feat: Update llama_cpp.py bindings

Author: abetlen

llama_cpp/llama_cpp.py

@@ -468,11 +468,13 @@ def byref(obj: CtypesCData, offset: Optional[int] = None) -> CtypesRef[CtypesCDa
 #     LLAMA_POOLING_TYPE_NONE = 0,
 #     LLAMA_POOLING_TYPE_MEAN = 1,
 #     LLAMA_POOLING_TYPE_CLS  = 2,
+#     LLAMA_POOLING_TYPE_LAST = 3,
 # };
 LLAMA_POOLING_TYPE_UNSPECIFIED = -1
 LLAMA_POOLING_TYPE_NONE = 0
 LLAMA_POOLING_TYPE_MEAN = 1
 LLAMA_POOLING_TYPE_CLS = 2
+LLAMA_POOLING_TYPE_LAST = 3   
 
 # enum llama_split_mode {
 #     LLAMA_SPLIT_MODE_NONE    = 0, // single GPU
@@ -761,7 +763,6 @@ class llama_model_params(ctypes.Structure):
 
 #     enum llama_rope_scaling_type rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
 #     enum llama_pooling_type      pooling_type;      // whether to pool (sum) embedding results by sequence id
-#                                                     // (ignored if no pooling layer)
 
 #     // ref: https://github.com/ggerganov/llama.cpp/pull/2054
 #     float    rope_freq_base;   // RoPE base frequency, 0 = from model
@@ -2316,6 +2317,16 @@ def llama_n_threads_batch(ctx: llama_context_p, /) -> int:
     ...
 
 
+# // Set whether the model is in embeddings model or not
+# // If true, embeddings will be returned but logits will not
+# LLAMA_API void llama_set_embeddings(struct llama_context * ctx, bool embeddings);
+@ctypes_function("llama_set_embeddings", [llama_context_p_ctypes, ctypes.c_bool], None)
+def llama_set_embeddings(ctx: llama_context_p, embeddings: bool, /):
+    """Set whether the model is in embeddings model or not
+    If true, embeddings will be returned but logits will not"""
+    ...
+
+
 # // Set whether to use causal attention or not
 # // If set to true, the model will only attend to the past tokens
 # LLAMA_API void llama_set_causal_attn(struct llama_context * ctx, bool causal_attn);