Add Llama3.2 1B as an new example model

.ci/scripts/test_eval_llama_mmlu.sh

@@ -35,6 +35,7 @@ run_and_verify() {
         exit 1
     fi
     $PYTHON_EXECUTABLE -m examples.models.llama.eval_llama \
+	--model llama2 \
 	-c stories110M.pt \
 	-p params.json \
 	-t tokenizer.model \

.ci/scripts/test_eval_llama_wikitext.sh

@@ -35,6 +35,7 @@ run_and_verify() {
         exit 1
     fi
     $PYTHON_EXECUTABLE -m examples.models.llama.eval_llama \
+	--model llama2 \
 	-c stories110M.pt \
 	-p params.json \
 	-t tokenizer.model \

.ci/scripts/test_llama.sh

@@ -206,7 +206,7 @@ if [[ "${QNN}" == "ON" ]]; then
   EXPORT_ARGS="${EXPORT_ARGS} -kv -v --qnn --disable_dynamic_shape"
 fi
 # Add dynamically linked library location
-$PYTHON_EXECUTABLE -m examples.models.llama.export_llama ${EXPORT_ARGS}
+$PYTHON_EXECUTABLE -m examples.models.llama.export_llama --model llama3 ${EXPORT_ARGS}
 
 # Create tokenizer.bin.
 echo "Creating tokenizer.bin"

.ci/scripts/test_llama_runner_eager.sh

@@ -35,6 +35,7 @@ run_and_verify() {
         exit 1
     fi
     $PYTHON_EXECUTABLE -m examples.models.llama.runner.eager \
+	--model llama2 \
 	-c stories110M.pt \
 	-p params.json \
 	-t tokenizer.model \

.ci/scripts/test_model.sh

@@ -77,7 +77,7 @@ test_model() {
     # Install requirements for export_llama
     bash examples/models/llama/install_requirements.sh
     # Test export_llama script: python3 -m examples.models.llama.export_llama
-    "${PYTHON_EXECUTABLE}" -m examples.models.llama.export_llama -c examples/models/llama/params/demo_rand_params.pth -p examples/models/llama/params/demo_config.json
+    "${PYTHON_EXECUTABLE}" -m examples.models.llama.export_llama --model llama2 -c examples/models/llama/params/demo_rand_params.pth -p examples/models/llama/params/demo_config.json
     run_portable_executor_runner
     rm "./${MODEL_NAME}.pte"
   fi

backends/vulkan/docs/android_demo.md

@@ -58,6 +58,7 @@ partially lower the Llama model to Vulkan.
 ```shell
 # The files will usually be downloaded to ~/.llama
 python -m examples.models.llama.export_llama \
+  --model llama3_2
   --disable_dynamic_shape --vulkan -kv --use_sdpa_with_kv_cache -d fp32 \
   -c ~/.llama/checkpoints/Llama3.2-1B/consolidated.00.pth \
   -p ~/.llama/checkpoints/Llama3.2-1B/params.json \

docs/source/llm/build-run-llama3-qualcomm-ai-engine-direct-backend.md

@@ -39,7 +39,7 @@ To export Llama 3 8B instruct with the Qualcomm AI Engine Direct Backend, ensure
 
