From 6e08281e588bbba1a5d180290a94a43f167f3a1a Mon Sep 17 00:00:00 2001
From: Kerfuffle <44031344+KerfuffleV2@users.noreply.github.com>
Date: Sun, 29 Oct 2023 11:31:40 -0600
Subject: Extend llama_kv_cache_seq_rm to allow matching any sequence (#3843)

* Extend llama_kv_cache_seq_rm to allow matichng any sequence

* Replace llama_kv_cache_tokens_rm with llama_kv_cache_clear

Use llama_kv_cache_clear for cache clearing

Change calls to llama_kv_cache_tokens_rm that want to delete by position to use llama_kv_cache_seq_rm functionality
---
 examples/batched-bench/batched-bench.cpp | 2 +-
 examples/llama-bench/llama-bench.cpp     | 4 ++--
 examples/main/main.cpp                   | 2 +-
 examples/perplexity/perplexity.cpp       | 6 +++---
 examples/server/server.cpp               | 2 +-
 5 files changed, 8 insertions(+), 8 deletions(-)

(limited to 'examples')

diff --git a/examples/batched-bench/batched-bench.cpp b/examples/batched-bench/batched-bench.cpp
index 43f9c971..533c55c1 100644
--- a/examples/batched-bench/batched-bench.cpp
+++ b/examples/batched-bench/batched-bench.cpp
@@ -185,7 +185,7 @@ int main(int argc, char ** argv) {
 
                 const auto t_pp_start = ggml_time_us();
 
-                llama_kv_cache_tokens_rm(ctx, -1, -1);
+                llama_kv_cache_clear(ctx);
 
                 if (!decode_helper(ctx, batch, ctx_params.n_batch)) {
                     LOG_TEE("%s: llama_decode() failed\n", __func__);
diff --git a/examples/llama-bench/llama-bench.cpp b/examples/llama-bench/llama-bench.cpp
index 20767d55..78039818 100644
--- a/examples/llama-bench/llama-bench.cpp
+++ b/examples/llama-bench/llama-bench.cpp
@@ -1037,7 +1037,7 @@ int main(int argc, char ** argv) {
 
         test t(inst, lmodel, ctx);
 
-        llama_kv_cache_tokens_rm(ctx, -1, -1);
+        llama_kv_cache_clear(ctx);
 
         // warmup run
         if (t.n_prompt > 0) {
@@ -1048,7 +1048,7 @@ int main(int argc, char ** argv) {
         }
 
         for (int i = 0; i < params.reps; i++) {
-            llama_kv_cache_tokens_rm(ctx, -1, -1);
+            llama_kv_cache_clear(ctx);
 
             uint64_t t_start = get_time_ns();
             if (t.n_prompt > 0) {
diff --git a/examples/main/main.cpp b/examples/main/main.cpp
index 3d9f670b..8a43b6ab 100644
--- a/examples/main/main.cpp
+++ b/examples/main/main.cpp
@@ -298,7 +298,7 @@ int main(int argc, char ** argv) {
         }
 
         // remove any "future" tokens that we might have inherited from the previous session
-        llama_kv_cache_tokens_rm(ctx, n_matching_session_tokens, -1);
+        llama_kv_cache_seq_rm(ctx, -1, n_matching_session_tokens, -1);
     }
 
     LOGLN(
diff --git a/examples/perplexity/perplexity.cpp b/examples/perplexity/perplexity.cpp
index 3c2542e8..bd2c73d8 100644
--- a/examples/perplexity/perplexity.cpp
+++ b/examples/perplexity/perplexity.cpp
@@ -210,7 +210,7 @@ static results_perplexity perplexity_v2(llama_context * ctx, const gpt_params &
         const auto t_start = std::chrono::high_resolution_clock::now();
 
         // clear the KV cache
-        llama_kv_cache_tokens_rm(ctx, -1, -1);
+        llama_kv_cache_clear(ctx);
 
         for (int j = 0; j < num_batches; ++j) {
             const int batch_start = start + j * n_batch;
@@ -339,7 +339,7 @@ static results_perplexity perplexity(llama_context * ctx, const gpt_params & par
         const auto t_start = std::chrono::high_resolution_clock::now();
 
         // clear the KV cache
-        llama_kv_cache_tokens_rm(ctx, -1, -1);
+        llama_kv_cache_clear(ctx);
 
         for (int j = 0; j < num_batches; ++j) {
             const int batch_start = start + j * n_batch;
@@ -573,7 +573,7 @@ static void hellaswag_score(llama_context * ctx, const gpt_params & params) {
         }
 
         // clear the KV cache
-        llama_kv_cache_tokens_rm(ctx, -1, -1);
+        llama_kv_cache_clear(ctx);
 
         auto logits = hellaswag_evaluate_tokens(ctx, query_embd, 0, params.n_batch, n_vocab);
         if (logits.empty()) {
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index 5b7e4139..c163c7f8 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -857,7 +857,7 @@ struct llama_server_context
 
     void kv_cache_clear() {
         // clear the entire KV cache
-        llama_kv_cache_tokens_rm(ctx, -1, -1);
+        llama_kv_cache_clear(ctx);
         clean_kv_cache = false;
     }
 
-- 
cgit v1.2.3