convert : support is_causal hyperparameter (#20746)

* convert : support is_causal hyperparameter

Check for the `is_causal` attribute in the Hugging Face model configuration and include it in the GGUF metadata.

* Update convert_hf_to_gguf.py

Co-authored-by: Sigbjørn Skjæret <redacted>
* style: fix F541 f-string is missing placeholders

---------

Co-authored-by: Sigbjørn Skjæret <redacted>

diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index 46469c862000065d2b5892a0388134a7e55e8b6e..8cfd0bf2f536f9e8710c3f1d1fced915f63aa9d9 100755 (executable)
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -1062,6 +1062,10 @@ class TextModel(ModelBase):
             self.gguf_writer.add_head_count_kv(n_head_kv)
             logger.info(f"gguf: key-value head count = {n_head_kv}")
 
+        if self.hparams.get("is_causal") is False:
+            self.gguf_writer.add_causal_attention(False)
+            logger.info("gguf: causal attention = False")
+
         # TODO: Handle "sliding_attention" similarly when models start implementing it
         rope_params = self.rope_parameters.get("full_attention", self.rope_parameters)
         if (rope_type := rope_params.get("rope_type")) is not None: