Not able to push the whole model to hugging face after training : help urgent , Hackathon submission EOD

[Data-Scientist-Sahil]

SO after training the model and pushing the adaptors to HF

model.push_to_hub("your_name/lora_model", token = "...") # Online saving
tokenizer.push_to_hub("your_name/lora_model", token = "...") # Online saving

this code flawlessly , but after trying to push the model in shards to HF , there was an error about model card
code -

if False: model.save_pretrained_merged("model", tokenizer, save_method = "merged_16bit",)
if False: model.push_to_hub_merged("hf/model", tokenizer, save_method = "merged_16bit", token = "")

error -

`TypeError                                 Traceback (most recent call last)
<ipython-input-14-848fa230afad> in <cell line: 2>()
      1 if True: model.save_pretrained_merged("model", tokenizer, save_method = "merged_16bit",)
----> 2 if True: model.push_to_hub_merged("rishika0704/javelin", tokenizer, save_method = "merged_16bit", token = os.getenv("hf_token"))

4 frames
/usr/local/lib/python3.10/dist-packages/transformers/utils/hub.py in create_and_tag_model_card(repo_id, tags, token, ignore_metadata_errors)
   1181     if tags is not None:
   1182         for model_tag in tags:
-> 1183             if model_tag not in model_card.data.tags:
   1184                 model_card.data.tags.append(model_tag)
   1185 

TypeError: argument of type 'NoneType' is not iterable`

Guys if anyone could help , please help me asap , this is regarding a submission in Hackathon

[codelion]

@Data-Scientist-Sahil There is some issue if the repo is not already created. See my comment here - https://discord.com/channels/1179035537009545276/1179035537529643040/1271746355093831793 see if you are on latest transformers.

[Data-Scientist-Sahil]

am unable to access the comment on that discord ...
would you please share again ...

[codelion]

Just join the unsloth discord, you may get better support there.

This was my comment -

The simplest thing may be to just do it in 2 steps. Finish the original fine-tune upload your lora then delete and clear the runtime. Then load the adapter from HF with 

    from unsloth import FastLanguageModel
    model, tokenizer = FastLanguageModel.from_pretrained(
        model_name = "your_model", # YOUR MODEL YOU USED FOR TRAINING
        max_seq_length = max_seq_length,
        dtype = dtype,
        load_in_4bit = load_in_4bit,
    )

and then push gguf  with 
model.push_to_hub_gguf("your_model", tokenizer, quantization_method = "q4_k_m", token = "")

The thing to be aware is that model.push_to_hub_gguf doesn't seem to create the repo if it doesn't exist. You can just create the model from the web and then push.

[Data-Scientist-Sahil]

how do i upload the Lora adapters after i clear the runtime and disconnected the GPU on google colab , so i tried to download all the files in model folder and then reupload it ... but ig that doesnt seem to work
i will join the discord channel

[codelion]

Start a new session, just load the model from folder and push to hub.

[rishiraj]

this should fix the issue huggingface/transformers#33315