Hugging Face's logo

Spaces:
shamiur
/
sunnyware
Running on Zero

App Files Community
1
sunnyware / src

Commit History

2 days ago

  1. Part 653: Filter pseudo-providers (none/unknown) from health dashboard

    shamiur
    88bf8d0

  2. Part 652: Provider Health Dashboard with 1h/24h stats, P95 latency, error tracking

    shamiur
    3685860

  3. Part 651: Exponential backoff + smart per-provider cooldowns

    shamiur
    3fd1057

3 days ago

  1. Part 650: Make local model log labels dynamic

    shamiur
    43f0321

  2. Part 648: Revert to Qwen2.5-0.5B — SmolLM-135M hallucinates

    shamiur
    b7cd8ae

  3. Part 647: Hardcode SmolLM path in _model_path() to bypass env vars

    shamiur
    5596c7b

  4. Part 646: Force SmolLM path in code to bypass HF secret limits

    shamiur
    9998949

  5. Part 645: Fix SmolLM 404 by using second-state community repo

    shamiur
    adcfbf8

  6. Part 645: Add SmolLM-135M download and preload logic

    shamiur
    c76112c

  7. Part 643: Beautiful formatting for time and search tools

    shamiur
    1d93cd2

  8. Part 642: Hardcode verbose=False to stop C++ log spam

    shamiur
    70751fd

  9. Part 641: Update SambaNova model to Meta-Llama-3.3-70B-Instruct

    shamiur
    6e02718

  10. Part 640: Switch SambaNova to 8B model for better free-tier reliability

    shamiur
    bf62a1b

  11. Part 639: Update Cerebras and SambaNova model IDs to current versions

    shamiur
    9c82c10

  12. Part 638: Add Cerebras, SambaNova, and Together AI as free fast providers

    shamiur
    6545bf5

  13. Part 635: Switch HF provider to Llama-3.3-70B-Instruct for reliable routing

    shamiur
    502726c

  14. Part 634: Add Hugging Face Inference Providers as a fast router

    shamiur
    16d4b0c

  15. Part 632: Hardcode n_ctx=512, restore n_batch=512 for local model speed

    shamiur
    11e6287

  16. Part 630: Optimize local model (n_ctx=512, n_batch=64) + add Test Connection button

    shamiur
    9b72451

  17. Part 629: Bump local timeout to 200s, reduce max_tokens to 32 for speed

    shamiur
    e04d79a

  18. Part 628: Bypass create_chat_completion (hangs) with manual ChatML + create_completion

    shamiur
    4583269

  19. Part 627: Revert local model config (repeat_penalty=1.0, restore stop tokens) to fix hang

    shamiur
    386e9ed

  20. Part 627: Diagnostic logging for local model hang

    shamiur
    bdca8d0

  21. Part 626: Bump local timeout to 120s, restore token limits for complete responses

    shamiur
    8e73e9f

  22. Part 625: Remove warmup, force n_threads=2, reduce n_batch and max_tokens

    shamiur
    a628934

  23. Part 624: Separate reasoning/response tokens, warmup, faster retry

    shamiur
    fe5989c

  24. Part 623: Local model performance - dynamic n_threads, max_tokens=256, stop tokens

    shamiur
    dcc633b

  25. Part 622: Fix math matcher for unicode operators (× ÷ −) and add Jarvis fast-path

    shamiur
    0e18c24

  26. Part 621: Deterministic tools (calculator, current_time) return immediately without LLM formatting

    shamiur
    ba9ca71

  27. Part 620: Remove Needle/MiniCPM, add native lock to prevent GGML_ASSERT crash

    shamiur
    16e5fb4

  28. Part 619: Fix llama.cpp crash by disabling tools for local model

    shamiur
    d7221df

  29. Part 618: Measure latency for both local and HTTP providers in the routing loop

    shamiur
    ebf7082

  30. Part 618: Fix rate_limit _load() spam and implement Round Robin / Fastest Response routing

    shamiur
    b60f748

4 days ago

  1. Part 617: Fix provider toggle by adding 'enabled' field to Provider class

    shamiur
    edb17dd

  2. Part 616: Ensure local provider is always shown in dashboard (ignore disable flag)

    shamiur
    8aacb95

  3. Part 615: Enable local model preload by removing 'if False'

    shamiur
    1296d95

  4. Part 615: Preload local model in lifespan

    shamiur
    d9da9f7

  5. Part 614: Fix local model status loading and badge display

    shamiur
    f7dbc03

  6. Part 613: Clickable local model toggle in Chat UI

    shamiur
    4ea9d77

  7. Part 610: Fix model resolver and legacy path for local provider

    shamiur
    ac6e68c

  8. Part 609: Local model toggle and provider cache invalidation

    shamiur
    ec15ff5

  9. Part 608: Local model status endpoint and UI badge

    shamiur
    cd4f2fb

  10. Part 607: Persist embedder + Qwen2.5 models on /data for HF Spaces

    shamiur
    602fee4

  11. Part 606: Enable local Qwen2.5 + embedder on HF Gradio Space

    shamiur
    0bf1fb7

  12. Part 605: Add local Ollama fallback (Qwen2.5-0.5B)

    shamiur
    d3faa71

  13. Part 604: Embedder status endpoint and Chat UI badge

    shamiur
    882cdf8

  14. Part 603: Slack bot integration and Library tabs

    shamiur
    1c8cea0

  15. Part 602: Reading indicator and forced RAG search

    shamiur
    2c40d10

  16. Part 601: Add scheduler env var and prompt hints

    shamiur
    341f355

  17. Part 601: RAG UI integration and fixes

    shamiur
    1e51229