save captioning in history.jsonl file with settings and selected images

[geroldmeisinger]

171

.jsonl expect a json object on every line. here is an (formatted) example. actual file is compact. the name of the file is !0_history.jsonl to always display on the beginning in most file managers:

{
    "date": "2024-06-09 18:19:46",
    "settings": {
        "model": "THUDM/cogvlm2-llama3-chat-19B-int4",
        "prompt": "describe this image",
        "caption_start": "",
        "caption_position": "Insert after last tag",
        "device": "GPU if available",
        "gpu_index": 0,
        "load_in_4_bit": true,
        "remove_tag_separators": false,
        "bad_words": "",
        "forced_words": "",
        "generation_parameters": {
            "min_new_tokens": 1,
            "max_new_tokens": 512,
            "num_beams": 1,
            "length_penalty": 1.0,
            "do_sample": false,
            "temperature": 1.0,
            "top_k": 50,
            "top_p": 1.0,
            "repetition_penalty": 1.0,
            "no_repeat_ngram_size": 3
        },
        "wd_tagger_settings": {
            "show_probabilities": true,
            "min_probability": 0.4,
            "max_tags": 30,
            "tags_to_exclude": ""
        }
    },
    "images": [
        "000000020.jpg",
        "000000020.webp"
    ]
}

[geroldmeisinger]

todo:

• on click of history entry restore settings
• add model revision to infos
• add history to "Views" and make it show again when closed

[geroldmeisinger]

new scheme:

{
    "date": "2024-06-10 09:52:04",
    "history_version": 0,
    "app": {
        "origin": "git@github.com:geroldmeisinger/taggui.git",
        "revision": "12a341e823c7bda229713e366491d3ce90fec1c2",
        "settings": {
            "model": "THUDM/cogvlm2-llama3-chat-19B-int4",
            "prompt": "describe this image",
            "caption_start": "",
            "caption_position": "Insert before first tag",
            "load_in_4_bit": true,
            "remove_tag_separators": true,
            "bad_words": "",
            "forced_words": "",
            "generation_parameters": {
                "min_new_tokens": 1,
                "max_new_tokens": 100,
                "num_beams": 1,
                "length_penalty": 1.0,
                "do_sample": false
            }
        }
    },
    "model": {
        "name": "THUDM/cogvlm2-llama3-chat-19B-int4",
        "type": ""
    },
    "images": [
        "000000020.webp"
    ]
}

model = model information according to model loader (transformers in this case) i want to add "commit hash", "endpoint" (=huggingface.co), "architecture" or "internal model name", quantization app = model information according to settings in application (taggui)

[geroldmeisinger]

@jhc13 : in history_list.py:45 HistoryListModel.append how do i get the models...

• git commit hash of download
• pretrained_model_name_or_path or model_name_or_path
• internal type/architecture as string
• repo url endpoint used in downloader (huggingface.co or any other url set)

transformers defines sooo many classes I'm confused about which type is actually used and where, and if the info is available in the model, the config or the hub.

https://huggingface.co/docs/transformers/main/en/model_doc/auto#transformers.AutoConfig.from_pretrained https://huggingface.co/docs/transformers/main/en/model_doc/auto#transformers.AutoModelForCausalLM https://huggingface.co/docs/transformers/main/en/model_doc/auto#transformers.AutoModel https://huggingface.co/docs/transformers/main/en/main_classes/model#transformers.PreTrainedModel https://huggingface.co/docs/transformers/main/en/main_classes/configuration#transformers.PretrainedConfig

[geroldmeisinger]

History View: