Hugging Face's logo

EschaLabs
/
escha-runtime-qwen3dense

sglang
quantization
cuda
inference
escha
qwen3
dense
Model card Files
xet
 Community
4
escha-runtime-qwen3dense

Commit History

30 days ago

  1. README: fix stale torch==2.9.* hard-pin claim, now that 1.2.2 is stable-ABI

    yzhwang
    Verified signature
    9f095a1

  1. 1.2.2: ship the same _C.so as the 35B runtime 1.0.3 (escha-kernels 9b72598; dense unaffected)

    yzhwang
    Verified signature
    a48aab5

  1. 1.2.1: MTP speculative decoding⁠

    yzhwang
    Verified signature
    e926dbf

  1. docs(cookbook): measured 16 GB recipe + The 16 GB tier, fp8 KV cache, 3 troubleshooting entries

    yzhwang
    Verified signature
    c095d63

  2. docs: 16 GB tier measured, developer-role 400, sustained-load stability note

    yzhwang
    Verified signature
    4da06dd

  1. Clarify that 64k context and 8-9 streams are not simultaneous

    yzhwang
    Verified signature
    36bb7bf

  2. Correct RADIX guidance (prefix cache is exact-match only); add long-context sizing

    yzhwang
    Verified signature
    de0e2fa

  3. escha 1.2.0: docs + changelog for tensor parallelism

    yzhwang
    Verified signature
    e05fa1a

  4. Remove superseded 1.1.1 wheel (replaced by 1.2.0)

    yzhwang
    Verified signature
    a48ecdc

  5. escha 1.2.0 wheel: tensor parallelism (gated on world_size>1)

    yzhwang
    Verified signature
    85c396e

  6. escha 1.1.1: docs + changelog for the cuda:0 device fix

    yzhwang
    Verified signature
    ca22b2f

  7. escha 1.1.1 wheel: rank device instead of hardcoded cuda:0

    yzhwang
    Verified signature
    4c39bd1

  8. Remove superseded 1.1.0 wheel (replaced by 1.1.1)

    yzhwang
    Verified signature
    9476e42

  1. Escha Runtime qwen3dense — escha 1.1.0+qwen3dense SGLang wheel, serve.sh, cookbook

    yzhwang
    bc55caa