 ```bash
 # Please note that calibration_data must include the prompt template for special tokens.
-python -m examples.models.llama.export_llama  -t <path_to_tokenizer.model>
+python -m examples.models.llama.export_llama --model llama3  -t <path_to_tokenizer.model>
 llama3/Meta-Llama-3-8B-Instruct/tokenizer.model -p <path_to_params.json> -c <path_to_checkpoint_for_Meta-Llama-3-8B-Instruct>  --use_kv_cache  --qnn --pt2e_quantize qnn_16a4w --disable_dynamic_shape --num_sharding 8 --calibration_tasks wikitext --calibration_limit 1 --calibration_seq_length 128 --optimized_rotation_path <path_to_optimized_matrix> --calibration_data "<|start_header_id|>system<|end_header_id|>\n\nYou are a funny chatbot.<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nCould you tell me about Facebook?<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
 ```
 

examples/demo-apps/android/LlamaDemo/docs/delegates/qualcomm_README.md

@@ -101,12 +101,12 @@ We support PTQ by default. The entire export may take ~20 minutes (Llama 3.1 8B)
 Examples:
 ```
 # 4 bits weight only quantize
-python -m examples.models.llama.export_llama --checkpoint "${MODEL_DIR}/consolidated.00.pth" -p "${MODEL_DIR}/params.json" -kv --disable_dynamic_shape --qnn --pt2e_quantize qnn_16a4w -d fp32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="test.pte”
+python -m examples.models.llama.export_llama --model llama3 --checkpoint "${MODEL_DIR}/consolidated.00.pth" -p "${MODEL_DIR}/params.json" -kv --disable_dynamic_shape --qnn --pt2e_quantize qnn_16a4w -d fp32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="test.pte”
 ```
 If the model is really big, it may require model sharding because the Qualcomm DSP is a 32bit system and has a 4GB size limit . For example for Llama 3 8B models, we need to shard the model into 4, but ExecuTorch still packages it into one PTE file. Here is an example:
 ```
 # 8 bits quantization with 4 shards
-python -m examples.models.llama.export_llama --checkpoint "${MODEL_DIR}/consolidated.00.pth" -p "${MODEL_DIR}/params.json" -kv --disable_dynamic_shape --qnn --pt2e_quantize qnn_8a8w -d fp32 --num_sharding 4 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="test.pte”
+python -m examples.models.llama.export_llama --model llama3 --checkpoint "${MODEL_DIR}/consolidated.00.pth" -p "${MODEL_DIR}/params.json" -kv --disable_dynamic_shape --qnn --pt2e_quantize qnn_8a8w -d fp32 --num_sharding 4 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="test.pte”
 ```
 Note: if you encountered issues below
 ```
@@ -158,7 +158,7 @@ To export Llama 3 8B instruct with the Qualcomm AI Engine Direct Backend, ensure
 * 8B models might need 16GB RAM on the device to run.
 ```
 # Please note that calibration_data must include the prompt template for special tokens.
-python -m examples.models.llama.export_llama  -t <path_to_tokenizer.model> -p <path_to_params.json> -c <path_to_checkpoint_for_Meta-Llama-3-8B-Instruct>  --use_kv_cache  --qnn --pt2e_quantize qnn_16a4w --disable_dynamic_shape --num_sharding 8 --calibration_tasks wikitext --calibration_limit 1 --calibration_seq_length 128 --optimized_rotation_path <path_to_optimized_matrix> --calibration_data "<|start_header_id|>system<|end_header_id|>\n\nYou are a funny chatbot.<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nCould you tell me about Facebook?<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
+python -m examples.models.llama.export_llama --model llama3 -t <path_to_tokenizer.model> -p <path_to_params.json> -c <path_to_checkpoint_for_Meta-Llama-3-8B-Instruct>  --use_kv_cache  --qnn --pt2e_quantize qnn_16a4w --disable_dynamic_shape --num_sharding 8 --calibration_tasks wikitext --calibration_limit 1 --calibration_seq_length 128 --optimized_rotation_path <path_to_optimized_matrix> --calibration_data "<|start_header_id|>system<|end_header_id|>\n\nYou are a funny chatbot.<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nCould you tell me about Facebook?<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
 ```
 
 ## Pushing Model and Tokenizer

examples/demo-apps/android/LlamaDemo/docs/delegates/xnnpack_README.md

@@ -56,14 +56,14 @@ In this demo app, we support text-only inference with up-to-date Llama models an
 Meta has released prequantized INT4 SpinQuant Llama 3.2 models that ExecuTorch supports on the XNNPACK backend.
 * Export Llama model and generate .pte file as below:
 ```
-python -m examples.models.llama.export_llama --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -d fp32 --xnnpack-extended-ops --preq_mode 8da4w_output_8da8w --preq_group_size 32 --max_seq_length 2048 --preq_embedding_quantize 8,0 --use_spin_quant native --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name "llama3_2_spinquant.pte"
+python -m examples.models.llama.export_llama --model llama3_2 --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -d fp32 --xnnpack-extended-ops --preq_mode 8da4w_output_8da8w --preq_group_size 32 --max_seq_length 2048 --preq_embedding_quantize 8,0 --use_spin_quant native --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name "llama3_2_spinquant.pte"
 ```
 
