Skip to content

Commit

Permalink
Remove deprecated builder_opt param from build command (#11259)
Browse files Browse the repository at this point in the history
Signed-off-by: Jan Lasek <janek.lasek@gmail.com>
  • Loading branch information
janekl authored and yashaswikarnati committed Nov 21, 2024
1 parent 0574b12 commit 9548ebb
Showing 1 changed file with 0 additions and 3 deletions.
3 changes: 0 additions & 3 deletions nemo/export/trt_llm/qnemo/qnemo_to_tensorrt_llm.py
Original file line number Diff line number Diff line change
Expand Up @@ -77,8 +77,6 @@ def qnemo_to_tensorrt_llm(

use_qdq = quant_algo in ["FP8", "W8A8_SQ_PER_CHANNEL"]

builder_opt = 4 if "RecurrentGemma" not in config.architecture else 0

speculative_decoding_mode = "medusa" if "Medusa" in config.architecture else None

build_cmd = "trtllm-build "
Expand All @@ -90,7 +88,6 @@ def qnemo_to_tensorrt_llm(
build_cmd += f"--max_input_len {max_input_len} "
build_cmd += f"--max_beam_width {max_beam_width} "
build_cmd += f"--max_prompt_embedding_table_size {max_prompt_embedding_table_size} "
build_cmd += f"--builder_opt {builder_opt} "
build_cmd += f"--paged_kv_cache {'enable' if paged_kv_cache else 'disable'} "
build_cmd += f"--use_paged_context_fmha {'enable' if paged_context_fmha else 'disable'} "
build_cmd += f"--remove_input_padding {'enable' if remove_input_padding else 'disable'} "
Expand Down

0 comments on commit 9548ebb

Please sign in to comment.