Skip to content

Commit

Permalink
Merge pull request #379 from deeppavlov/dev
Browse files Browse the repository at this point in the history 
Release v0.17.7
  • Loading branch information
@dilyararimovna
dilyararimovna authored Apr 7, 2023
2 parents 375a987 + b8dfdf5 commit ed5d384
Show file tree
Hide file tree
Showing 5 changed files with 19 additions and 7 deletions.
10 changes: 10 additions & 0 deletions MODELS.md
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
## Models used in Generative Assistants

Here you may find a list of models that currently available for use in Generative Assistants.

| model name | model link | open-source? | size (billion parameters) | GPU usage | max tokens (prompt + response) | description |
|------------|---------------------------------------------------------|--------------------------|---------------------------|---------------------------|--------------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| BLOOMZ 7B | [link](https://huggingface.co/bigscience/bloomz-7b1) | yes | 7.1B | 33GB | 2,048 tokens | An open-source multilingual task-oriented large language model. BLOOMZ 7B1 comes from BLOOMZ model family (featuring 560M, 1.1B, 1.7B, 3B, 7.1B, and 176B parameter versions). Each of the models is a [BLOOM](https://huggingface.co/bigscience/bloom) model of corresponding size, fine-tuned on cross-lingual task-instruction dataset (46 languages, 16 NLP tasks). For more details about BLOOM, refer to [this paper](https://arxiv.org/pdf/2211.05100.pdf). For more details about BLOOMZ and its dataset, refer to [this paper](https://arxiv.org/pdf/2211.01786.pdf). |
| GPT-J 6B | [link](https://huggingface.co/EleutherAI/gpt-j-6b) | yes | 6B | 25GB | 2,048 tokens | An open-source large language model. English-only, not fine-tuned for instruction following, not capable of code generation. For more details, refer to this [GitHub repo](https://github.com/kingoflolz/mesh-transformer-jax) |
| GPT-3.5 | [link](https://platform.openai.com/docs/models/gpt-3-5) | no (paid access via API) | supposedly, 175B | - (cannot be run locally) | 4,097 tokens | Based on text-davinci-003 -- the largest and most capable of GPT-3/GPT-3.5 models family (featuring davinci, curie, babbage, ada models) not optimized for chat. Unlike earlier GPT-3 models, also able to understand and generate code. Unlike GPT-3.5 turbo, not optimised for chat. For more details, refer to [OpenAI website](https://platform.openai.com/docs/models/gpt-3-5). |
| ChatGPT | [link](https://platform.openai.com/docs/models/gpt-3-5) | no (paid access via API) | supposedly, 175B | - (cannot be run locally) | 4,096 tokens | Based on gpt-3.5-turbo -- the most capable of the entire GPT-3/GPT-3.5 models family. Optimized for chat. Able to understand and generate code. For more details, refer to [OpenAI website](https://platform.openai.com/docs/models/gpt-3-5). |
4 changes: 2 additions & 2 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -242,7 +242,7 @@ Dream Architecture is presented in the following image:
| MIDAS Predictor | 30 MB RAM | BERT-based model trained on a semantic classes subset of MIDAS dataset |
| NER | 2.2 GB RAM, 5 GB GPU | extracts person names, names of locations, organizations from uncased text |
| News API Annotator | 80 MB RAM | extracts the latest news about entities or topics using the GNews API. DeepPavlov Dream deployments utilize our own API key. |
| Personality Catcher | 30 MB RAM | |
| Personality Catcher | 30 MB RAM | the skill is to change the system's personality description via chatting interface, it works as a system command, the response is system-like message |
| Prompt Selector | 50 MB RAM | Annotator utilizing Sentence Ranker to rank prompts and selecting `N_SENTENCES_TO_RETURN` most relevant prompts (based on questions provided in prompts) |
| Property Extraction | 6.3 GiB RAM | extracts user attributes from utterances |
| Rake Keywords | 40 MB RAM | extracts keywords from utterances with the help of RAKE algorithm |
Expand Down Expand Up @@ -303,7 +303,7 @@ Dream Architecture is presented in the following image:
| DFF Program Y Wide Skill | 110 MB RAM | **[New DFF version]** Chatbot Program Y (https://github.com/keiffster/program-y) adapted for Dream socialbot, which includes only very general templates (with lower confidence) |
| Small Talk Skill | 35 MB RAM | asks questions using the hand-written scripts for 25 topics, including but not limited to love, sports, work, pets, etc. |
| SuperBowl Skill | 30 MB RAM | supports FAQ, facts, and scripts for SuperBowl |
| Text QA | 1.8 GB RAM, 2.8 GB GPU | |
| Text QA | 1.8 GB RAM, 2.8 GB GPU | The service finds the answer of a factoid question in text. |
| Valentine's Day Skill | 30 MB RAM | supports FAQ, facts, and scripts for Valentine's Day |
| Wikidata Dial Skill | 100 MB RAM | generates an utterance using Wikidata triplets. Not turned on, needs improvement |
| DFF Animals Skill | 200 MB RAM | is created using DFF and has three branches of conversation about animals: user's pets, pets of the socialbot, and wild animals |
Expand Down
2 changes: 1 addition & 1 deletion services/dialogpt/component.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@ dialogpt:
name: dialogpt
display_name: DialoGPT
container_name: dialogpt
component_type: null
component_type: Generative
model_type: NN-based
is_customizable: false
author: DeepPavlov
Expand Down
6 changes: 3 additions & 3 deletions services/text_qa/component.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,11 +2,11 @@ text-qa:
name: text_qa
display_name: Text QA
container_name: text-qa
component_type: null
component_type: Retrieval
model_type: NN-based
is_customizable: false
author: DeepPavlov
description: ''
description: The service finds the answer of a factoid question in text.
ram_usage: 1.8G
gpu_usage: 2.8G
port: 8078
Expand All @@ -22,7 +22,7 @@ text-qa-ru:
name: text_qa
display_name: Text QA
container_name: text-qa-ru
component_type: null
component_type: Retrieval
model_type: NN-based
is_customizable: false
author: DeepPavlov
Expand Down
4 changes: 3 additions & 1 deletion skills/personality_catcher/component.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,9 @@ personality-catcher:
model_type: Dictionary/Pattern-based
is_customizable: false
author: DeepPavlov
description: ''
description: the skill is to change the system's personality description
via chatting interface, it works as a system command,
the response is system-like message
ram_usage: 100M
gpu_usage: null
port: 8010
Expand Down

0 comments on commit ed5d384

Please sign in to comment.