Feat/dialogpt en

annotators/IntentCatcher/data/intent_phrases.json

@@ -559,7 +559,7 @@
         },
         "what_can_you_do": {
             "phrases": [
-                "((so )){0,1}(tell me ){0,1}what can you do",
+                "((so )){0,1}(tell me ){0,1}what (else ){0,1}can you do",
                 "(tell me ){0,1}what are you able to do",
                 "(tell me ){0,1}what you can do",
                 "what are your ((skills)|(abilities)|(features))",

assistant_dists/dream/cpu.yml

@@ -43,4 +43,6 @@ services:
     environment:
       DEVICE: cpu
       CUDA_VISIBLE_DEVICES: ""
-
+  dialogpt:
+    environment:
+      CUDA_VISIBLE_DEVICES: ""

assistant_dists/dream/dev.yml

@@ -394,6 +394,11 @@ services:
       - "./annotators/midas_predictor:/src"
     ports:
       - 8121:8121
+  dialogpt:
+    volumes:
+      - "./services/dialogpt:/src"
+    ports:
+      - 8125:8125
   dff-template-skill:
     volumes:
       - "./skills/dff_template_skill:/src"

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, dff-template-skill:8120"
+          midas-predictor:8121, dialogpt:8125, dff-template-skill:8120"
       WAIT_HOSTS_TIMEOUT: ${WAIT_TIMEOUT:-480}
   convers-evaluator-annotator:
     env_file: [.env]
@@ -1136,6 +1136,25 @@ services:
           memory: 50M
         reservations:
           memory: 50M
+
+  dialogpt:
+    env_file: [ .env ]
+    build:
+      args:
+        SERVICE_PORT: 8125
+        SERVICE_NAME: dialogpt
+        PRETRAINED_MODEL_NAME_OR_PATH: microsoft/DialoGPT-small
+      context: ./services/dialogpt/
+    command: flask run -h 0.0.0.0 -p 8125
+    environment:
+      - CUDA_VISIBLE_DEVICES=0
+      - FLASK_APP=server
+    deploy:
+      resources:
+        limits:
+          memory: 2G
+        reservations:
+          memory: 2G
 
   dff-template-skill:
     env_file: [.env]

assistant_dists/dream/gpu1.yml

@@ -183,6 +183,10 @@ services:
     restart: unless-stopped
   midas-predictor:
     restart: unless-stopped
+  dialogpt:
+    restart: unless-stopped
+    environment:
+      - CUDA_VISIBLE_DEVICES=9
   dff-template-skill:
     restart: unless-stopped
 version: '3.7'

assistant_dists/dream/pipeline_conf.json

@@ -951,6 +951,19 @@
                 ],
                 "state_manager_method": "add_hypothesis"
             },
+            "dialogpt": {
+                "connector": {
+                    "protocol": "http",
+                    "timeout": 2,
+                    "url": "http://dialogpt:8125/respond"
+                },
+                "dialog_formatter": "state_formatters.dp_formatters:last_utt_and_history_dialog",
+                "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",

assistant_dists/dream/proxy.yml

@@ -584,7 +584,15 @@ services:
     environment:
       - PROXY_PASS=dream.deeppavlov.ai:8121
       - PORT=8121
-
+
+  dialogpt:
+    command: [ "nginx", "-g", "daemon off;" ]
+    build:
+      context: dp/proxy/
+      dockerfile: Dockerfile
+    environment:
+      - PROXY_PASS=dream.deeppavlov.ai:8125
+
   dff-template-skill:
     command: [ "nginx", "-g", "daemon off;" ]
     build:

assistant_dists/dream/test.yml

@@ -123,5 +123,8 @@ services:
   midas-predictor:
     environment:
       - CUDA_VISIBLE_DEVICES=6
+  dialogpt:
+    environment:
+      - CUDA_VISIBLE_DEVICES=6
   dff-template-skill:
 version: '3.7'

services/dialogpt/Dockerfile

@@ -0,0 +1,20 @@
+# syntax=docker/dockerfile:experimental
+
+FROM pytorch/pytorch:1.5-cuda10.1-cudnn7-runtime
+
+WORKDIR /src
+
+ARG PRETRAINED_MODEL_NAME_OR_PATH
+ENV PRETRAINED_MODEL_NAME_OR_PATH ${PRETRAINED_MODEL_NAME_OR_PATH}
+ARG SERVICE_PORT
+ENV SERVICE_PORT ${SERVICE_PORT}
+
+COPY ./requirements.txt /src/requirements.txt
+RUN pip install -r /src/requirements.txt
+
+RUN python -c "from transformers import AutoTokenizer; AutoTokenizer.from_pretrained('${PRETRAINED_MODEL_NAME_OR_PATH}');"
+RUN python -c "from transformers import AutoModelForCausalLM; AutoModelForCausalLM.from_pretrained('${PRETRAINED_MODEL_NAME_OR_PATH}');"
+
+COPY . /src
+
+CMD gunicorn --workers=1 server:app -b 0.0.0.0:${SERVICE_PORT} --timeout=300

services/dialogpt/README.md

@@ -0,0 +1,3 @@
+GPU RAM = 1Gb
+cpu time = 0.15 sec 
+gpu time = 0.05 sec 

services/dialogpt/requirements.txt

@@ -0,0 +1,9 @@
+transformers==4.6.0
+flask==1.1.1
+itsdangerous==2.0.1
+gunicorn==19.9.0
+requests==2.22.0
+sentry-sdk[flask]==0.14.1
+healthcheck==1.3.3
+jinja2<=3.0.3
+Werkzeug<=2.0.3

