Update sampling API for llama.cpp

[abetlen]

Updates llama-cpp-python to use the new llama.cpp sampler api.

Known Issues:

• Grammar parser seems to silently fail when there is a syntax error in the gbnf grammar

[e-c-d]

You probably know this already, but the llama_context_params.seed field should be removed everywhere:

diff --git a/llama_cpp/llama_cpp.py b/llama_cpp/llama_cpp.py
index efec065..fa53135 100644
--- a/llama_cpp/llama_cpp.py
+++ b/llama_cpp/llama_cpp.py
@@ -850,11 +850,10 @@ class llama_model_params(ctypes.Structure):
 # };
 class llama_context_params(ctypes.Structure):
     """Parameters for llama_context

     Attributes:
-        seed (int): RNG seed, -1 for random
         n_ctx (int): text context, 0 = from model
         n_batch (int): logical maximum batch size that can be submitted to llama_decode
         n_ubatch (int): physical maximum batch size
         n_seq_max (int): max number of sequences (i.e. distinct states for recurrent models)
         n_threads (int): number of threads to use for generation
@@ -881,11 +880,10 @@ class llama_context_params(ctypes.Structure):
         abort_callback (ggml_abort_callback): abort callback if it returns true, execution of llama_decode() will be aborted
         abort_callback_data (ctypes.ctypes.c_void_p): data for abort_callback
     """

     if TYPE_CHECKING:
-        seed: int
         n_ctx: int
         n_batch: int
         n_ubatch: int
         n_seq_max: int
         n_threads: int
@@ -911,11 +909,10 @@ class llama_context_params(ctypes.Structure):
         flash_attn: bool
         abort_callback: Callable[[ctypes.c_void_p], bool]
         abort_callback_data: ctypes.c_void_p

     _fields_ = [
-        ("seed", ctypes.c_uint32),
         ("n_ctx", ctypes.c_uint32),
         ("n_batch", ctypes.c_uint32),
         ("n_ubatch", ctypes.c_uint32),
         ("n_seq_max", ctypes.c_uint32),
         ("n_threads", ctypes.c_int32),

[abetlen]

@e-c-d yes thank you forgot to push this!