Skip to content

Commit

Permalink
Merge pull request #1335 from opentensor/release/5.0.0
Browse files Browse the repository at this point in the history 
Release/5.0.0
  • Loading branch information
@ifrit98
ifrit98 committed May 18, 2023
2 parents 07e84e8 + 76f5fc9 commit bc10fb1
Show file tree
Hide file tree
Showing 301 changed files with 15,592 additions and 24,016 deletions.
158 changes: 79 additions & 79 deletions .test_durations
View file

Large diffs are not rendered by default.

5 changes: 5 additions & 0 deletions CHANGELOG.md
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,10 @@
# Changelog

## 5.0.0 / 2023-05-17

**Full Changelog**: https://github.com/opentensor/bittensor/compare/v4.0.1...v5.0.0


## 4.0.1 / 2023-04-21

* Fix btcli my_delegates bug by @camfairchild in ef32a4da0d0827ab5977af1454d66ffe97cbc572
Expand Down
2 changes: 1 addition & 1 deletion DEVELOPMENT_WORKFLOW.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -37,7 +37,7 @@ Each of these branches have a specific purpose and are bound to strict rules as
- May branch off from: `nobunaga`
- Must merge back into: `nobunaga`
- Branch naming convention:
- Anything except master, nobunaga, nakamoto, release/* or hotfix/*
- Anything except master, nobunaga, finney, release/* or hotfix/*
- Suggested: `feature/<ticket>/<descriptive-sentence>`

Feature branches are used to develop new features for the upcoming or a distant future release. When starting development of a feature, the target release in which this feature will be incorporated may well be unknown at that point.
Expand Down
323 changes: 10 additions & 313 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -9,47 +9,25 @@

### Internet-scale Neural Networks <!-- omit in toc -->

[Discord](https://discord.gg/bittensor) • [Docs](https://docs.bittensor.com/) • [Network](https://www.bittensor.com/network) • [Research](https://drive.google.com/file/d/1VnsobL6lIAAqcA1_Tbm8AYIQscfJV4KU) • [Code](https://github.com/opentensor/BitTensor)
[Discord](https://discord.gg/bittensor) • [Network](https://taostats.io/) • [Research](https://drive.google.com/file/d/1VnsobL6lIAAqcA1_Tbm8AYIQscfJV4KU)

</div>

This repository contains Bittensor's Python API, which can be used for the following purposes:

1. Querying the Bittensor network as a [client](https://github.com/opentensor/bittensor#31-client).
2. Running and building Bittensor miners and validators for [mining TAO](https://github.com/opentensor/bittensor#43-running-a-template-miner).
3. Pulling network [state information](https://github.com/opentensor/bittensor#3-using-bittensor).
4. Managing [TAO wallets](https://github.com/opentensor/bittensor#41-cli), balances, transfers, etc.
1. Querying the Bittensor network as a client.
2. Running and building Bittensor miners and validators.
3. Pulling network state information.
4. Managing TAO wallets, balances, transfers, etc.

Bittensor is a mining network, similar to Bitcoin, that includes built-in incentives designed to encourage miners to provide value by hosting trained or training machine learning models. These models can be queried by clients seeking inference over inputs, such as token-based text generations or numerical embeddings from a large foundation model like GPT-NeoX-20B.

Token-based incentives are designed to drive the network's growth and distribute the value generated by the network directly to the individuals producing that value, without intermediaries. The network is open to all participants, and no individual or group has full control over what is learned, who can profit from it, or who can access it.

To learn more about Bittensor, please read our [paper](https://drive.google.com/file/d/1VnsobL6lIAAqcA1_Tbm8AYIQscfJV4KU/view).