 ### For Llama 3.2 1B and 3B QAT+LoRA models
 Meta has released prequantized INT4 QAT+LoRA Llama 3.2 models that ExecuTorch supports on the XNNPACK backend.
 * Export Llama model and generate .pte file as below:
 ```
-python -m examples.models.llama.export_llama --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -qat -lora 16 -kv --use_sdpa_with_kv_cache -X -d fp32 --xnnpack-extended-ops --preq_mode 8da4w_output_8da8w --preq_group_size 32 --max_seq_length 2048 --preq_embedding_quantize 8,0 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name "llama3_2_qat_lora.pte"
+python -m examples.models.llama.export_llama --model llama3_2 --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -qat -lora 16 -kv --use_sdpa_with_kv_cache -X -d fp32 --xnnpack-extended-ops --preq_mode 8da4w_output_8da8w --preq_group_size 32 --max_seq_length 2048 --preq_embedding_quantize 8,0 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name "llama3_2_qat_lora.pte"
 ```
 
 ### For Llama 3.2 1B and 3B BF16 models
@@ -72,7 +72,7 @@ We have supported BF16 as a data type on the XNNPACK backend for Llama 3.2 1B/3B
 * Export Llama model and generate .pte file as below:
 
 ```
-python -m examples.models.llama.export_llama --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -d bf16 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama3_2_bf16.pte"
+python -m examples.models.llama.export_llama --model llama3_2 --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -d bf16 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama3_2_bf16.pte"
 ```
 
 For more detail using Llama 3.2 lightweight models including prompt template, please go to our official [website](https://www.llama.com/docs/model-cards-and-prompt-formats/llama3_2#-llama-3.2-lightweight-models-(1b/3b)-).
@@ -87,7 +87,7 @@ To safeguard your application, you can use our Llama Guard models for prompt cla
 * We prepared this model using the following command
 
 ```
-python -m examples.models.llama.export_llama --checkpoint <path-to-pruned-llama-guard-1b-checkpoint.pth> --params <path-to-your-params.json> -d fp32 -kv --use_sdpa_with_kv_cache --quantization_mode 8da4w --group_size 256 --xnnpack --max_seq_length 8193 --embedding-quantize 4,32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_prune_map <path-to-your-llama_guard-pruned-layers-map.json> --output_name="llama_guard_3_1b_pruned_xnnpack.pte"
+python -m examples.models.llama.export_llama --model llama3_2 --checkpoint <path-to-pruned-llama-guard-1b-checkpoint.pth> --params <path-to-your-params.json> -d fp32 -kv --use_sdpa_with_kv_cache --quantization_mode 8da4w --group_size 256 --xnnpack --max_seq_length 8193 --embedding-quantize 4,32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_prune_map <path-to-your-llama_guard-pruned-layers-map.json> --output_name="llama_guard_3_1b_pruned_xnnpack.pte"
 ```
 
 
@@ -97,7 +97,7 @@ python -m examples.models.llama.export_llama --checkpoint <path-to-pruned-llama-
 * Export Llama model and generate .pte file as below:
 
 ```
-python -m examples.models.llama.export_llama --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -qmode 8da4w --group_size 128 -d fp32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama.pte"
+python -m examples.models.llama.export_llama --model llama3_2 --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -qmode 8da4w --group_size 128 -d fp32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama.pte"
 ```
 
 You may wonder what the ‘--metadata’ flag is doing. This flag helps export the model with proper special tokens added that the runner can detect EOS tokens easily.

