Cannot build monsterapi/gemma-2b-lora-maths-orca-200k using LLM Conversion script

[hassanabidpk]

I am trying to convert monsterapi/gemma-2b-lora-maths-orca-200k to TFLite Flatbuffers file as mentioned here

But getting errors. I am reusing this script

Added below functions

def gemma2b_lora_download(token):
  REPO_ID = "monsterapi/gemma-2b-lora-maths-orca-200k"
  FILENAMES = ["tokenizer.json", "tokenizer_config.json", "adapter_model.safetensors"]
  os.environ['HF_TOKEN'] = "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"
  with out:
    for filename in FILENAMES:
      hf_hub_download(repo_id=REPO_ID, filename=filename, local_dir="./gemma-2b-lora-it")

def gemma2b_lora_convert_config(backend):
  input_ckpt = '/content/gemma-2b-lora-it/'
  vocab_model_file = '/content/gemma-2b-lora-it/'
  output_dir = '/content/intermediate/gemma-2b-lora-it/'
  output_tflite_file = f'/content/converted_models/gemma_base_lora_{backend}.bin'
  lora_output_tflite_file = f'/content/converted_models/gemma_lora_{backend}.bin'
  return converter.ConversionConfig(input_ckpt=input_ckpt, ckpt_format='safetensors', model_type='GEMMA_2B', backend=backend, output_dir=output_dir, combine_file_only=False, vocab_model_file=vocab_model_file, output_tflite_file=output_tflite_file, lora_output_tflite_file=lora_output_tflite_file)

Every time Colab enteprise is crashing

Is there any example to convert this trained model to .bin format for running on Android?

[kalyan2789g]

Hi @hassanabidpk,

We are looking into the issue. Could you please confirm the gemma model name/configuration which you are working? Please share the complete error logs to analyse the issue in detail.

We will keep you posted on the progress.

Thanks.

[hassanabidpk]

I am using this Gemma2B trained model : https://huggingface.co/monsterapi/gemma-2b-lora-maths-orca-200k based on google/gemma-2b

I am attaching
app.log
llm_conversion_notebook.txt
notebook and app.log file

[kuaashish]

Hi @hassanabidpk,

It appears that the model you provided has been modified and is not the same as the one available at https://huggingface.co/google/gemma-2b. As a result, it may not be supported, and the script may not work in this case. We will assign this to our team to investigate further. Any assistance we can get would be helpful not only for us but also for other community members trying to convert unsupported Hugging Face models.

Thank you!!