From a06bff505201a043613529293b58a1b13ee85a03 Mon Sep 17 00:00:00 2001 From: Luis Vega Date: Wed, 16 Apr 2025 21:03:07 -0700 Subject: [PATCH] Fix rotary_emb param in NemotronH attention (#3646) Signed-off-by: Luis Vega --- tensorrt_llm/_torch/models/modeling_nemotron_h.py | 1 - 1 file changed, 1 deletion(-) diff --git a/tensorrt_llm/_torch/models/modeling_nemotron_h.py b/tensorrt_llm/_torch/models/modeling_nemotron_h.py index 611781e19e..9fc148dbfa 100644 --- a/tensorrt_llm/_torch/models/modeling_nemotron_h.py +++ b/tensorrt_llm/_torch/models/modeling_nemotron_h.py @@ -77,7 +77,6 @@ class TransformerLayer(Attention): num_key_value_heads=config.num_key_value_heads, max_position_embeddings=config.max_position_embeddings, bias=config.attention_bias, - rotary_emb=None, pos_embd_params=None, layer_idx=layer_idx, dtype=config.torch_dtype,