Spaces:
Running
Running
Update README.md
Browse files
README.md
CHANGED
@@ -1,29 +1,58 @@
|
|
1 |
---
|
2 |
title: README
|
3 |
-
emoji:
|
4 |
colorFrom: blue
|
5 |
colorTo: red
|
6 |
sdk: static
|
7 |
pinned: false
|
|
|
8 |
---
|
9 |
|
10 |
# Reactive AI
|
11 |
-
We are working on our own
|
12 |
-
between interactions/sequences instead of between tokens/elements in sequence and provides reactive communication patterns.
|
13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
Our primary architecture - **Reactor** - is planned as the first _**awareness AGI model**_, that's modelling awareness as an _Infinite Chain-of-Thoughts_,
|
15 |
connected to _Short-Term and Long-Term Memory_ (_Attention-based Memory System_) and _Receptors/Effectors_ systems for real-time reactive processing.
|
16 |
It will be able to constantly and autonomously learn from interactions in _Continouos Live Learning_ process.
|
17 |
|
|
|
18 |
While the **Reactor** is the main goal, it's extremely hard to achieve, as it's definitely the most advanced neural network ensemble ever.
|
19 |
|
20 |
That's why we designed simplified architectures, for incremental transformation from language/reasoning models to awareness model:
|
21 |
- **Reactive Transformer** is introducing _Attention-based Memory System_ and adding _Short-Term Memory_ to Transformer language models
|
22 |
- **Preactor** is adding _Long-Term Memory_ and ability to learn from interactions
|
23 |
|
24 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
|
26 |
-
More info soon
|
27 |
|
28 |
## RxNN Platform
|
29 |
We are working on complete Reactive Neural Networks development framework - [RxNN github](https://github.com/RxAI-dev/RxNN)
|
|
|
1 |
---
|
2 |
title: README
|
3 |
+
emoji: 👁
|
4 |
colorFrom: blue
|
5 |
colorTo: red
|
6 |
sdk: static
|
7 |
pinned: false
|
8 |
+
short_description: Reactive AI - Reactive Neural Networks and Event-Driven AI
|
9 |
---
|
10 |
|
11 |
# Reactive AI
|
12 |
+
We are working on our own ideas of Reactive Neural Networks (RxNN) and Event-Driven AI, advancing from language models to AGI awareness models.
|
|
|
13 |
|
14 |
+
## Reactive Neural Networks and Event-Driven AI
|
15 |
+
Reactive Neural Networks (RxNN) are memory-augmented neural networks with higher levels of recurrence (inter-sequence vs. intra-sequence in RNNs),
|
16 |
+
focused on processing single interactions with access to previous interactions via memory layers. We call this _**event-driven real-time processing**_
|
17 |
+
to distinguish it from classical _data-driven processing_ of the full conversation history in each interaction. This difference is crucial in case
|
18 |
+
of AGI and awareness - the key feature of humans awareness, is that we remember what we were doing 10 mins ago, without recalling the whole-day history - we
|
19 |
+
are working in real-time - just like event-driven _Reactive Neural Networks_.
|
20 |
+
|
21 |
+
In Event-Driven AI models are processing the data in reaction to environment or internal events, and are emitting other response events as a result.
|
22 |
+
Processing of input and output events by the model is called the interaction. Event or an interaction could occur in any point in continous time. Models
|
23 |
+
have to be stateful and remember the data between the interactions.
|
24 |
+
|
25 |
+
_**Strong Reactive Neural Networks**_ like **Reactor** could emit and listen to its internal events, while the _**Weak Reactive Neural Networks**_ are
|
26 |
+
working only on environment events.
|
27 |
+
|
28 |
+
## Reactor AGI
|
29 |
Our primary architecture - **Reactor** - is planned as the first _**awareness AGI model**_, that's modelling awareness as an _Infinite Chain-of-Thoughts_,
|
30 |
connected to _Short-Term and Long-Term Memory_ (_Attention-based Memory System_) and _Receptors/Effectors_ systems for real-time reactive processing.
|
31 |
It will be able to constantly and autonomously learn from interactions in _Continouos Live Learning_ process.
|
32 |
|
33 |
+
## Reactive Language Models (RxLM)
|
34 |
While the **Reactor** is the main goal, it's extremely hard to achieve, as it's definitely the most advanced neural network ensemble ever.
|
35 |
|
36 |
That's why we designed simplified architectures, for incremental transformation from language/reasoning models to awareness model:
|
37 |
- **Reactive Transformer** is introducing _Attention-based Memory System_ and adding _Short-Term Memory_ to Transformer language models
|
38 |
- **Preactor** is adding _Long-Term Memory_ and ability to learn from interactions
|
39 |
|
40 |
+
### RxT-Alpha Open Research
|
41 |
+
We are currently working on **Reactive Transformer Proof-of-Concept - RxT-Alpha**, especially on the new reinforcement learning stage - **Memory Reinforcement Learning**,
|
42 |
+
that's required for our reactive models, between the _Supervised Fine-Tuning_ and _Reinforcement Learning from Human Feedback for reactive models (RxRLHF)_. The research
|
43 |
+
is open, we are publishing the results of all separate steps, just after finishing them.
|
44 |
+
|
45 |
+
The Proof-of-Concept includes 3 small scale models based on **Reactive Transformer** architecture:
|
46 |
+
- RxT-Alpha-Micro (~11M params) - pre-training and fine-tuning finished, MRL in progress - training based on small synthetic datasets
|
47 |
+
- RxT-Alpha-Mini (~70M params) - pre-training in progress - training on real data
|
48 |
+
- RxT-Alpha (~530M/0.5B params) - pre-training in progress - training on real data
|
49 |
+
|
50 |
+
All the models have theoretically infinite context, limited only for single interaction (message + response), but in practice it's limited by short-term memory
|
51 |
+
capacity (it will be improved in Preactor). Limits are:
|
52 |
+
- RxT-Alpha-Micro - 256 tokens for single interaction, 6 * 256 for STM size (768kb), expected length of a smooth conversation min. ~4k tokens
|
53 |
+
- RxT-Alpha-Mini - 1024 tokens for single interaction, 8 * 1024 for STM size (8mb), expected length of a smooth conversation min. ~16k tokens
|
54 |
+
- RxT-Alpha - 2048 tokens for single interaction, 12 * 2048 for STM size (50mb), expected length of a smooth conversation min. ~32k tokens
|
55 |
|
|
|
56 |
|
57 |
## RxNN Platform
|
58 |
We are working on complete Reactive Neural Networks development framework - [RxNN github](https://github.com/RxAI-dev/RxNN)
|