Nov 13 2 days ago

1. 20:52

   Forward project context through a dedicated Gradio input instead of filtered chat history

   SEE7
   Verified signature

   d2287b7
2. 20:50

   Hold partial Harmony markers to keep streamed text monotonic

   SEE7
   Verified signature

   7610b69
3. 20:49

   Request complete Gradio stream snapshots before converting to OpenAI deltas

   SEE7
   Verified signature

   163174c
4. 20:46

   Record completed public Space integration tests

   SEE7
   Verified signature

   cc4fe65
5. 20:46

   Record completed public Space integration tests

   SEE7
   Verified signature

   a06e5f1
6. 20:46

   Stream Thinking content as a stable prefix and strip Harmony role transitions

   SEE7
   Verified signature

   effa79e
7. 20:44

   Handle native Mxfp4Config objects during startup validation

   SEE7
   Verified signature

   434757e
8. 20:42

   Add real Instruct and Thinking endpoint smoke tests

   SEE7
   Verified signature

   8ae83bd
9. 20:40

   Pin GPT-OSS MXFP4 compatible kernel runtime

   SEE7
   Verified signature

   d6065cf
10. 20:40

    Use supported native GPT-OSS attention for model loading

    SEE7
    Verified signature

    5065e85
11. 20:39

    Pin AstheralPlus Space to published trained MXFP4 checkpoint

    SEE7
    Verified signature

    3578bc8
12. 20:37

    Publish browser integration test for Projects with mocked completion transport

    SEE7
    Verified signature

    90236d7
13. 20:37

    Align Transformers with the trained model and exported-weight verification

    SEE7
    Verified signature

    156e352
14. 20:29

    Keep complete chat URLs and reset project context for new conversations

    SEE7
    Verified signature

    eddf75a
15. 20:26

    Implement 20.9B native reasoning inference and persistent project workspaces

    SEE7
    Verified signature

    84979ea
16. 15:31

    Preserve ZeroGPU platform errors using session queue events instead of null call errors

    SEE7
    Verified signature

    8be89d0
17. 15:29

    Expose sanitized quota readiness and preserve precise budget errors before queue

    SEE7
    Verified signature

    a06dbd3
18. 15:19

    Remove public account/model headers and visitor OAuth checks; keep server quota guard with retries

    SEE7
    Verified signature

    ffd5312
19. 15:10

    Add coding workspace, text/code references, per-model profiles and Continue reply

    SEE7
    Verified signature

    4b43fe7
20. 05:17

    Allow 4096-token replies and 110-second generation with 180-second quota reserve

    SEE7
    Verified signature

    7244dd9
21. 04:57

    Pin Astheral chat to tested V3 weights; preserve UI and quota guard

    SEE7
    Verified signature

    92a2cd2
22. 03:49

    Keep selected GGUF identity in responses and reset missing temporary models

    SEE7
    Verified signature

    0786d8f
23. 03:47

    Register the GGUF ZeroGPU handler during application startup

    SEE7
    Verified signature

    9c78f5e
24. 03:46

    Add private GGUF upload, model selection and quota-guarded CUDA inference

    SEE7
    Verified signature

    b66be8b
25. 03:38

    Serve llama-ui routes directly without Gradio SSR interception

    SEE7
    Verified signature

    766c469
26. 03:36

    Preserve Hugging Face visitor headers through the Gradio bridge

    SEE7
    Verified signature

    e2cae95
27. 03:34

    Install real llama-ui with streaming Astheral ZeroGPU adapter and existing quota guard

    SEE7
    Verified signature

    66474d0
28. 03:22

    Pause GPU generation before final three minutes of included quota

    SEE7
    Verified signature

    8ff525a
29. 03:15

    Correct Gradio examples for response settings inputs

    SEE7
    Verified signature

    94f80ad
30. 03:13

    Deploy private Astheral demo with real weights on ZeroGPU

    SEE7
    Verified signature

    0339afd
31. 03:11

    initial commit

    SEE7
    Verified signature

    7312ec4