[chattable]

https://ai.facebook.com/blog/blender-bot-2-an-open-source-chatbot-that-builds-long-term-memory-and-searches-the-internet

[Denis ROBERT]

In testing, we’ve found that BlenderBot 2.0 outperforms the conversational abilities of the best existing systems.

They all have the best chatbot

[squarebear]

It certainly isn't the first chatbot to have a long term memory, as they claim.

[chattable]

i wonder if blender bot 2 have short term memory as well?
it would need it for roleplay.
i wonder if they made another ai dungeon with this would it be any good at it?
i wonder how good is it at daydreaming?

[frankinstien]

"You may talk with our models. The 2.7B can be interacted with on a 16gb P100 GPU or better. The 9.4B parameter model requires at least two 32gb V100 GPUs to interact with."

P100 GPU $4555.00
Double-Precision. Performance 4.7 TeraFLOPS.
Single-Precision. Performance 9.3 TeraFLOPS.
Half-Precision. Performance 18.7 TeraFLOPS.

V100 GPU $12,610.31
FP16 (half) performance 28.26 TFLOPS
FP32 (float) performance 14.13 TFLOPS
FP64 (double) performance 7.066 TFLOPS

Compared to my Vega 56 which cost me, at the time, $425.00.
FP16 (half) performance 21.09 TFLOPS
FP32 (float) performance 10.54 TFLOPS
FP64 (double) performance 659.0 GFLOPS

While the V100 has 34% more computing power, its cost is 35 times greater and the P100 compared to the Vega 56 has 13% less computing power and costs over ten times more! Also, note that double or single precision computation isn't necessary for neural networks. I wish I had bought two of them when they were at bargain prices...

[chattable]

my gpu for my computer is a amd radeon hd.

[frankinstien]

my gpu for my computer is a amd radeon hd.

I'm referring to my AI server:



The server has two 10 core Intel Xeons for a total of 40 threads and the m2.NVMe SSDs are 3TBs

[chattable]

oh!

[MikeB]

While the V100 has 34% more computing power, its cost is 35 times greater and the P100 compared to the Vega 56 has 13% less computing power and costs over ten times more

The V100 also has 5120 cores, and the P100 has 3584... some of the later Quadro GPU's have a lot of cores too but are more reasonably priced.

[MagnusWootton]

my gtx980 has 2048!  very fast beast,  good for raytracing and ai.

[squarebear]

Finally got around to testing this today.
 
First touch ("Hello") throws off a _POTENTIALLY_UNSAFE__ tag after processing for 31 seconds!
It was downhill from there until it crashed. 

[infurl]

I didn't find any online versions to try out just now so I'm assuming you followed the instructions here and set it up yourself.

https://www.parl.ai/docs/tutorial_quick.html

If that's the case, what were you running it on?

[squarebear]

It's not online anywhere that I could see. Yes, we set it up ourselves.
On AWS, it is a c5.xlarge, which is listed as being an Intel Xeon with 4 vCPU and 8 GiB memory.  All-core turbo frequency up to 3.4GHz, and single core turbo frequency up to 3.5 GHz. 

[infurl]

Thanks for the info. From what you reported, it sounds underpowered. I don't even have a laptop that small anymore. If I get time I'll give it a try myself though I don't know when I'll be able to.

At the very least you should try to use a lot more memory, 32GB if it's an option.

[chattable]

i asked the people on the botlibre websight if they could set blender bot 2.0 up with botlibre for desktop.
i am waiting for a response.