* Added a lock object into `SafeLlamaModelHandle` which all calls to `llama_decode` (in the `SafeLLamaContextHandle`) lock first. This prevents two contexts from running inference on the same model at the same time, which seems to be unsafe in llama.cpp.
* Modified the lock to be global over _all_ inferences. This seems to be necessary (at least with the CUDA backend).
- 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.
- Properly displaying `LLamaToken`
- Removed all tokenisation code in `SafeLLamaContextHandle` - just pass it all through to the `SafeLlamaModelHandle`
- Improved `SafeLlamaModelHandle` tokenisation:
- Renting an array, for one less allocation
- Not using `&tokens[0]` to take a pointer to an array, this is redundant and doesn't work on empty arrays
- Made `NativeApi` into a `static class` (it's not intended to be instantiated)
- Moved `LLamaTokenType` enum out into a separate file
- Made `LLamaSeqId` and `LLamaPos` into `record struct`, convenient to have equality etc
- `AntipromptProcessor` accepts chunks of text and returns a value indicating if any antiprompt has been detected.
- `StreamingTokenDecoder` decodes tokens into text, maintaining some internal state to handle single characters which are encoded as multiple tokens.
Added tests for these classes and updated StatelessExecutor to use them.
Removed most DeTokenize methods, marked the rest as obsolete (should always use a `StreamingTokenDecoder`).
- Just convert it to a `string`, nice and simple
- Write the bytes to a `Span<byte>` no allocations
- Write the chars to a `StringBuilder` potentially no allocations
- Used those methods to add a `Clone` method to SafeLLamaContextHandle
- Simplified `LLamaContext` by using the new methods
- Sealed `LLamaContext` and `LLamaEmbedder`
This change _only_ implements the low level API and makes no effort to update the LlamaSharp higher level abstraction.
It is built upon llama `b3f138d`, necessary DLLs are **not** included in this commit.
Martin Evans
Yaohui Liu