- [1. Documentation](#1-documentation)
- [2. Install](#2-install)
- [3. Using Bittensor](#3-using-bittensor)
- [3.1. Client](#31-client)
- [3.2. Server](#32-server)
- [3.3. Validator](#33-validator)
- [4. Features](#4-features)
- [4.1. Using the CLI](#41-cli)
- [4.2. Selecting the network to join](#42-selecting-the-network-to-join)
- [4.3. Running a template miner](#43-running-a-template-miner)
- [4.4. Running a template server](#44-running-a-template-server)
- [4.5. Syncing with the chain/ Finding the ranks/stake/uids of other nodes](#46-syncing-with-the-chain-finding-the-ranksstakeuids-of-other-nodes)
- [4.6. Finding and creating the endpoints for other nodes in the network](#47-finding-and-creating-the-endpoints-for-other-nodes-in-the-network)
- [4.7. Querying others in the network](#48-querying-others-in-the-network)
- [5. Release](#5-release)
- [6. License](#6-license)
- [7. Acknowledgments](#7-acknowledgments)

## 1. Documentation

https://docs.bittensor.com/

## 2. Install
Three ways to install Bittensor
# Install
There are three ways to install Bittensor

1. Through the installer:
```
Expand All @@ -67,291 +45,10 @@ $ git clone https://github.com/opentensor/bittensor.git
$ python3 -m pip install -e bittensor/
```

## 3. Using Bittensor

The following examples showcase how to use the Bittensor API for 3 separate purposes.

### 3.1. Client

Querying the network for generations.

```python
import bittensor
wallet = bittensor.wallet().create_if_non_existent()
graph = bittensor.metagraph().sync()
print ( bittensor.dendrite( wallet = wallet ).generate
(
endpoints = graph.endpoints[graph.incentive.sort()[1][-1]], // The highest ranked peer.
prompt = "The quick brown fox jumped over the lazy dog",
num_to_generate = 20
)
)
```

Querying the network for representations.

```python
import bittensor
wallet = bittensor.wallet().create_if_non_existent()
graph = bittensor.metagraph().sync()
print ( bittensor.dendrite( wallet = wallet ).text_last_hidden_state
(
endpoints = graph.endpoints[graph.incentive.sort()[1][-1]], // The highest ranked peer.
inputs = "The quick brown fox jumped over the lazy dog"
)
)
...
// Apply model.
...
loss.backward() // Accumulate gradients on endpoints.
```

### 3.2. Server

Serving a custom model.

```python
import bittensor
import torch
from transformers import GPT2Model, GPT2Config

model = GPT2Model( GPT2Config(vocab_size = bittensor.__vocab_size__, n_embd = bittensor.__network_dim__ , n_head = 8))
optimizer = torch.optim.SGD( [ {"params": model.parameters()} ], lr = 0.01 )

def forward_text( pubkey, inputs_x ):
return model( inputs_x )

def backward_text( pubkey, inputs_x, grads_dy ):
with torch.enable_grad():
outputs_y = model( inputs_x.to(device) ).last_hidden_state
torch.autograd.backward (
tensors = [ outputs_y.to(device) ],
grad_tensors = [ grads_dy.to(device) ]
)
optimizer.step()
optimizer.zero_grad()

wallet = bittensor.wallet().create().register()
axon = bittensor.axon (
wallet = wallet,
forward_text = forward_text,
backward_text = backward_text
).start().serve()
```

### 3.3. Validator

Validating models by setting weights.

```python
import bittensor
import torch

graph = bittensor.metagraph().sync()
dataset = bittensor.dataset()
chain_weights = torch.ones( [graph.n.item()], dtype = torch.float32 )

for batch in dataset.dataloader( 10 ):
...
// Train chain_weights.
...
bittensor.subtensor().set_weights (
weights = chain_weights,
uids = graph.uids,
wait_for_inclusion = True,
wallet = bittensor.wallet(),
)
```
## 4. Features

### 4.1. CLI

Creating a new wallet.
```bash
$ btcli new_coldkey
$ btcli new_hotkey
```

Listing your wallets
```bash
$ btcli list
```

Registering a wallet
```bash
$ btcli register
```

Running a miner
```bash
$ btcli run
```

Checking balances
```bash
$ btcli overview
```

Checking the incentive mechanism.
```bash
$ btcli metagraph
```

Transfering funds
```bash
$ btcli transfer
```

Staking/Unstaking from a hotkey
```bash
$ btcli stake
$ btcli unstake
```

### 4.2. Selecting the network to join
There are two open Bittensor networks: staging (Nobunaga) and main (Nakamoto, Local).

