_context, sent]]
+            context_ids += [context_id]
+    context_ids = np.array(context_ids)
+    try:
+        scores = requests.post(SENTENCE_RANKER_SERVICE_URL, json={"sentence_pairs": pairs}, timeout=1.5).json()[0][
+            "batch"
+        ]
+        scores = np.array(scores)
+        for i, context in enumerate(contexts):
+            curr_ids = np.where(context_ids == i)[0]
+            most_relevant_sent_ids = np.argsort(scores[curr_ids])[::-1][:N_SENTENCES_OT_RETURN]
+            curr_result = {
+                "persona": [PERSONA_SENTENCES[_id] for _id in most_relevant_sent_ids],
+                "max_similarity": scores[curr_ids][most_relevant_sent_ids[0]],
+            }
+            logger.info(f"Persona: {curr_result['persona']}")
+            result += [curr_result]
+    except Exception as exc:
+        logger.exception(exc)
+        sentry_sdk.capture_exception(exc)
+        result = [{"persona": [], "max_similarity": 0.0}] * len(contexts)
+
+    total_time = time.time() - st_time
+    logger.info(f"relative-persona-extractor exec time: {total_time:.3f}s")
+    return result
+
+
+@app.route("/respond", methods=["POST"])
+def respond():
+    result = get_result(request)
+    return jsonify(result)
+
+
+@app.route("/respond_batch", methods=["POST"])
+def respond_batch():
+    result = get_result(request)
+    return jsonify([{"batch": result}])
+
+
+if __name__ == "__main__":
+    app.run(debug=False, host="0.0.0.0", port=3000)
diff --git a/annotators/relative_persona_extractor/test.py b/annotators/relative_persona_extractor/test.py
@@ -0,0 +1,45 @@
+import requests
+
+
+def main():
+    url = "http://0.0.0.0:8133/respond"
+    input_data = {
+        "contexts": [
+            [
+                "How do you spend your spare time?",
+                "I like to watch movies and eat pizza.",
+                "Cool! What else do you like?",
+            ],
+            [
+                "I like to go to the cinema on fridays",
+                "great. how do you spend your spare time?",
+                "I like to watch movies",
+            ],
+        ]
+    }
+    gold = [
+        {
+            "max_similarity": 0.6948127746582031,
+            "persona": [
+                "I like Italian food especially pasta and pizza.",
+                "I like to watch football and basketball on TV.",
+                "I like watching travel video blogs.",
+            ],
+        },
+        {
+            "max_similarity": 0.6451027989387512,
+            "persona": [
+                "I like watching travel video blogs.",
+                "I like to watch football and basketball on TV.",
+                "I like Italian food especially pasta and pizza.",
+            ],
+        },
+    ]
+
+    result = requests.post(url, json=input_data).json()
+    assert result == gold, print(result)
+    print("Success!")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/annotators/relative_persona_extractor/test.sh b/annotators/relative_persona_extractor/test.sh
@@ -0,0 +1,3 @@
+#!/bin/bash
+
+python test.py
diff --git a/assistant_dists/dream/cpu.yml b/assistant_dists/dream/cpu.yml
@@ -49,8 +49,13 @@ services:
   dialogpt:
     environment:
       CUDA_VISIBLE_DEVICES: ""
-  infilling:
+  sentence-ranker:
     environment:
+      DEVICE: cpu
+      CUDA_VISIBLE_DEVICES: ""
+  dialogpt-persona-based:
+    environment:
+      DEVICE: cpu
       CUDA_VISIBLE_DEVICES: ""
   intent-catcher:
     environment:

