o
    °ƒðjö3  ã                   @   s  d dl Z d dlZd dlZd dlZd dlmZ 	 ddd„Zdd„ Zdd„ Zd	d
„ Z	dd„ Z
			ddd„Zdd„ Zdd„ Zdd„ Zedkr}eƒ Zedeƒ d dlmZ eejƒ ejsfejsaJ dƒ‚eeƒZnejZd dlmZ eeeƒZeD ]Zeeƒ qvdS dS )é    N)ÚTensorProtoc                 C   s¬  t  ¡ }|jdddtdd� |jdddtdd� |jd	d
dtddd� |jdddtddd� |jddtddd� |jddtddd� |jddtddd� |jddtddd� |jddtddd� |jddtd d d� |jd!dtd d"d� |jd#dtd d$d� |jd%dd&g d'¢d(d)� |jd*d+dd,d-d.� |jdd/� |jd0dtd1d2d� |jd3dd,d4d.� |jdd5� |jd6dd,d7d.� |jdd8� |jd9d:dd,d;� |jdd<� | | ¡S )=Nz-iz--inputFz2Set the input file for reading the profile results)ÚrequiredÚtypeÚhelpz-mz--modelzIonnx model path to run profiling. Required when --input is not specified.z-bz--batch_sizeé   zbatch size of input)r   r   Údefaultr   z-sz--sequence_lengthé    zsequence length of inputz--past_sequence_lengthzpast sequence length for gpt2z--global_lengthz&number of global tokens for longformerz	--samplesiè  z\number of samples to test. Set it large enough to reduce the variance of performance result.z--thresholdg{®Gáz„?zfThreshold of run time ratio among all nodes. Nodes with larger ratio will show in top expensive nodes.z--thread_numéÿÿÿÿznumber of threads to usez--input_ids_namez"input name for input IDs, for bertz--segment_ids_namez$input name for segment IDs, for bertz--input_mask_namez'input name for attention mask, for bertz--dummy_inputsr   )ÚbertÚgpt2Ú
longformerr   zEType of model inputs. The default will create dummy inputs with ones.)r   r   Úchoicesr   z-gz	--use_gpuÚ
store_truezuse GPU)r   Úactionr   )Úuse_gpuz
--providerÚcudazExecution provider to usez--basic_optimizationz_Enable only basic graph optimizations. By default, all optimizations are enabled in OnnxRuntime)Úbasic_optimizationz--kernel_time_onlyz.Only include the kernel time and no fence time)Úkernel_time_onlyz-vz	--verbose)r   r   )Úverbose)ÚargparseÚArgumentParserÚadd_argumentÚstrÚintÚfloatÚset_defaultsÚ
parse_args)ÚargvÚparser© r   ú£/tmp/claude-1000/-var-www-html-academy/3f42efd1-c2b7-4d08-9b05-30d86bb35479/scratchpad/piper/venv/lib/python3.10/site-packages/onnxruntime/transformers/profiler.pyÚparse_arguments   sð   ûûú	ú	ûûûûûûûûûûüü
r!   c                 C   sD   ddl m} || ||| |dd�}|D ]}| d |¡}	q| ¡ }
|
S )Nr   )Úcreate_onnxruntime_sessionT)Úenable_all_optimizationÚnum_threadsÚenable_profiling)Úbenchmark_helperr"   ÚrunÚend_profiling)Úonnx_model_pathr   Úproviderr   Ú
thread_numÚ
all_inputsr"   ÚsessionÚinputsÚ_Úprofile_filer   r   r    Úrun_profile�   s   ú	r1   c                 C   s&   t |  d¡ƒtkrt| |  d¡ƒS d S )NÚvalue)r   Ú
WhichOneofr   Úgetattr)Údimr   r   r    Úget_dim_from_type_proto°   s   &r6   c                 C   s   dd„ | j jjD ƒS )Nc                 S   s   g | ]}t |ƒ‘qS r   )r6   )Ú.0Údr   r   r    Ú
<listcomp>µ   s    z-get_shape_from_type_proto.<locals>.<listcomp>)Útensor_typeÚshaper5   )Ú
type_protor   r   r    Úget_shape_from_type_proto´   s   r=   c                    s  i ‰ |   ¡ D ]m}t|jƒ}g }t|ƒD ]\}}t|tƒr!| |¡ qt|ƒdkr+ dS t|ƒdkr7|||d < t|ƒdkrC|||d < |jjj	}	|	t
jt
jt
jfv sTJ ‚|	t
jkr\tjn
|	t
jkrdtjntj}
tj||
d�}|ˆ |j< q‡ fdd„t|ƒD ƒ}|S )a  Create dummy inputs for ONNX model.

    Args:
        onnx_model (OnnxModel): ONNX model
        batch_size (int): batch size
        sequence_length (int): sequence length
        samples (int): number of samples

    Returns:
        List[Dict]: list of inputs
    é   Nr   r   ©Údtypec                    ó   g | ]}ˆ ‘qS r   r   ©r7   r/   ©Údummy_inputsr   r    r9   Þ   ó    z'create_dummy_inputs.<locals>.<listcomp>)Ú'get_graph_inputs_excluding_initializersr=   r   Ú	enumerateÚ
isinstancer   ÚappendÚlenr:   Ú	elem_typer   ÚFLOATÚINT32ÚINT64ÚnumpyÚfloat32Úint64Úint32ÚonesÚnameÚrange)Ú
onnx_modelÚ
batch_sizeÚsequence_lengthÚsamplesÚgraph_inputr;   Úsymbol_dimsÚir5   rK   Ú	data_typeÚdatar,   r   rC   r    Úcreate_dummy_inputs¸   s0   


€

ÿýr_   c                 C   sB   ddl m}m} || |||ƒ\}	}
}||||dd|	|
|dd�	}|S )a-  Create dummy inputs for BERT model.