examples/demo-apps/apple_ios/LLaMA/docs/delegates/mps_README.md

@@ -45,9 +45,9 @@ Install the required packages to export the model
 sh examples/models/llama/install_requirements.sh
 ```
 
-Export the model
+Export the model (Llama 3 in this case)
 ```
-python -m examples.models.llama.export_llama --checkpoint "${MODEL_DIR}/consolidated.00.pth" --params "${MODEL_DIR}/params.json" -kv --use_sdpa_with_kv_cache --mps -d fp32 --disable_dynamic_shape -qmode 8da4w -G 32
+python -m examples.models.llama.export_llama --model llama3 --checkpoint "${MODEL_DIR}/consolidated.00.pth" --params "${MODEL_DIR}/params.json" -kv --use_sdpa_with_kv_cache --mps -d fp32 --disable_dynamic_shape -qmode 8da4w -G 32
 ```
 
 ## Pushing Model and Tokenizer

examples/demo-apps/apple_ios/LLaMA/docs/delegates/xnnpack_README.md

@@ -48,14 +48,14 @@ sh examples/models/llama/install_requirements.sh
 Meta has released prequantized INT4 SpinQuant Llama 3.2 models that ExecuTorch supports on the XNNPACK backend.
 * Export Llama model and generate .pte file as below:
 ```
-python -m examples.models.llama.export_llama --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -d fp32 --xnnpack-extended-ops --preq_mode 8da4w_output_8da8w --preq_group_size 32 --max_seq_length 2048 --preq_embedding_quantize 8,0 --use_spin_quant native --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name "llama3_2_spinquant.pte"
+python -m examples.models.llama.export_llama --model llama3_2 --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -d fp32 --xnnpack-extended-ops --preq_mode 8da4w_output_8da8w --preq_group_size 32 --max_seq_length 2048 --preq_embedding_quantize 8,0 --use_spin_quant native --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name "llama3_2_spinquant.pte"
 ```
 
 ### For Llama 3.2 1B and 3B QAT+LoRA models
 Meta has released prequantized INT4 QAT+LoRA Llama 3.2 models that ExecuTorch supports on the XNNPACK backend.
 * Export Llama model and generate .pte file as below:
 ```
-python -m examples.models.llama.export_llama --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -qat -lora 16 -kv --use_sdpa_with_kv_cache -X -d fp32 --xnnpack-extended-ops --preq_mode 8da4w_output_8da8w --preq_group_size 32 --max_seq_length 2048 --preq_embedding_quantize 8,0 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name "llama3_2_qat_lora.pte"
+python -m examples.models.llama.export_llama --model llama3_2 --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -qat -lora 16 -kv --use_sdpa_with_kv_cache -X -d fp32 --xnnpack-extended-ops --preq_mode 8da4w_output_8da8w --preq_group_size 32 --max_seq_length 2048 --preq_embedding_quantize 8,0 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name "llama3_2_qat_lora.pte"
 ```
 
 ### For Llama 3.2 1B and 3B BF16 models
@@ -64,7 +64,7 @@ We have supported BF16 as a data type on the XNNPACK backend for Llama 3.2 1B/3B
 * Export Llama model and generate .pte file as below:
 
