ParlAI

ParlAI copied to clipboard

Add FlashAttention Kernel in Triton

1

**Tldr: Add implementations of [FlashAttention](https://arxiv.org/abs/2205.14135) using OpenAI's triton language.** **Background**: - FlashAttention: an IO-aware exact attention algorithm that uses tiling to reduce the number of memory reads/writes, 15% end-to-end speedup...

pearlli98

BlenderBot3: inference on a particular module

5

Hi, I was wondering how can I inference a particular module of the BB3 model? For example, the generate dialogue response module takes in “Full context + knowledge + memory...

lidiyam

Hi, I'm trying to host a chat service with blenderbot 3 using multiple gpus. Here is my config file ``` tasks: default: onboard_world: MessengerBotChatOnboardWorld task_world: MessengerBotChatTaskWorld timeout: 1800 agents_required: 1...

ryanshea10

**Patch description** Made a small update to the PACER model plugin that allows for it to be added to other agents (like the `ExpandedAttentionDecoderAndPacerAgent`) without doubly loading the base model...

JackUrb

Feature: Add the ability to log model artifacts to Weights & Biases

3

**Patch description** This PR adds the ability to log model artifacts to [Weights & Biases](https://wandb.ai/site). Logging model artifacts allows complete reproducibility of the experiments that resulted in the trained model...

parambharat

Blenderbot 3 Import Error

5

Hi I am trying to run Blenderbot 3 on Ubuntu 20.04 tried with RTX 2080Ti and RTX 3080Ti and it worked exactly once (one turn) after which it keeps giving...

shleeku

**Bug description** Please enter a clear and concise description of what the bug is. I see this : https://parl.ai/projects/bb3/ When I run this ```parlai interactive --init-opt gen/opt_bb3 --opt-server API_SERVER --loglevel...

ildoonet

Commands for Seeker Training on Dialog GPT2 or T5?

7

Hey I was wondering what the set of commands are to get parlai to train t5 or GPT2 to train with seeker parameters where the search query is the copy...

ArEnSc

Fixed T5 and GPT2 related bugs in the hugging_face agents

3

**Patch description** Previously T5 dictionary was undefined and elements of the huggingface dictionary agent were uninitialized leading to immediate exceptions in the train step for T5 and at the beginning...

michaelsaxon

Fix seed from opt in rng sampler of multiwoz_v22

2

**Patch description** Since the support for setting seed in `train_model.py` has been added, we can let user fix the seed in dataset sampler also. **Testing steps** **Other information**

prajjwal1