graph : fix equal_seq() check (#14986)

ggml-ci

Author: ggerganov

src/llama-context.cpp

@@ -113,6 +113,15 @@ llama_context::llama_context(
         }
     }
 
+    {
+        const char * LLAMA_GRAPH_REUSE_DISABLE = getenv("LLAMA_GRAPH_REUSE_DISABLE");
+        graph_reuse_disable = LLAMA_GRAPH_REUSE_DISABLE ? (atoi(LLAMA_GRAPH_REUSE_DISABLE) != 0) : graph_reuse_disable;
+
+        if (graph_reuse_disable) {
+            LLAMA_LOG_WARN("%s: graph reuse disabled\n", __func__);
+        }
+    }
+
     const uint32_t n_ctx_per_seq = cparams.n_ctx / cparams.n_seq_max;
 
     LLAMA_LOG_INFO("%s: n_seq_max     = %u\n",   __func__, cparams.n_seq_max);
@@ -716,7 +725,7 @@ llm_graph_result * llama_context::process_ubatch(const llama_ubatch & ubatch, ll
     // in order to correctly reuse a graph, it's full topology has to be uniquely determined by these parameters
     const auto gparams = graph_params(res, ubatch, mctx, gtype);
 
-    if (res->can_reuse(gparams)) {
+    if (!graph_reuse_disable && res->can_reuse(gparams)) {
         //LLAMA_LOG_DEBUG("%s: reusing previous graph\n", __func__);
 
         n_reused++;

src/llama-context.h

@@ -291,6 +291,9 @@ struct llama_context {
     // ref: https://github.com/ggml-org/llama.cpp/pull/14285
     bool supports_set_rows = false;
 
+    // env: LLAMA_GRAPH_REUSE_DISABLE
+    bool graph_reuse_disable = false;
+
     // perf
     mutable int64_t t_start_us  = 0;
     mutable int64_t t_load_us   = 0;

src/llama-graph.h

@@ -423,7 +423,9 @@ struct llm_graph_params {
                 (!ubatch.embd  && !other.ubatch.embd)
             );
 
-        if (can_reuse_ubatch && !ubatch.equal_seqs()) {
+        // when we split the batch using "equal_seqs" we have to verify that the participating sequences are the same
+        //   the reason is because the set of attention streams would be different for different sequences
+        if (can_reuse_ubatch && ubatch.equal_seqs()) {
             if (!ubatch.data) {
                 // if the old ubatch does not own it's data, then we cannot guarantee that it is still alive, and
                 //   therefore we cannot perform the sequence id check. normally should never happen