OpenMol/PubChemSFT

• all_clean.json

  • remove overlapped parts with ChEBI-20 test
  • remove no description SMILES
  • Format:

    {
      SMILES <str>:
      [
        ["Please describe the molecule", DESCRIPTION],
        ...,
      ]
    }
    

  • Stats
    • max tokens length: 6113
    • min tokens length: 20
    • mean tokens length: 191
    • median tokens length: 149
    • Total 326,689 single turn dialogue. Total 293,302 SMILES examples.

• Size:
  Train: 264,391
  Valid: 33,072
  Test: 32,987

• conversation template

  'conversation':{
    [
      "from":  "human",
      "value": <QUERY>, # randomly sample from query templates,
    ],
    [
      "from":  "gpt",
      "value": <TEXT>, # description about the given molecule
    ],
  }
  

• Each entry in the pickle contains:

  "graph":
    [
      "edge_index":, # array like (int64) 
      "edge_feat":,  # array like (int64) 
      "node_feat":,  # array like (int64)
      "num_nodes":,  # int
    ],
    "conversation:: # see above
  

• 14 query templates

    # bear me, we still use <image> token instead of somewhat <molecule> token
    {
     '<image>\nCould you give me a brief overview of this molecule?',
     '<image>\nCould you provide a description of this molecule?',
     '<image>\nDescribe this molecule.',
     '<image>\nPlease give me some details about this molecule.',
     '<image>\nProvide a brief overview of this molecule.',
     '<image>\nProvide a description of this molecule.',
     '<image>\nWhat can you tell me about this molecule?',
     'Could you give me a brief overview of this molecule?\n<image>',
     'Could you provide a description of this molecule?\n<image>',
     'Describe this molecule.\n<image>',
     'Please give me some details about this molecule.\n<image>',
     'Provide a brief overview of this molecule.\n<image>',
     'Provide a description of this molecule.\n<image>',
     'What can you tell me about this molecule?\n<image>'
    }