diff --git a/convert-llama-h5-to-gguf.py b/convert-llama-h5-to-gguf.py
index d7706f618..56063f375 100644
--- a/convert-llama-h5-to-gguf.py
+++ b/convert-llama-h5-to-gguf.py
@@ -95,12 +95,21 @@ if "_name_or_path" in hparams:
 else:
     hf_repo=""
 
+if "max_sequence_length" in hparams:
+    ctx_length = hparams["max_sequence_length"]
+elif "max_position_embeddings" in hparams:
+    ctx_length = hparams["max_position_embeddings"]
+else:
+    print("gguf: can not find ctx length parameter.")
+    sys.exit()
+
+
 gguf_writer.add_architecture(llm_arch)
 gguf_writer.add_name(last_dir)
 gguf_writer.add_file_type("All tensors F32" if ftype == 0 else "Most tensors F16, some F32")
 gguf_writer.add_source_hf_repo(hf_repo)
 gguf_writer.add_tensor_data_layout(llm_arch, "Meta AI original pth")
-gguf_writer.add_context_length(llm_arch, hparams["max_position_embeddings"])
+gguf_writer.add_context_length(llm_arch, ctx_length)
 gguf_writer.add_embedding_length(llm_arch, hparams["hidden_size"])
 gguf_writer.add_block_count(llm_arch, block_count)
 gguf_writer.add_feed_forward_length(llm_arch, hparams["intermediate_size"])