Hugging Face's logo

Spaces:
hardkpentium101
/
indicRAG
Sleeping

App Files Community
indicRAG / backend /src

Commit History

  1. downgrade datasets and add trust rc kwarg

    hardkpentium101
    806abbf

  1. update embedding gen to use vector storage files

    hardkpentium101
    c31bec2

  2. sanitize token in embeddings generator

    hardkpentium101
    e226194

  3. remove write permission args from hf login

    hardkpentium101
    cff0c00

  4. fix llm manager

    hardkpentium101
    504adfa

  5. fix llm manager

    hardkpentium101
    c95fe64

  6. sanitize hf token

    hardkpentium101
    0ce52c8

  7. sanitize hf token

    hardkpentium101
    5d092bc

  1. Switch to Qwen-1.5-1.8B-Chat - verified multilingual model with good Indic support⁠

    hardkpentium101 +1: Qwen-Coder
    3862877

  2. Pass HF_TOKEN explicitly to model loading - more reliable in Docker⁠

    hardkpentium101 +1: Qwen-Coder
    aefd0f7

  3. Switch to AI4Bharat IndicLLM - better support for 11 Indic languages⁠

    hardkpentium101 +1: Qwen-Coder
    057cc64

  4. update max new token to 1024

    hardkpentium101
    a9b1188

  1. Simplify validation - use general patterns, not hardcoded lists⁠

    hardkpentium101 +1: Qwen-Coder
    dd9966e

  2. Set generation_config on model only, not passed to pipeline - fixes duplicate arg error⁠

    hardkpentium101 +1: Qwen-Coder
    53643df

  3. Suppress max_new_tokens/max_length warning - cosmetic only⁠

    hardkpentium101 +1: Qwen-Coder
    e9cffd1

  4. Create gen_config once, pass to pipeline once - clean implementation⁠

    hardkpentium101 +1: Qwen-Coder
    288b50e

  5. Use GenerationConfig object passed to pipeline - proper format⁠

    hardkpentium101 +1: Qwen-Coder
    bff2384

  6. Pass generation params directly to pipeline - no GenerationConfig object⁠

    hardkpentium101 +1: Qwen-Coder
    2bbaeb3

  7. Remove redundant config override - set generation_config once⁠

    hardkpentium101 +1: Qwen-Coder
    b0da3b5

  8. Force override model.config max_length and max_new_tokens to fix warning⁠

    hardkpentium101 +1: Qwen-Coder
    52b77b9

  9. Use structured prompt format (CONTEXT/OBJECTIVE/STYLE/TONE/AUDIENCE/RESPONSE)⁠

    hardkpentium101 +1: Qwen-Coder
    bea43dd

  10. Simplify prompt and validation - robust creative output⁠

    hardkpentium101 +1: Qwen-Coder
    e6027ef

  11. Fix syntax error: missing quote in invalid_patterns⁠

    hardkpentium101 +1: Qwen-Coder
    9a520d5

  12. Use comprehensive creative writer prompt with strict output validation⁠

    hardkpentium101 +1: Qwen-Coder
    1c91cf7

  13. Fix prompt leakage: simpler prompt format, stricter filtering for exact leakage patterns⁠

    hardkpentium101 +1: Qwen-Coder
    3b06a13

  14. Set max_length=None in GenerationConfig to override model defaults⁠

    hardkpentium101 +1: Qwen-Coder
    9164057

  15. Fix max_length conflict by explicitly setting max_length=None in pipeline⁠

    hardkpentium101 +1: Qwen-Coder
    355f389

  16. Add poem/story specifications: type, theme, length, style⁠

    hardkpentium101 +1: Qwen-Coder
    5737e4a

  17. Add output validation and strengthen prompt against meta-commentary⁠

    hardkpentium101 +1: Qwen-Coder
    e9c1b7a

  18. Add user-selected language to prompt for proper language response⁠

    hardkpentium101 +1: Qwen-Coder
    5c2c171

  19. Restructure prompt for direct creative output, improve meta-commentary filtering⁠

    hardkpentium101 +1: Qwen-Coder
    5d1a0cf

  20. Improve output cleaning: remove </s>, [INST], bracket numbers⁠

    hardkpentium101 +1: Qwen-Coder
    bdbc8d1

  21. Strengthen prompt: never refer to or explain context⁠

    hardkpentium101 +1: Qwen-Coder
    eae5493

  22. Add output cleaning to filter informal/garbled text⁠

    hardkpentium101 +1: Qwen-Coder
    ff4f2a3

  23. Update prompt for creative writing, increase temperature to 0.9 and top_p to 0.92⁠

    hardkpentium101 +1: Qwen-Coder
    bebbcff

  24. Set generation_config on model directly to avoid duplicate param error⁠

    hardkpentium101 +1: Qwen-Coder
    177f8b5

  25. Use GenerationConfig to avoid parameter conflict warnings⁠

    hardkpentium101 +1: Qwen-Coder
    d8b5182

  26. Fix max_length conflict, set max_new_tokens to 1024⁠

    hardkpentium101 +1: Qwen-Coder
    93d3286

  27. Increase max_tokens to 4096, send top 3 docs, truncate context to 800 chars⁠

    hardkpentium101 +1: Qwen-Coder
    cf6e686

  28. Increase max_new_tokens to 1024 for longer responses⁠

    hardkpentium101 +1: Qwen-Coder
    27f1789

  29. Update prompt for Hindi literature expertise and multilingual support⁠

    hardkpentium101 +1: Qwen-Coder
    8e3ac92

  30. Fix prompt template and increase max_new_tokens to 512⁠

    hardkpentium101 +1: Qwen-Coder
    c8b8fcf

  31. Fix CPU inference: auto-detect GPU, use float16 on CPU⁠

    hardkpentium101 +1: Qwen-Coder
    7e8fd52

  32. Use bitsandbytes 4-bit quantization instead of AirLLM (more stable)⁠

    hardkpentium101 +1: Qwen-Coder
    83eb81f

  33. Add airllm and optimum to requirements⁠

    hardkpentium101 +1: Qwen-Coder
    4e571e5

  34. Use AirLLM 4-bit quantization for Sarvam-1 (uses ~1.5GB RAM)⁠

    hardkpentium101 +1: Qwen-Coder
    c47fb58

  35. Switch to TinyLlama-1.1B with float16 for lower memory⁠

    hardkpentium101 +1: Qwen-Coder
    916bdad

  36. Tighten prompt to reduce meta-commentary⁠

    hardkpentium101 +1: Qwen-Coder
    4dd4fff

  37. Improve system prompt for better RAG responses⁠

    hardkpentium101 +1: Qwen-Coder
    a059043

  38. Add debug logging for prompt/response⁠

    hardkpentium101 +1: Qwen-Coder
    71ceb5b