Hugging Face's logo

mlboydaisuke
/
Qwen3.8-27B-CoreAI

Image-Text-to-Text
coreai
qwen3_5
aimodel
apple-silicon
on-device
qwen3.8
hybrid
gated-deltanet
vlm
speculative-decoding
Model card Files
xet
 Community
Qwen3.8-27B-CoreAI / gpu-pipelined

Commit History

  1. qwen3_8_27b_verify_s9_int4lin_d4_hpost: int4 S=9 verify + hidden out for the MTP drafter (29.0 tok/s code, lossless)

    mlboydaisuke
    Verified signature
    aaccef9

  2. qwen3_8_27b_decode_int4lin: int4 text decoder (22.2 tok/s M4 Max, gate 15/16)

    mlboydaisuke
    Verified signature
    53ed45d

  3. qwen3_8_27b_mtp_s9_int8hu_block32_sym: MTP S=9 replay graph (shared-KV sibling)

    mlboydaisuke
    Verified signature
    4d6f7de

  4. qwen3_8_27b_mtp_s1_int8hu_block32_sym: MTP head as S=1 stateful drafter (int8)

    mlboydaisuke
    Verified signature
    577a703

  1. Qwen3.8-27B: remove pf32 VL decoder (fp16 overflow on real images; pf16 replaces it)

    mlboydaisuke
    Verified signature
    927bbee

  2. Qwen3.8-27B: VL decoder pf16 (pf32 chunk overflows fp16 content-dependently)

    mlboydaisuke
    Verified signature
    b2086ed

  3. Qwen3.8-27B: qwen3_8_27b_vl_decode_int8hu_block32_sym_pf32

    mlboydaisuke
    Verified signature
    df6b93a

  4. Qwen3.8-27B: qwen3_8_27b_vision_fp16

    mlboydaisuke
    Verified signature
    3863519

  5. Qwen3.8-27B: qwen3_8_27b_decode_int8hu_block32_sym

    mlboydaisuke
    Verified signature
    d18974d