diff --git a/assistant_dists/dream/dev.yml b/assistant_dists/dream/dev.yml
@@ -410,11 +410,23 @@ services:
       - "./services/dialogpt:/src"
     ports:
       - 8125:8125
-  infilling:
+  sentence-ranker:
     volumes:
-      - "./services/infilling:/src"
+      - "./services/sentence_ranker:/src"
     ports:
-      - 8122:8122
+      - 8128:8128
+  dialogpt-persona-based:
+    volumes:
+      - "./services/dialogpt_persona_based:/src"
+      - "./common:/src/common"
+    ports:
+      - 8131:8131
+  relative-persona-extractor:
+    volumes:
+      - "./annotators/relative_persona_extractor:/src"
+      - "./common:/src/common"
+    ports:
+      - 8133:8133
   dff-template-skill:
     volumes:
       - "./skills/dff_template_skill:/src"

diff --git a/assistant_dists/dream/docker-compose.override.yml b/assistant_dists/dream/docker-compose.override.yml
@@ -19,7 +19,7 @@ services:
           dff-funfact-skill:8104, dff-bot-persona-skill:8105, news-api-annotator:8112,
           dff-gossip-skill:8109, dff-wiki-skill:8111, dff-gaming-skill:8115, topic-recommendation:8113,
           user-persona-extractor:8114, wiki-facts:8116, dff-music-skill:8099, entity-detection:8103, dff-art-skill:8117,
-          midas-predictor:8121, dialogpt:8125, storygpt:8126, prompt-storygpt:8127, infilling:8122, dff-template-skill:8120"
+          midas-predictor:8121, dialogpt:8125, storygpt:8126, prompt-storygpt:8127, dff-template-skill:8120"
       WAIT_HOSTS_TIMEOUT: ${WAIT_TIMEOUT:-480}
   convers-evaluator-annotator:
     env_file: [.env]
@@ -1187,22 +1187,63 @@ services:
         reservations:
           memory: 2G
 
-  infilling:
+  dialogpt-persona-based:
     env_file: [ .env ]
     build:
-      context: ./services/infilling/
       args:
-        SERVICE_PORT: 8122
-    command: flask run -h 0.0.0.0 -p 8122
+        SERVICE_PORT: 8131
+        SERVICE_NAME: dialogpt_persona_based
+        PRETRAINED_MODEL_NAME_OR_PATH: dim/dialogpt-medium-persona-chat
+        MAX_PERSONA_SENTENCES: 3
+      context: .
+      dockerfile: ./services/dialogpt_persona_based/Dockerfile
+    command: flask run -h 0.0.0.0 -p 8131
+    environment:
+      - CUDA_VISIBLE_DEVICES=0
+      - FLASK_APP=server
+    deploy:
+      resources:
+        limits:
+          memory: 2G
+        reservations:
+          memory: 2G
+
+  relative-persona-extractor:
+    env_file: [ .env ]
+    build:
+      args:
+        SERVICE_PORT: 8133
+        SERVICE_NAME: relative_persona_extractor
+        N_SENTENCES_OT_RETURN: 3
+      context: .
+      dockerfile: ./annotators/relative_persona_extractor/Dockerfile
+    command: flask run -h 0.0.0.0 -p 8133
+    environment:
+      - FLASK_APP=server
+    deploy:
+      resources:
+        limits:
+          memory: 80M
+        reservations:
+          memory: 80M
+
+  sentence-ranker:
+    env_file: [ .env ]
+    build:
+      args:
+        SERVICE_PORT: 8128
+        PRETRAINED_MODEL_NAME_OR_PATH: sentence-transformers/bert-base-nli-mean-tokens
+      context: ./services/sentence_ranker/
+    command: flask run -h 0.0.0.0 -p 8128
     environment:
       - CUDA_VISIBLE_DEVICES=0
       - FLASK_APP=server
     deploy:
       resources:
         limits:
-          memory: 2.5G # ?
+          memory: 3G
         reservations:
-          memory: 2.5G # ?
+          memory: 3G
 
