Merge branch 'master' into xsn/server_mtmd

Author: ngxson

convert_hf_to_gguf.py

@@ -5686,7 +5686,12 @@ def set_gguf_parameters(self):
         rope_dim = hparams.get("head_dim") or hparams["hidden_size"] // hparams["num_attention_heads"]
 
         self.gguf_writer.add_rope_dimension_count(rope_dim)
-        self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.NONE)
+        if (self.hparams.get("rope_scaling") or {}).get("type") == "yarn" and "factor" in self.hparams["rope_scaling"]:
+            self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.YARN)
+            self.gguf_writer.add_rope_scaling_factor(self.hparams["rope_scaling"]["factor"])
+            self.gguf_writer.add_rope_scaling_orig_ctx_len(self.hparams["rope_scaling"]["original_max_position_embeddings"])
+        else:
+            self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.NONE)
         self.gguf_writer.add_leading_dense_block_count(hparams["first_k_dense_replace"])
         self.gguf_writer.add_vocab_size(hparams["vocab_size"])
         self.gguf_writer.add_expert_feed_forward_length(hparams["moe_intermediate_size"])

ggml/src/ggml-sycl/ggml-sycl.cpp

@@ -3873,6 +3873,9 @@ static bool ggml_backend_sycl_device_supports_op(ggml_backend_dev_t dev, const g
                 if (a->ne[3] != b->ne[3]) {
                     return false;
                 }
+                if (!ggml_is_contiguous(b)) {
+                    return false;
+                }
                 ggml_type a_type = a->type;
                 if (a_type == GGML_TYPE_IQ4_NL  || a_type == GGML_TYPE_IQ4_XS ||
                     a_type == GGML_TYPE_IQ3_XXS || a_type == GGML_TYPE_IQ3_S  ||