-
Notifications
You must be signed in to change notification settings - Fork 40
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
SN1-195: add inference task #344
Conversation
Hollyqui
commented
Aug 21, 2024
- Added Dynamic Model Loading/unloading to allow validators to run multiple reference models
- Added inference task to allow us to force miners to generate with a specific model (layer to be used for ensembling)
- Made scoring loop run asynchronously from task generation
Co-authored-by: Dmytro Bobrenko <17252809+dbobrenko@users.noreply.github.com>
…macrocosm-os/prompting into SN1-110-move-from-pip-to-poetry
* initial push on openai miner * deprecates langchain and hf miner, adds openai miner * brings back pm2 autorun command * Prevent multi-turn on translation and sentiment * updates gitignore * bumps bittensor version * adds block to logged wandb data * updates version * Add Unit Tests * Improve multi-turn prompting * Reiterate need for pronouns * Stress importance of pronouns * SN1-12 * drops block from log data * Switch last_update_block to local var instead of metagraph attribute * Update README.md * update logging to use python rather than loguru refs #200 * Reduce timout to 15 seconds * Change event logging * Remove loguru requirement * Increment Version Number * v2.5.2 (#287) * Point to macrocosmos entity * Adjust project name * Increase spec version to 2.5.2 --------- Co-authored-by: bkb2135 <bkb2135@columbia.edu> Co-authored-by: p-ferreira <38992619+p-ferreira@users.noreply.github.com> Co-authored-by: p-ferreira <pe_drojunior@hotmail.com> Co-authored-by: bkb2135 <98138173+bkb2135@users.noreply.github.com> Co-authored-by: Chengzhe <chengzhe8919@gmail.com>
Merge Main into pre-staging
- Added pytest and cleaned poetry.lock - Make vllm optional via validator extra - Made `angle-emb` optional
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Make sure to change PR request to pre-staging
branch, and merge with v2.7.1 changes
README.md
Outdated
|
||
# Compute Requirements | ||
|
||
1. To run a **validator**, you will need at least 70GB of VRAM. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We have vali/miner requirements and instructions defined in assets/validator.md and assets/miner.md
Make sure to merge with v2.7.0+
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I believe this should already be merged with v2.7.1, no?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
So it is merged with 2.7.1 but pre-staging has a different commit history, so github doesn't realise. I would suggest force pushing this into pre-staging later to get the correct commit history which is then also aligned with staging/main
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We moved miner/validator info into separate markdown files, main README is currently short and concise with links to detailed instructions
Looks like you've restored the previous version of README
@@ -45,7 +42,7 @@ def check_uid_availability( | |||
return True | |||
|
|||
|
|||
def get_random_uids(self: BaseNeuron, k: int, exclude: list[int] = None) -> np.ndarray: | |||
def get_random_uids(k: int | None = 10**6, exclude: list[int] = None, own_uid: int | None = None) -> np.ndarray: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We should make k
mandatory, it's bounded to ~1000
def get_random_uids(k: int | None = 10**6, exclude: list[int] = None, own_uid: int | None = None) -> np.ndarray: | |
def get_random_uids(k: int, exclude: list[int] = None, own_uid: int | None = None) -> np.ndarray: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why is that needed - there's only 1k miners in the network, right? So passing k=None simply means you sample all miners
Co-authored-by: Dmytro Bobrenko <17252809+dbobrenko@users.noreply.github.com>
Co-authored-by: Dmytro Bobrenko <17252809+dbobrenko@users.noreply.github.com>
- Adding dataset to retrieve random website, which feeds the inference task
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
README.md
Outdated
|
||
# Compute Requirements | ||
|
||
1. To run a **validator**, you will need at least 70GB of VRAM. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We moved miner/validator info into separate markdown files, main README is currently short and concise with links to detailed instructions
Looks like you've restored the previous version of README