Commit History

  1. Patch remote config typo (ministral3 -> mistral3) before loading model

    Sahil Seemant
    b40b372
  2. Pin transformers to 4.50.0 for mistral3 support

    Sahil Seemant
    d4a8dce
  3. Bypass TokenizersBackend bug by using Nemo tokenizer for Ministral

    Sahil Seemant
    9ae233f
  4. Pin transformers to 4.48.2 to avoid broken remote versions

    Sahil Seemant
    1fd607f
  5. Add offload_folder to support low-memory loading on Hugging Face

    Sahil Seemant
    e10f558
  6. Fix Streamlit set_page_config order and print specific ImportError

    Sahil Seemant
    68a7bfa
  7. Final cloud stability fixes for PEFT adapter loading

    Sahil Seemant
    7e69116
  8. Fix NoneType error by conditionally passing quantization_config

    Sahil Seemant
    0f06dcb
  9. Fix FP8 quantization conflict by skipping manual config for Mistral-3

    Sahil Seemant
    6e608c4
  10. Fix TypeError by using explicit BitsAndBytesConfig for quantization

    Sahil Seemant
    309cfde

  1. Fix TokenizersBackend error by using slow tokenizer and adding mistral-common

    Sahil Seemant
    402e3e2
  2. Fix OSError by switching to AutoTokenizer for cloud loading

    Sahil Seemant
    2634f8f
  3. Optimize cloud build: use Python 3.11 and remove vllm

    Sahil Seemant
    bedb746
  4. Fix VLM architecture incompatibility (Ministral/Qwen)

    Sahil Seemant
    cb2afbd
  5. Corrected Hugging Face repository IDs and case

    Sahil Seemant
    6b16d45
  6. Fix Streamlit popover attribute error by pinning version 1.35.0

    Sahil Seemant
    739bd96
  7. Fix Hugging Face Space configuration

    Sahil Seemant
    be490c4
  8. Add Cloud compatibility & Transformers fallback

    Sahil Seemant
    74320ca
  9. Fix cloud deployment crash and add transformers fallback

    Sahil Seemant
    ed87d92
  10. Simplified project to local MLX only

    Sahil Seemant
    06e7bdc