- Nobunaga (staging)
- Nakamoto (main)
- Local (localhost, mirrors nakamoto)

```bash
$ export NETWORK=local
$ python (..) --subtensor.network $NETWORK
or
>> btcli run --subtensor.network $NETWORK
```

### 4.3. Running a template miner

The following command will run Bittensor's template miner

```bash
$ cd bittensor
$ python ./bittensor/_neuron/text/template_miner/main.py
```
or
```python3
>> import bittensor
>> bittensor.neurons.text.template_miner.neuron().run()
```

OR with customized settings

```bash
$ cd bittensor
$ python3 ./bittensor/_neuron/text/template_miner/main.py --wallet.name <WALLET NAME> --wallet.hotkey <HOTKEY NAME>
```

For the full list of settings, please run

```bash
$ python3 ~/.bittensor/bittensor/bittensor/_neuron/neurons/text/template_miner/main.py --help
```

### 4.4. Running a template server

The template server follows a similar structure as the template miner.

```bash
$ cd bittensor
$ python3 ./bittensor/_neuron/text/core_server/main.py --wallet.name <WALLET NAME> --wallet.hotkey <HOTKEY NAME>
```
or
```python3
>> import bittensor
>> bittensor.neurons.text.core_server.neuron().run()
```

For the full list of settings, please run

```bash
$ cd bittensor
$ python3 ./bittensor/_neuron/text/core_server/main.py --help
```


### 4.5. Serving an endpoint on the network

Endpoints are served to the bittensor network through the axon. The axon is instantiated via a wallet which holds an account on the Bittensor network.

```python
import bittensor

wallet = bittensor.wallet().create().register()
axon = bittensor.axon (
wallet = wallet,
forward_text = forward_text,
backward_text = backward_text
).start().serve()
```

### 4.6. Syncing with the chain/ Finding the ranks/stake/uids of other nodes

Information from the chain is collected/formated by the metagraph.

```bash
btcli metagraph
```
and
```python
import bittensor

meta = bittensor.metagraph()
meta.sync()

# --- uid ---
print(meta.uids)

# --- hotkeys ---
print(meta.hotkeys)

# --- ranks ---
print(meta.R)

# --- stake ---
print(meta.S)

```

### 4.7. Finding and creating the endpoints for other nodes in the network

```python
import bittensor

meta = bittensor.metagraph()
meta.sync()

### Address for the node uid 0
endpoint_as_tensor = meta.endpoints[0]
endpoint_as_object = meta.endpoint_objs[0]
```

### 4.8. Querying others in the network

```python
import bittensor

meta = bittensor.metagraph()
meta.sync()

### Address for the node uid 0
endpoint_0 = meta.endpoints[0]

### Creating the wallet, and dendrite
wallet = bittensor.wallet().create().register()
den = bittensor.dendrite(wallet = wallet)
representations, _, _ = den.forward_text (
endpoints = endpoint_0,
inputs = "Hello World"
)

```

## 5. Release
## Release
The release manager should follow the instructions of the [RELEASE_GUIDELINES.md](./RELEASE_GUIDELINES.md) document.

## 6. License
## License
The MIT License (MIT)
Copyright © 2021 Yuma Rao

Expand All @@ -362,5 +59,5 @@ The above copyright notice and this permission notice shall be included in all c
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.


## 7. Acknowledgments
## Acknowledgments
**learning-at-home/hivemind**
2 changes: 1 addition & 1 deletion VERSION
View file
Original file line number Diff line number Diff line change
@@ -1 +1 @@
4.0.1
5.0.0
Loading

0 comments on commit bc10fb1

Please sign in to comment.