From: Georgi Gerganov <redacted>
Date: Thu, 24 Apr 2025 19:29:22 +0000 (+0300)
Subject: embeddings : fix batch sizes (#13076)
X-Git-Tag: upstream/0.0.5185^0
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=226251ed56b85190e18a1cca963c45b888f4953c;p=pkg%2Fggml%2Fsources%2Fllama.cpp

embeddings : fix batch sizes (#13076)

ggml-ci
---

diff --git a/examples/embedding/embedding.cpp b/examples/embedding/embedding.cpp
index 6f089041..06fce236 100644
--- a/examples/embedding/embedding.cpp
+++ b/examples/embedding/embedding.cpp
@@ -89,6 +89,13 @@ int main(int argc, char ** argv) {
     common_init();
 
     params.embedding = true;
+
+    // utilize the full context
+    if (params.n_batch < params.n_ctx) {
+        LOG_WRN("%s: setting batch size to %d\n", __func__, params.n_ctx);
+        params.n_batch = params.n_ctx;
+    }
+
     // For non-causal models, batch size must be equal to ubatch size
     params.n_ubatch = params.n_batch;
 
@@ -134,7 +141,6 @@ int main(int argc, char ** argv) {
 
     // max batch size
     const uint64_t n_batch = params.n_batch;
-    GGML_ASSERT(params.n_batch >= params.n_ctx);
 
     // tokenize the prompts and trim
     std::vector<std::vector<int32_t>> inputs;