 ```
-python -m examples.models.llama.export_llama --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -d bf16 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama3_2_bf16.pte"
+python -m examples.models.llama.export_llama --model llama3_2 --checkpoint <path-to-your-checkpoint.pth> --params <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -d bf16 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama3_2_bf16.pte"
 ```
 
 For more detail using Llama 3.2 lightweight models including prompt template, please go to our official [website](https://www.llama.com/docs/model-cards-and-prompt-formats/llama3_2#-llama-3.2-lightweight-models-(1b/3b)-).
@@ -73,7 +73,7 @@ For more detail using Llama 3.2 lightweight models including prompt template, pl
 
 Export the model
 ```
-python -m examples.models.llama.export_llama --checkpoint <path-to-your-checkpoint.pth> -p <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -qmode 8da4w  --group_size 128 -d fp32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --embedding-quantize 4,32 --output_name="llama3_kv_sdpa_xnn_qe_4_32.pte"
+python -m examples.models.llama.export_llama --model llama3_2 --checkpoint <path-to-your-checkpoint.pth> -p <path-to-your-params.json> -kv --use_sdpa_with_kv_cache -X -qmode 8da4w  --group_size 128 -d fp32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --embedding-quantize 4,32 --output_name="llama3_kv_sdpa_xnn_qe_4_32.pte"
 ```
 
 ### For LLaVA model

examples/models/llama/README.md

@@ -166,6 +166,7 @@ LLAMA_CHECKPOINT=path/to/checkpoint.pth
 LLAMA_PARAMS=path/to/params.json
 
 python -m examples.models.llama.export_llama \
+  --model llama3_2
   --checkpoint "${LLAMA_CHECKPOINT:?}" \
   --params "${LLAMA_PARAMS:?}" \
   -kv \
@@ -187,6 +188,7 @@ LLAMA_QUANTIZED_CHECKPOINT=path/to/spinquant/checkpoint.pth
 LLAMA_PARAMS=path/to/spinquant/params.json
 
 python -m examples.models.llama.export_llama \
+   --model llama3_2
    --checkpoint "${LLAMA_QUANTIZED_CHECKPOINT:?}" \
    --params "${LLAMA_PARAMS:?}" \
    --use_sdpa_with_kv_cache \
@@ -212,6 +214,7 @@ LLAMA_QUANTIZED_CHECKPOINT=path/to/qlora/checkpoint.pth
 LLAMA_PARAMS=path/to/qlora/params.json
 
 python -m examples.models.llama.export_llama \
+   --model llama3_2
    --checkpoint "${LLAMA_QUANTIZED_CHECKPOINT:?}" \
    --params "${LLAMA_PARAMS:?}" \
    -qat \
@@ -237,9 +240,20 @@ You can export and run the original Llama 3 8B instruct model.
 
 2. Export model and generate `.pte` file
     ```
-    python -m examples.models.llama.export_llama --checkpoint <consolidated.00.pth> -p <params.json> -kv --use_sdpa_with_kv_cache -X -qmode 8da4w  --group_size 128 -d fp32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --embedding-quantize 4,32 --output_name="llama3_kv_sdpa_xnn_qe_4_32.pte"
+    python -m examples.models.llama.export_llama
+	    --model llama3
+	    --checkpoint <consolidated.00.pth>
+		-p <params.json>
+		-kv
+		--use_sdpa_with_kv_cache
+		-X
+		-qmode 8da4w
+		--group_size 128
+		-d fp32
+		--metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}'
+		--embedding-quantize 4,32
+		--output_name="llama3_kv_sdpa_xnn_qe_4_32.pte"
     ```
-
     Due to the larger vocabulary size of Llama 3, we recommend quantizing the embeddings with `--embedding-quantize 4,32` as shown above to further reduce the model size.
 
 ## Step 3: Run on your computer to validate

examples/models/llama/UTILS.md

@@ -19,17 +19,17 @@ From `executorch` root:
     ```
 3. Export model and generate `.pte` file.
     ```
-    python -m examples.models.llama.export_llama -c stories110M.pt -p params.json -X -kv
+    python -m examples.models.llama.export_llama --model llama3 -c stories110M.pt -p params.json -X -kv
     ```
 
 ## Smaller model delegated to other backends
 
 Currently we supported lowering the stories model to other backends, including, CoreML, MPS and QNN. Please refer to the instruction
 for each backend ([CoreML](https://pytorch.org/executorch/main/build-run-coreml.html), [MPS](https://pytorch.org/executorch/main/build-run-mps.html), [QNN](https://pytorch.org/executorch/main/build-run-qualcomm-ai-engine-direct-backend.html)) before trying to lower them. After the backend library is installed, the script to export a lowered model is
 
-- Lower to CoreML: `python -m examples.models.llama.export_llama -kv --disable_dynamic_shape --coreml -c stories110M.pt -p params.json `
-- MPS: `python -m examples.models.llama.export_llama -kv --disable_dynamic_shape --mps -c stories110M.pt -p params.json `
-- QNN: `python -m examples.models.llama.export_llama -kv --disable_dynamic_shape --qnn -c stories110M.pt -p params.json `
+- Lower to CoreML: `python -m examples.models.llama.export_llama --model llama3 -kv --disable_dynamic_shape --coreml -c stories110M.pt -p params.json `
+- MPS: `python -m examples.models.llama.export_llama --model llama3 -kv --disable_dynamic_shape --mps -c stories110M.pt -p params.json `
+- QNN: `python -m examples.models.llama.export_llama --model llama3 -kv --disable_dynamic_shape --qnn -c stories110M.pt -p params.json `
 
 The iOS LLAMA app supports the CoreML and MPS model and the Android LLAMA app supports the QNN model. On Android, it also allow to cross compiler the llama runner binary, push to the device and run.
 

examples/models/llama/eval_llama_lib.py

@@ -191,7 +191,7 @@ def gen_eval_wrapper(
 
     pt2e_quant_params, quantizers, quant_dtype = get_quantizer_and_quant_params(args)
     # GPTFastEvalWrapper: Create a wrapper around a pre-exported model
-    manager: LLMEdgeManager = _prepare_for_llama_export(model_name, args)
+    manager: LLMEdgeManager = _prepare_for_llama_export(args)
 
     if len(quantizers) != 0:
         manager = manager.export().pt2e_quantize(quantizers)