feat: add mistral 7b 128k (#589)

* feat: add mistral 7b 128k

* fix: model name

* fix: model name

* fix: configs

* fix: config names and cards

* fix: more cards

* fix: mem

* fix: no proxy available

MODELS.md

@@ -14,3 +14,4 @@
 | [Anthropic Claude Instant v1](https://docs.anthropic.com/claude/reference/complete_post)     | anthropic-api-claude-instant-v1   | no (paid access via API)             | supposedly, 52B  | - (cannot be run locally) | 9,000 tokens                   | available under subscription plan, commercial use allowed  | A smaller model with far lower latency, sampling at roughly 40 words/sec! Its output quality is somewhat lower than the latest claude-1 model, particularly for complex tasks. However, it is much less expensive and blazing fast. NB: paid. You must provide your Anthropic API key to use the model. Your Anthropic API account will be charged according to your usage.                                                                                    |
 | Russian XGLM 4.5B (private weights)                                                          | transformers-lm-ruxglm            | no                                   | 4.5B             | 15GB                      | 2,048 tokens                   | Not available yet                                          | A private large language model for the Russian language which was fine-tuned for instruction following by Dmitry Kosenko in Summer 2023. This model is up and running on our servers and can be used for free.                                                                                                                                                                                                                                                 |
 | [ruGPT-3.5-13B](https://huggingface.co/ai-forever/ruGPT-3.5-13B)                             | transformers-lm-rugpt35           | yes                                  | 13B              | 35GB (half-precision)     | 2,048 tokens                   | MIT                                                        | A large language model for the Russian language which was used for trainig GigaChat. This model is up and running on our servers and can be used for free.                                                                                                                                                                                                                                                                                                     |
+| [Mistral 7B 128k Tokens](https://huggingface.co/NousResearch/Yarn-Mistral-7b-128k)           | transformers-lm-mistral-7b-128k   | yes                                  | 7B               | 20GB (half-precision)     | 128,000 tokens                 | Apache 2.0 , commercial use is allowed                     | An open-source English-only large language model which was fine-tuned for instruction following but is NOT capable of code generation. NB: free of charge. This model is up and running on our servers and can be used for free.                                                                                                                                                                                                                               |

assistant_dists/document_based_qa_transformers/cpu.yml

@@ -0,0 +1,10 @@
+version: '3.7'
+services:
+  combined-classification:
+    environment:
+      DEVICE: cpu
+      CUDA_VISIBLE_DEVICES: ""
+  sentence-ranker:
+    environment:
+      DEVICE: cpu
+      CUDA_VISIBLE_DEVICES: ""

assistant_dists/document_based_qa_transformers/db_conf.json

@@ -0,0 +1,6 @@
+{
+    "host": "DB_HOST",
+    "port": "DB_PORT",
+    "name": "DB_NAME",
+    "env": true
+}

assistant_dists/document_based_qa_transformers/dev.yml

@@ -0,0 +1,52 @@
+# С такими volumes удобно дебажить, не нужно пересобирать контейнер каждый раз при изменении кода
+services:
+  agent:
+    volumes:
+      - ".:/dp-agent"
+    ports:
+      - 4242:4242
+  files:
+    ports:
+      - 3000:3000
+    volumes:
+      - "~/.deeppavlov/file_server:/tmp"
+  sentseg:
+    volumes:
+      - "./annotators/SentSeg:/src"
+    ports:
+      - 8011:8011
+  combined-classification:
+    volumes:
+      - "./common:/src/common"
+      - "./annotators/combined_classification:/src"
+    ports:
+      - 8087:8087
+  sentence-ranker:
+    volumes:
+      - "./services/sentence_ranker:/src"
+      - "~/.deeppavlov/cache:/root/.cache"
+    ports:
+      - 8128:8128
+  transformers-lm-mistral-7b-128k:
+    volumes:
+      - "./services/transformers_lm:/src"
+      - "./common:/src/common"
+      - "~/.deeppavlov/cache:/root/.cache"
+    ports:
+      - 8185:8185
+  doc-retriever:
+    volumes:
+      - "./annotators/doc_retriever:/src"
+      - "./common:/src/common"
+      - "./documents:/src/documents"
+    ports:
+      - 8165:8165
+  dff-document-qa-transformers-llm-skill:
+    volumes:
+      - "./skills/dff_document_qa_llm_skill:/src"
+      - "./common:/src/common"
+      - "./documents:/src/documents"
+    ports:
+      - 8186:8186
+
+version: "3.7"