services/dialogpt/server.py

@@ -0,0 +1,82 @@
+import logging
+import time
+import os
+
+import sentry_sdk
+import torch
+from flask import Flask, request, jsonify
+from sentry_sdk.integrations.flask import FlaskIntegration
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+sentry_sdk.init(dsn=os.getenv("SENTRY_DSN"), integrations=[FlaskIntegration()])
+
+
+logging.basicConfig(format="%(asctime)s - %(name)s - %(levelname)s - %(message)s", level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+PRETRAINED_MODEL_NAME_OR_PATH = os.environ.get("PRETRAINED_MODEL_NAME_OR_PATH")
+logging.info(f"PRETRAINED_MODEL_NAME_OR_PATH = {PRETRAINED_MODEL_NAME_OR_PATH}")
+DEFAULT_CONFIDENCE = 0.9
+ZERO_CONFIDENCE = 0.0
+MAX_HISTORY_DEPTH = 3
+
+try:
+    tokenizer = AutoTokenizer.from_pretrained(PRETRAINED_MODEL_NAME_OR_PATH)
+    model = AutoModelForCausalLM.from_pretrained(PRETRAINED_MODEL_NAME_OR_PATH)
+    if torch.cuda.is_available():
+        model.to("cuda")
+        logger.info("dialogpt is set to run on cuda")
+
+    logger.info("dialogpt is ready")
+except Exception as e:
+    sentry_sdk.capture_exception(e)
+    logger.exception(e)
+    raise e
+
+app = Flask(__name__)
+logging.getLogger("werkzeug").setLevel("WARNING")
+
+
+def generate_response(context, model, tokenizer):
+    encoded_context = []
+    for uttr in context[-MAX_HISTORY_DEPTH:]:
+        encoded_context += [tokenizer.encode(uttr + tokenizer.eos_token, return_tensors="pt")]
+    bot_input_ids = torch.cat(encoded_context, dim=-1)
+
+    with torch.no_grad():
+        if torch.cuda.is_available():
+            bot_input_ids = bot_input_ids.to("cuda")
+        chat_history_ids = model.generate(
+            bot_input_ids, do_sample=True, max_length=50, top_k=3, pad_token_id=tokenizer.eos_token_id
+        )
+        if torch.cuda.is_available():
+            chat_history_ids = chat_history_ids.cpu()
+    return tokenizer.decode(chat_history_ids[:, bot_input_ids.shape[-1] :][0], skip_special_tokens=True)
+
+
+@app.route("/respond", methods=["POST"])
+def respond():
+    st_time = time.time()
+    contexts = request.json.get("utterances_histories", [])
+
+    try:
+        responses = []
+        confidences = []
+        for context in contexts:
+            response = generate_response(context, model, tokenizer)
+            if len(response) > 3:
+                # drop too short responses
+                responses += [response]
+                confidences += [DEFAULT_CONFIDENCE]
+            else:
+                responses += [""]
+                confidences += [ZERO_CONFIDENCE]
+    except Exception as exc:
+        logger.exception(exc)
+        sentry_sdk.capture_exception(exc)
+        responses = [""] * len(contexts)
+        confidences = [ZERO_CONFIDENCE] * len(contexts)
+
+    total_time = time.time() - st_time
+    logger.info(f"masked_lm exec time: {total_time:.3f}s")
+    return jsonify(list(zip(responses, confidences)))

services/dialogpt/test.py

@@ -0,0 +1,17 @@
+import requests
+
+
+def test_respond():
+    url = "http://0.0.0.0:8125/respond"
+
+    contexts = [["hi", "hi. how are you?"], ["let's chat about movies", "cool. what movies do you like?"]]
+    gold_result = [["I'm good, how are you?", 0.9], ["I like the new one.", 0.9]]
+    result = requests.post(url, json={"utterances_histories": contexts}).json()
+    assert [
+        len(sample[0]) > 0 and sample[1] > 0.0 for sample in result
+    ], f"Got\n{result}\n, but expected:\n{gold_result}"
+    print("Success")
+
+
+if __name__ == "__main__":
+    test_respond()

services/dialogpt/test.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+python test.py

skill_selectors/rule_based_selector/connector.py

@@ -185,6 +185,7 @@ async def send(self, payload: Dict, callback: Callable):
                 skills_for_uttr.append("personal_info_skill")
                 skills_for_uttr.append("meta_script_skill")
                 skills_for_uttr.append("dummy_skill")
+                skills_for_uttr.append("dialogpt")
                 if len(dialog["utterances"]) < 20:
                     skills_for_uttr.append("dff_friendship_skill")
 

tests/dream/assert_test_dialogs.py

@@ -43,6 +43,7 @@
         "dff_travel_skill",
         "dff_wiki_skill",
         "game_cooperative_skill",
+        "dialogpt",
     ],
 }
 

tests/runtests.sh

@@ -148,7 +148,7 @@ if [[ "$MODE" == "test_skills" || "$MODE" == "all" ]]; then
                      fact-random fact-retrieval dff-intent-responder-skill badlisted-words \
                      dff-gossip-skill dff-wiki-skill topic-recommendation dff-science-skill personal-info-skill \
                      user-persona-extractor small-talk-skill wiki-facts dff-art-skill dff-funfact-skill \
-                     meta-script-skill spelling-preprocessing dff-gaming-skill \
+                     meta-script-skill spelling-preprocessing dff-gaming-skill dialogpt \
                      dff-music-skill dff-bot-persona-skill entity-detection midas-predictor; do
 
         echo "Run tests for $container"