   storygpt:
     env_file: [ .env ]
@@ -1261,4 +1302,3 @@ services:
         reservations:
           memory: 128M
 version: '3.7'
-
diff --git a/assistant_dists/dream/gpu1.yml b/assistant_dists/dream/gpu1.yml
@@ -193,10 +193,14 @@ services:
     restart: unless-stopped
     environment:
       - CUDA_VISIBLE_DEVICES=9
-  infilling:
+  dialogpt-persona-based:
     restart: unless-stopped
     environment:
       - CUDA_VISIBLE_DEVICES=7
+  sentence-ranker:
+    restart: unless-stopped
+    environment:
+      - CUDA_VISIBLE_DEVICES=9
   dff-template-skill:
     restart: unless-stopped
 version: '3.7'
diff --git a/assistant_dists/dream/pipeline_conf.json b/assistant_dists/dream/pipeline_conf.json
@@ -440,7 +440,20 @@
                     "annotators.midas_classification",
                     "annotators.combined_classification"
                 ]
-            }
+            },
+            "relative_persona_extractor": {
+                "connector": {
+                "protocol": "http",
+                "timeout": 2,
+                "url": "http://relative-persona-extractor:8133/respond"
+                },
+                "dialog_formatter": "state_formatters.dp_formatters:context_formatter_dialog",
+                "response_formatter": "state_formatters.dp_formatters:simple_formatter_service",
+                "state_manager_method": "add_annotation",
+                "previous_services": [
+                    "annotators.spelling_preprocessing"
+                ]
+          }
         },
         "skill_selectors": {
             "rule_based_selector": {
@@ -977,6 +990,19 @@
                 ],
                 "state_manager_method": "add_hypothesis"
             },
+            "dialogpt_persona_based": {
+                "connector": {
+                "protocol": "http",
+                "timeout": 3,
+                "url": "http://dialogpt-persona-based:8131/respond"
+                },
+                "dialog_formatter": "state_formatters.dp_formatters:persona_bot_formatter",
+                "response_formatter": "state_formatters.dp_formatters:skill_with_attributes_formatter_service",
+                "previous_services": [
+                    "skill_selectors"
+                ],
+                "state_manager_method": "add_hypothesis"
+            },
             "dff_template_skill": {
                 "connector": {
                     "protocol": "http",

diff --git a/assistant_dists/dream/proxy.yml b/assistant_dists/dream/proxy.yml
@@ -576,14 +576,32 @@ services:
       - PROXY_PASS=dream.deeppavlov.ai:8125
       - PORT=8125
 
-  infilling:
+  sentence-ranker:
     command: [ "nginx", "-g", "daemon off;" ]
     build:
       context: dp/proxy/
       dockerfile: Dockerfile
     environment:
-      - PROXY_PASS=dream.deeppavlov.ai:8122
-      - PORT=8122
+      - PROXY_PASS=dream.deeppavlov.ai:8128
+      - PORT=8128
+
+  dialogpt-persona-based:
+    command: [ "nginx", "-g", "daemon off;" ]
+    build:
+      context: dp/proxy/
+      dockerfile: Dockerfile
+    environment:
+      - PROXY_PASS=dream.deeppavlov.ai:8131
+      - PORT=8125
+
+  relative-persona-extractor:
+    command: [ "nginx", "-g", "daemon off;" ]
+    build:
+      context: dp/proxy/
+      dockerfile: Dockerfile
+    environment:
+      - PROXY_PASS=dream.deeppavlov.ai:8133
+      - PORT=8133
 
   dff-short-story-skill:
     command: [ "nginx", "-g", "daemon off;" ]

diff --git a/assistant_dists/dream/test.yml b/assistant_dists/dream/test.yml
@@ -125,8 +125,11 @@ services:
   prompt-storygpt:
     environment:
       - CUDA_VISIBLE_DEVICES=8
-  infilling:
+  dialogpt-persona-based:
     environment:
-      - CUDA_VISIBLE_DEVICES=7
+      - CUDA_VISIBLE_DEVICES=9
+  sentence-ranker:
+    environment:
+      - CUDA_VISIBLE_DEVICES=9
   dff-template-skill:
 version: '3.7'