assistant_dists/document_based_qa_transformers/docker-compose.override.yml

@@ -0,0 +1,158 @@
+services:
+  agent:
+    command: sh -c 'bin/wait && python -m deeppavlov_agent.run agent.pipeline_config=assistant_dists/document_based_qa_transformers/pipeline_conf.json'
+    environment:
+      WAIT_HOSTS: "sentseg:8011, combined-classification:8087, ranking-based-response-selector:8002,
+        sentence-ranker:8128, transformers-lm-mistral-7b-128k:8185, doc-retriever:8165, dff-document-qa-transformers-llm-skill:8186"
+      WAIT_HOSTS_TIMEOUT: ${WAIT_TIMEOUT:-1000}
+      HIGH_PRIORITY_INTENTS: 1
+      RESTRICTION_FOR_SENSITIVE_CASE: 1
+      ALWAYS_TURN_ON_ALL_SKILLS: 0
+      LANGUAGE: EN
+
+  files:
+    image: julienmeerschart/simple-file-upload-download-server
+
+  sentseg:
+    env_file: [ .env ]
+    build:
+      context: ./annotators/SentSeg/
+      dockerfile: Dockerfile-test
+    command: flask run -h 0.0.0.0 -p 8011
+    environment:
+      - FLASK_APP=server
+    deploy:
+      resources:
+        limits:
+          memory: 1.5G
+        reservations:
+          memory: 1.5G
+
+  doc-retriever:
+    env_file: [ .env ]
+    build:
+      context: .
+      dockerfile: ./annotators/doc_retriever/Dockerfile
+      args:
+        SERVICE_PORT: 8165
+        SERVICE_NAME: doc_retriever
+        CONFIG_PATH: ./doc_retriever_config.json
+        DOC_PATH_OR_LINK: http://files.deeppavlov.ai/dream_data/documents_for_qa/test_file_dream_repo.html,http://files.deeppavlov.ai/dream_data/documents_for_qa/alphabet_financial_report.txt,http://files.deeppavlov.ai/dream_data/documents_for_qa/test_file_jurafsky_chatbots.pdf
+        PARAGRAPHS_NUM: 5
+        FILE_SERVER_TIMEOUT: 30
+    command: python -m flask run -h 0.0.0.0 -p 8165
+    environment:
+      - FLASK_APP=server
+      - CUDA_VISIBLE_DEVICES=0
+    deploy:
+      resources:
+        limits:
+          memory: 5G
+        reservations:
+          memory: 5G
+
+  combined-classification:
+    env_file: [ .env ]
+    build:
+      args:
+        CONFIG: combined_classifier.json
+        SERVICE_PORT: 8087
+      context: .
+      dockerfile: ./annotators/combined_classification/Dockerfile
+    command: gunicorn --workers=1 server:app -b 0.0.0.0:8087 --timeout 600
+    environment:
+      - CUDA_VISIBLE_DEVICES=0
+    deploy:
+      resources:
+        limits:
+          memory: 2G
+        reservations:
+          memory: 2G
+
+  ranking-based-response-selector:
+    env_file: [ .env ]
+    build:
+      args:
+        SERVICE_PORT: 8002
+        SERVICE_NAME: response_selector
+        LANGUAGE: EN
+        SENTENCE_RANKER_ANNOTATION_NAME: sentence_ranker
+        SENTENCE_RANKER_SERVICE_URL: http://sentence-ranker:8128/respond
+        SENTENCE_RANKER_TIMEOUT: 3
+        N_UTTERANCES_CONTEXT: 5
+        FILTER_TOXIC_OR_BADLISTED: 1
+      context: .
+      dockerfile: ./response_selectors/ranking_based_response_selector/Dockerfile
+    command: flask run -h 0.0.0.0 -p 8002
+    environment:
+      - FLASK_APP=server
+    deploy:
+      resources:
+        limits:
+          memory: 100M
+        reservations:
+          memory: 100M
+
+  sentence-ranker:
+    env_file: [ .env ]
+    build:
+      args:
+        SERVICE_PORT: 8128
+        SERVICE_NAME: sentence_ranker
+        PRETRAINED_MODEL_NAME_OR_PATH: sentence-transformers/all-MiniLM-L6-v2
+      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: 3G
+        reservations:
+          memory: 3G
+
+  transformers-lm-mistral-7b-128k:
+    env_file: [ .env ]
+    build:
+      args:
+        SERVICE_PORT: 8185
+        SERVICE_NAME: transformers_lm_mistral_7b_128k
+        PRETRAINED_MODEL_NAME_OR_PATH: NousResearch/Yarn-Mistral-7b-128k
+        HALF_PRECISION: 1
+        USE_FLASH_ATTENTION_2: 1
+      context: .
+      dockerfile: ./services/transformers_lm/Dockerfile
+    command: flask run -h 0.0.0.0 -p 8185
+    environment:
+      - CUDA_VISIBLE_DEVICES=0
+      - FLASK_APP=server
+    deploy:
+      resources:
+        limits:
+          memory: 50G
+        reservations:
+          memory: 50G
+
+  dff-document-qa-transformers-llm-skill:
+    env_file: [ .env ]
+    build:
+      args:
+        SERVICE_PORT: 8186
+        SERVICE_NAME: dff_document_qa_llm_skill
+        GENERATIVE_SERVICE_URL: http://transformers-lm-mistral-7b-128k:8185/respond
+        GENERATIVE_SERVICE_CONFIG: transformers_mistral.json
+        GENERATIVE_TIMEOUT: 120
+        N_UTTERANCES_CONTEXT: 7
+        FILE_SERVER_TIMEOUT: 30
+        DOCUMENT_PROMPT_FILE: common/prompts/document_qa_instruction.json
+      context: .
+      dockerfile: ./skills/dff_document_qa_llm_skill/Dockerfile
+    deploy:
+      resources:
+        limits:
+          memory: 128M
+        reservations:
+          memory: 128M
+
+version: '3.7'