You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The primary purpose of the Forge ecosystem is to delineate infra concerns from model concerns thereby making RL experimentation easier. Forge delivers this by providing clear RL abstractions and one scalable implementation of these abstractions. When you need fine-grained control over placement, fault handling/redirecting training loads during a run, or communication patterns, the primitives are there. When you don’t, you can focus purely on your RL algorithm.
8
+
The primary purpose of the TorchForge ecosystem is to delineate infra concerns from model concerns thereby making RL experimentation easier. TorchForge delivers this by providing clear RL abstractions and one scalable implementation of these abstractions. When you need fine-grained control over placement, fault handling/redirecting training loads during a run, or communication patterns, the primitives are there. When you don’t, you can focus purely on your RL algorithm.
9
9
10
10
Key features:
11
11
- Usability for rapid research (isolating the RL loop from infrastructure)
12
12
- Hackability for power users (all parts of the RL loop can be easily modified without interacting with infrastructure)
13
13
- Scalability (ability to shift between async and synchronous training and across thousands of GPUs)
14
14
15
-
> ⚠️ **Early Development Warning**Forge is currently in an experimental
15
+
> ⚠️ **Early Development Warning**TorchForge is currently in an experimental
16
16
> stage. You should expect bugs, incomplete features, and APIs that may change
17
17
> in future versions. The project welcomes bugfixes, but to make sure things are
18
18
> well coordinated you should discuss any significant change before starting the
@@ -31,11 +31,11 @@ You can also find our notebook tutorials (coming soon)
31
31
32
32
### Basic
33
33
34
-
Forge requires the latest PyTorch nightly with [Monarch](https://github.com/meta-pytorch/monarch), [vLLM](https://github.com/vllm-project/vllm), and [torchtitan](https://github.com/pytorch/torchtitan). For convenience,
34
+
TorchForge requires the latest PyTorch nightly with [Monarch](https://github.com/meta-pytorch/monarch), [vLLM](https://github.com/vllm-project/vllm), and [torchtitan](https://github.com/pytorch/torchtitan). For convenience,
35
35
we have pre-packaged these dependencies as wheels in assets/wheels. (Note that the basic install script
36
36
uses [DNF](https://docs.fedoraproject.org/en-US/quick-docs/dnf/), but could be easily extended to other Linux OS.)
37
37
38
-
Forge requires the Github CLI (gh) to download a compatible vLLM package. See [here](https://github.com/cli/cli#installation) for gh install instructions before continuting. Please login to gh with your Github account before continuing with `gh auth login`. You may use either https or ssh as the protocol for authentication.
38
+
TorchForge requires the Github CLI (gh) to download a compatible vLLM package. See [here](https://github.com/cli/cli#installation) for gh install instructions before continuting. Please login to gh with your Github account before continuing with `gh auth login`. You may use either https or ssh as the protocol for authentication.
39
39
40
40
```bash
41
41
conda create -n forge python=3.10
@@ -56,11 +56,6 @@ If you need to re-build the wheels for whatever reason, you can do so with:
56
56
./scripts/build_wheels.sh
57
57
```
58
58
59
-
For your information, since the vLLM wheel is too large for GitHub, we uploaded it as a release in the `install.sh` script:
0 commit comments