Set triton kernels as default. Remove deprecated config fields.

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "/nfs-gpu/xlstm/converted_model_checkpoints/dclm_mLSTMv1_7B_ctx8192_sep_finetune_2024-11-29T17:03:51_0_550000",
+  "_name_or_path": "NX-AI/xLSTM-7b",
   "add_embedding_dropout": false,
   "add_forward_backend_padding": false,
   "add_out_norm": true,
@@ -9,18 +9,23 @@
   "architectures": [
     "xLSTMForCausalLM"
   ],
+  "autocast_kernel_dtype": "bfloat16",
   "bos_token_id": 0,
   "cell_norm_eps": 1e-06,
+  "chunk_size": 64,
+  "chunkwise_kernel": "chunkwise--triton_xl_chunk",
   "embedding_dim": 4096,
   "eos_token_id": 2,
+  "eps": 1e-06,
   "ffn_proj_factor": 2.667,
   "ffn_round_up_to_multiple_of": 64,
   "force_bos_token_insert": true,
-  "forward_backend_name": "chunkwise--triton_limit_chunk",
   "gate_soft_cap": 15.0,
   "head_dim": 512,
   "igate_bias_init_range": -10.0,
+  "inference_state_dtype": "float32",
   "mlstm_round_up_to_multiple_of": 64,
+  "mode": "inference",
   "model_type": "xlstm",
   "norm_eps": 1e-06,
   "norm_reduction_force_float32": true,
@@ -30,12 +35,14 @@
   "pad_token_id": 1,
   "qk_dim_factor": 0.5,
   "return_last_states": true,
-  "step_backend_name": "triton_fused",
+  "sequence_kernel": "native_sequence__triton",
+  "step_kernel": "triton",
   "tie_word_embeddings": false,
   "torch_dtype": "float32",
-  "transformers_version": "4.47.0.dev0",
+  "transformers_version": "4.48.0.dev0",
   "use_bias": false,
   "use_cache": true,
   "v_dim_factor": 1.0,
-  "vocab_size": 50304
+  "vocab_size": 50304,
+  "weight_mode": "single"
 }