Skip to content

LLM inference in C/C++

C++ ggml

spec : remove tokens assert from dsv4 graph_mtp (#30444)

Oliver Simons296538adea19
diff --git a/src/models/deepseek4.cpp b/src/models/deepseek4.cpp
index 5df4c04c54..50864b2cbe 100644
--- a/src/models/deepseek4.cpp
+++ b/src/models/deepseek4.cpp
@@ -1366,7 +1366,6 @@ llama_model_deepseek4::graph_mtp::graph_mtp(const llama_model & model, const llm
cparams.nextn_layer_offset < (int) hparams.n_layer_nextn &&
"nextn_layer_offset out of range [0, n_layer_nextn)");
ASSERT_EMBD_OR_TOKEN(ubatch);
- GGML_ASSERT(ubatch.token && "DEEPSEEK4 MTP requires token input");
const int64_t hc = hparams.dsv4_hc_mult;
GGML_ASSERT(hparams.n_embd_out() == (uint32_t) (n_embd*hc) && "DEEPSEEK4 MTP hidden width mismatch");