LLamaSharp

1074 Commits

8 Branches

333 MB

26 Download

Tree: c7d0dc915a

Commit Graph

Author	SHA1	Message	Date
Martin Evans	c7d0dc915a	Assorted small changes to clean up some code warnings	1 year ago
Martin Evans	949861a581	- Added a `Modify` method to `Conversation`. This grants temporary access to directly modify the KV cache. - Re-implmented `Rewind` as an extension method using `Modify` internally - Implemented `ShiftLeft`, which shifts everything over except for some starting tokens. This is the same as the `StatelessExecutor` out-of-context handling. - Starting batch at epoch 1, this ensures that conversations (starting at zero) are below the current epoch. It also means `0` can always be used as a value guaranteed to be below the current epoch.	1 year ago
Martin Evans	b0acecf080	Created a new `BatchedExecutor` which processes multiple "Conversations" in one single inference batch. This is faster, even when the conversations are unrelated, and is much faster if the conversations share some overlap (e.g. a common system prompt prefix). Conversations can be "forked", to create a copy of a conversation at a given point. This allows e.g. prompting a conversation with a system prefix just once and then forking it again and again for each individual conversation. Conversations can also be "rewound" to an earlier state. Added two new examples, demonstrating forking and rewinding.	1 year ago

Author

SHA1

Message

Date

Martin Evans

c7d0dc915a

Assorted small changes to clean up some code warnings

1 year ago

Martin Evans

949861a581

- Added a `Modify` method to `Conversation`. This grants **temporary** access to directly modify the KV cache.

- Re-implmented `Rewind` as an extension method using `Modify` internally
 - Implemented `ShiftLeft`, which shifts everything over except for some starting tokens. This is the same as the `StatelessExecutor` out-of-context handling.
 - Starting batch at epoch 1, this ensures that conversations (starting at zero) are below the current epoch. It also means `0` can always be used as a value guaranteed to be below the current epoch.

1 year ago

Martin Evans

b0acecf080

Created a new `BatchedExecutor` which processes multiple "Conversations" in one single inference batch. This is faster, even when the conversations are unrelated, and is much faster if the conversations share some overlap (e.g. a common system prompt prefix).

Conversations can be "forked", to create a copy of a conversation at a given point. This allows e.g. prompting a conversation with a system prefix just once and then forking it again and again for each individual conversation. Conversations can also be "rewound" to an earlier state.

Added two new examples, demonstrating forking and rewinding.

1 year ago

3 Commits (c7d0dc915aae3d38d6adeeb22cfbff5cadfaf770)