    Args:
        onnx_model (OnnxModel): ONNX model
        batch_size (int): batch size
        sequence_length (int): sequence length
        samples (int): number of samples
        input_ids_name (str, optional): Name of graph input for input IDs. Defaults to None.
        segment_ids_name (str, optional): Name of graph input for segment IDs. Defaults to None.
        input_mask_name (str, optional): Name of graph input for attention mask. Defaults to None.

    Returns:
        List[Dict]: list of inputs
    r   )Úfind_bert_inputsÚgenerate_test_dataé{   F)Ú
test_casesÚseedr   Ú	input_idsÚsegment_idsÚ
input_maskÚrandom_mask_length)Úbert_test_datar`   ra   )rV   rW   rX   rY   Úinput_ids_nameÚsegment_ids_nameÚinput_mask_namer`   ra   re   rf   rg   r,   r   r   r    Úcreate_bert_inputsâ   s   ÷rm   c                    sæ   ||||| dœ}i ‰ |   ¡ D ]V}t|jƒ}t|ƒD ]\}}	t|	tƒr4|	|vr.td|	› �ƒ‚||	 ||< q|jjj}
|
t	j
t	jt	jfv sFJ ‚|
t	j
krNtjn
|
t	jkrVtjntj}tj||d�}|ˆ |j< q‡ fdd„t|ƒD ƒ}|S )aÂ  Create dummy inputs for GPT-2 model.

    Args:
        onnx_model (OnnxModel): ONNX model
        batch_size (int): batch size
        sequence_length (int): sequence length
        past_sequence_length (int): past sequence length
        samples (int): number of samples

    Raises:
        RuntimeError: symbolic is not supported. Use the tool convert_to_onnx.py to export ONNX model instead.

    Returns:
        List[Dict]: list of inputs
    )rW   Úseq_lenÚpast_seq_lenÚtotal_seq_lenúsymbol is not supported: r?   c                    rA   r   r   rB   rC   r   r    r9   7  rE   z&create_gpt2_inputs.<locals>.<listcomp>)rF   r=   r   rG   rH   r   ÚRuntimeErrorr:   rK   r   rL   rM   rN   rO   rP   rQ   rR   rS   rT   rU   )rV   rW   rX   Úpast_sequence_lengthrY   ÚsymbolsrZ   r;   r\   r5   rK   r]   r^   r,   r   rC   r    Úcreate_gpt2_inputs  s0   ü

€

ÿýru   c                    s  ||dœ}i ‰ |   ¡ D ]m}t|jƒ}t|ƒD ]\}}	t|	tƒr0|	|vr*td|	› �ƒ‚||	 ||< q|jjj}
|
t	j
t	jt	jfv sBJ ‚|
t	j
krJtjn
|
t	jkrRtjntj}d|jv rltj||d�}d|dd…d|…f< ntj||d�}|ˆ |j< q‡ fdd„t|ƒD ƒ}|S )	aÎ  Create dummy inputs for Longformer model.

    Args:
        onnx_model (OnnxModel): ONNX model
        batch_size (int): batch size
        sequence_length (int): sequence length
        global_length (int): number of global tokens
        samples (int): number of samples

    Raises:
        RuntimeError: symbolic is not supported. Use the tool convert_longformer_to_onnx.py to export ONNX model instead.

    Returns:
        List[Dict]: list of inputs
    )rW   rX   rq   Úglobalr?   r   Nc                    rA   r   r   rB   rC   r   r    r9   f  rE   z,create_longformer_inputs.<locals>.<listcomp>)rF   r=   r   rG   rH   r   rr   r:   rK   r   rL   rM   rN   rO   rP   rQ   rR   rT   ÚzerosrS   rU   )rV   rW   rX   Úglobal_lengthrY   rt   rZ   r;   r\   r5   rK   r]   r^   r,   r   rC   r    Úcreate_longformer_inputs;  s.   


€

ÿý
ry   c                 C   s  | j dkr| j ntjdd�}dtjvrt|ƒtjd< ddlm} ddlm	} ||| j
ƒƒ}d }| jdkrEt|| j| j| j| j| j| jƒ}n.| jdkrWt|| j| j| j| jƒ}n| jd	krit|| j| j| j| jƒ}n
t|| j| j| jƒ}t| j
| j| j| j| j |ƒ}|S )
Nr   F)ÚlogicalÚOMP_NUM_THREADS)Úload)Ú	OnnxModelr
   r   r   )r+   ÚpsutilÚ	cpu_countÚosÚenvironr   Úonnxr|   rV   r}   ÚmodelrD   rm   rW   rX   rY   rj   rk   rl   ru   rs   ry   rx   r_   r1   r   r*   r   )Úargsr$   r|   r}   rV   r,   r0   r   r   r    r'   j  sV   

ù
	û
ûú	r'   Ú__main__Ú	Arguments)Úsetup_loggerzMrequires either --model to run profiling or --input to read profiling results)Úprocess_results)N)NNN)r   r€   rO   r~   r‚   r   r!   r1   r6   r=   r_   rm   ru   ry   r'   Ú__name__Ú	argumentsÚprintr&   r‡   r   Úinputrƒ   r0   Úprofile_result_processorrˆ   ÚresultsÚliner   r   r   r    Ú<module>   sD    
	 /
ù)0/6




î