fix(embedding): set kv_unified=True when embedding=True to enable batch processing

Author: SanjanaB123

.gitignore

@@ -179,3 +179,5 @@ cython_debug/
 
 # downloaded model .bin files
 docker/open_llama/*.bin
+test_model.gguf
+.claude/

llama_cpp/llama.py

@@ -402,7 +402,7 @@ def __init__(
                 self.n_batch,
                 llama_cpp.llama_max_parallel_sequences(),
             )
-
+            self.context_params.kv_unified = True
         self._ctx = self._stack.enter_context(
             contextlib.closing(
                 internals.LlamaContext(