Hugging Face's logo

jabbatheduck
/
ninfer-ext-models

Image-Text-to-Text
Trellis
NInfer
English
Chinese
code
ninfer-ext
exl3
nvfp4
Mixture of Experts
expert-offload
blackwell
quantization
cuda
text-generation
vision
speculative-decoding
gemma4
qat
Model card Files
xet
 Community
ninfer-ext-models

Commit History

2 days ago

  1. Refresh measured numbers: repo-protocol perplexity and the same-session MTP sweep

    jabbatheduck
    Verified signature
    916f6f4

  2. Upload gemma4_26b_a4b_qat.ninfer (text + vision + MTP, QAT int4 experts)

    jabbatheduck
    Verified signature
    ceacbad

  3. Index the Gemma 4 entries

    jabbatheduck
    Verified signature
    8298762

  4. Add Gemma 4 26B-A4B IT QAT card, checksums and conversion report

    jabbatheduck
    Verified signature
    b909beb

5 days ago

  1. Delete .spd2.bin with huggingface_hub

    jabbatheduck
    Verified signature
    5824a8c

  2. Upload .spd2.bin with huggingface_hub

    jabbatheduck
    Verified signature
    632f359

  3. Delete .speedtest.bin with huggingface_hub

    jabbatheduck
    Verified signature
    6c1d4d7

  4. Upload .speedtest.bin with huggingface_hub

    jabbatheduck
    Verified signature
    c2c6bdd

  5. Delete .hf_xtest2.txt with huggingface_hub

    jabbatheduck
    Verified signature
    8a44bd3

  6. Delete .hf_xtest.txt with huggingface_hub

    jabbatheduck
    Verified signature
    cfed9af

  7. Upload .hf_xtest2.txt with huggingface_hub

    jabbatheduck
    Verified signature
    ab4841e

  8. Upload .hf_xtest.txt with huggingface_hub

    jabbatheduck
    Verified signature
    9960b6e

  9. Upload gemma4-31b-it-m1/gemma4_31b_it_m1.ninfer.conversion.json with huggingface_hub

    jabbatheduck
    Verified signature
    665864d

  10. Upload gemma4-31b-it-m1/README.md with huggingface_hub

    jabbatheduck
    Verified signature
    df35d03

  11. Upload gemma4-31b-it-m1/SHA256SUMS with huggingface_hub

    jabbatheduck
    Verified signature
    6211a32

8 days ago

  1. Upload README.md with huggingface_hub

    jabbatheduck
    Verified signature
    6c5147c

  2. Upload folder using huggingface_hub

    jabbatheduck
    Verified signature
    bc2b429

10 days ago

  1. Card: add Qwen3.8-Flash-Next EXL3 4.0 and 3.5 bpw

    jabbatheduck
    Verified signature
    b20a34d

  2. Add qwen3.8-flash-next-exl3-4bpw (EXL3 routed experts, Text + MTP)

    jabbatheduck
    Verified signature
    2097b81

  3. Add qwen3.8-flash-next-exl3-3p5bpw (EXL3 routed experts, Text + MTP)

    jabbatheduck
    Verified signature
    9533c2d

13 days ago

  1. Qwen3.8-Flash-Next: label prefill with ngram residency; note mapped is ~1.7x faster

    jabbatheduck
    Verified signature
    737e67e

  2. Qwen3.8-Flash-Next: label prefill with ngram residency; note mapped is ~1.7x faster

    jabbatheduck
    Verified signature
    e51e356

  3. Qwen3.8-Flash-Next: one prefill row, measured on the serve route

    jabbatheduck
    Verified signature
    396817e

  4. Qwen3.8-Flash-Next: one prefill row, measured on the serve route

    jabbatheduck
    Verified signature
    3729ae5

  5. Qwen3.8-Flash-Next: add long-context prefill after QSA block-key pooling

    jabbatheduck
    Verified signature
    69ce943

  6. Qwen3.8-Flash-Next: add long-context prefill after QSA block-key pooling

    jabbatheduck
    Verified signature
    ae43b15

14 days ago

  1. docs: mark Qwen3.8-Flash-Next files available

    jabbatheduck
    Verified signature
    95404da

  2. Add Qwen3.8-Flash-Next NVFP4 (ModelOpt, 119 GB, 4 shards)

    jabbatheduck
    Verified signature
    f72128d

  3. Upload README.md with huggingface_hub

    jabbatheduck
    Verified signature
    bdade29

  4. Upload README.md with huggingface_hub

    jabbatheduck
    Verified signature
    d2a172d

16 days ago

  1. Performance: rebuilt artifacts, draft head, MTP at parity with Q4

    jabbatheduck
    Verified signature
    b7197a5

  2. Rebuild with the indexed draft head and the fork chat template

    jabbatheduck
    Verified signature
    eec3230

  3. Rebuild with the indexed draft head and the fork chat template

    jabbatheduck
    Verified signature
    864390f

  4. Rebuild with the indexed draft head and the fork chat template

    jabbatheduck
    Verified signature
    322a0bc

  5. Rebuild with the indexed draft head and the fork chat template

    jabbatheduck
    Verified signature
    5cfa2cd

  6. Fill in the MTP row for Q4/NVFP4 and record the missing draft head

    jabbatheduck
    Verified signature
    3c98f5d

  7. Add the 3.5 bpw column: slower on both regimes, size tier only

    jabbatheduck
    Verified signature
    2cd6192

  8. Refresh performance: prefill 1.17x of Q4, decode at parity, MTP K=3 124 tok/s

    jabbatheduck
    Verified signature
    cf80ef8

17 days ago

  1. Add EXL3 3.5 bpw; record KLD and the prefill gap

    jabbatheduck
    Verified signature
    ed67413

  2. Add Qwen3.8-27B EXL3 3.5 bpw

    jabbatheduck
    Verified signature
    6302c0f

  3. Upload qwen3_8_27b_exl3_3p5bpw.ninfer.conversion.json with huggingface_hub

    jabbatheduck
    Verified signature
    2e67450

  4. Add Qwen3.8-27B EXL3 4.0 bpw

    jabbatheduck
    Verified signature
    e1efd4e

  5. Upload qwen3_8_27b_exl3_4bpw.ninfer.conversion.json with huggingface_hub

    jabbatheduck
    Verified signature
    0082f57

  6. Upload README.md with huggingface_hub

    jabbatheduck
    Verified signature
    2f7c1bc

  7. initial commit

    jabbatheduck
    Verified signature
    3452ff2