[BUG] post_process_function on rerank_pipeline_with_bge-rerank-m3-v2_model_deployed_on_Sagemaker.md

[tkykenmt]

What is the bug?
post_process_function on rerank_pipeline_with_bge-rerank-m3-v2_model_deployed_on_Sagemaker.md has a logic issue which returns inappropriate sorted result. Need to fix with updated code.

https://github.com/opensearch-project/ml-commons/blob/main/docs/tutorials/rerank/rerank_pipeline_with_bge-rerank-m3-v2_model_deployed_on_Sagemaker.md

How can one reproduce the bug?
Steps to reproduce the behavior:

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

What is the expected behavior?
A clear and concise description of what you expected to happen.

What is your host/environment?

• OS: [e.g. iOS]
• Version [e.g. 22]
• Plugins

Do you have any screenshots?
If applicable, add screenshots to help explain your problem.

Do you have any additional context?
Add any other context about the problem.

[mingshl]

Hi @tkykenmt thanks for making the fix, but more details to the PR would help the community understand the issue better,

This is how I interpret your fix if I get this right,

previously, the cross encorder model return the model output ordering the highest scores first, but it might cause the mismatch with the document order,

for example:

sample model response is ordering by the highest score first and rerank the document index num,

 [
  {index: 0, score: 0.95},
  {index: 2, score: 0.3},
  {index: 1, score: 0.2}
]

and the document was sending to the model in the order of

 [
  {index: 0},
  {index: 1},
  {index: 2}
]

and the post process function handles the model output following original document index sorting, so the scores are adding to the proper documents.

after the new pro-processing function, it will be

[
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.95]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.2]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.3]}
]

please correct me if I get this wrong

[tkykenmt]

Hi @mingshl, thank you for clarification. My intention is as follows.

sample model response is ordering by the highest score first and rerank the document index num:

 [
  {index: 0, score: 0.95},
  {index: 2, score: 0.3},
  {index: 1, score: 0.2}
]

and the document was sending to the model in the order of:

 [
  {index: 0},
  {index: 1},
  {index: 2}
]

Current post process function also aims to handles the model output following original document index sorting, but reorder logic is incorrect. The logic can generate result with accidentally overwritten another item such as:

[
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.95]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.2]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.2]},
]

After the new pro-processing function, the issue should be fixed and result will be:

[
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.95]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.2]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.3]}
]

[mingshl]

Hi @mingshl, thank you for clarification. My intention is as follows.

sample model response is ordering by the highest score first and rerank the document index num:

 [
  {index: 0, score: 0.95},
  {index: 2, score: 0.3},
  {index: 1, score: 0.2}
]

and the document was sending to the model in the order of:

 [
  {index: 0},
  {index: 1},
  {index: 2}
]

Current post process function also aims to handles the model output following original document index sorting, but reorder logic is incorrect. The logic can generate result with accidentally overwritten another item such as:

[
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.95]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.2]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.2]},
]

After the new pro-processing function, the issue should be fixed and result will be:

[
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.95]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.2]},
  {"name": "similarity", "data_type": "FLOAT32", "shape": [1], "data": [0.3]}
]

thanks for the explaination, this helps a lot to understand the issue, can you add this to the issue description as well? Thank you!