How to use from
llama.cpp
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Grawnch/TokenWing-Model-Assets
# Run inference directly in the terminal:
llama cli -hf Grawnch/TokenWing-Model-Assets
Install from WinGet (Windows)
winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Grawnch/TokenWing-Model-Assets
# Run inference directly in the terminal:
llama cli -hf Grawnch/TokenWing-Model-Assets
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf Grawnch/TokenWing-Model-Assets
# Run inference directly in the terminal:
./build/bin/llama cli -hf Grawnch/TokenWing-Model-Assets
Use Docker
docker model run hf.co/Grawnch/TokenWing-Model-Assets
Quick Links

TokenWing Model Assets (Ling 3.0 Flash, Flint 40B, & Qwen3.8-Flash-Next WV-v5)

Pre-computed .correction, .vram-experts.json, .experts.layer-order.json, and .usage-seed.json companion files for:

  • runmyai/Ling-3.0-Flash-TokenWing (Ling-3.0-flash-MXFP4_MOE.*)
  • runmyai/Flint-40B-TokenWing (flint-UD-Q4_K_XL.*)
  • Grawnch/Qwen3.8-Flash-Next-WV-v5-3.5bpw-TokenWing (Qwen3.8-Flash-Next-WV-v5-3.5bpw.*)
  • Grawnch/Qwen3.8-Flash-Next-WV-v5-Sub2.5bpw-TokenWing (Qwen3.8-Flash-Next-WV-v5-Sub2.5bpw.*)
  • Grawnch/Qwen3.8-Flash-Next-WV-v5-2.26bpw-TokenWing (Qwen3.8-Flash-Next-WV-v5-2.26bpw.*)

TokenWing Quant-Plugin Compatibility (--quant-plugin)

Checkpoint Plugin Release (grawnchtoader/tokenwing-assets) Plugin ABI EXT0..EXT5 Type IDs Target TokenWing Branch
Qwen3.8-Flash-Next-WV-v5-Sub2.5bpw-TokenWing wv-plugin-v1.3 ABI v2 112..117 + tokenwing.quant_plugin.types runmyai/tokenwing dev (>= 6a3e018f)
Qwen3.8-Flash-Next-WV-v5-3.5bpw-TokenWing wv-plugin-v1.3 ABI v2 112..117 + tokenwing.quant_plugin.types runmyai/tokenwing dev (>= 6a3e018f)
Qwen3.8-Flash-Next-WV-v5-2.26bpw-TokenWing wv-plugin-v1.1 ABI v1 43..48 Legacy ABI v1 builds

Files & SHA-256 Checksums

File Model Size (Bytes) SHA-256
Ling-3.0-flash-MXFP4_MOE.correction Ling 3.0 Flash 372,310,016 11bb5bf0064a63b7357152eadb76717c89c1cfb9235311676dfae0d2daae06ed
Ling-3.0-flash-MXFP4_MOE.vram-experts.json Ling 3.0 Flash 408,021 68a510e131f6e04b188a0d198d8c2e9d9d4522015fb3c153d38abb1753d2e0b2
Ling-3.0-flash-MXFP4_MOE.experts.layer-order.json Ling 3.0 Flash 180 65d65266e1ea4c0c799125da383ef9370b075b61ee77c406ced5fbf6fa0598ab
flint-UD-Q4_K_XL.correction Flint 40B 190,906,368 c1c955bdd7c2f0e1533559e5639a05ebad60afbb6c4e86ce9829091f3a801772
flint-UD-Q4_K_XL.vram-experts.json Flint 40B 477,237 8e3ca18ccb936fc61db53cb2a543ee957e1704e0e4464f9b6f8d198d509ec39b
flint-UD-Q4_K_XL.experts.layer-order.json Flint 40B 195 560b160f5316b8faea89dac810739216ca9a4feef9f8dd2d0cb759fed56aaf1e
Qwen3.8-Flash-Next-WV-v5-3.5bpw.correction Qwen3.8-Flash-Next WV-v5 3.5bpw 482,476,032 40f7b3c83bc6c9b8b12dc24bf74aa905f97ffa50052ef780d2559b1c8c94d102
Qwen3.8-Flash-Next-WV-v5-3.5bpw.vram-experts.json Qwen3.8-Flash-Next WV-v5 3.5bpw (H=96) 368,442 bf2c4ca9f7e2aa0689d966490264bc6ee31b2bb4ce4a34fe34b15f03f71c1bb4
Qwen3.8-Flash-Next-WV-v5-3.5bpw.experts.layer-order.json Qwen3.8-Flash-Next WV-v5 3.5bpw (H=96) 528 22fcc9221f39df6e62549ef01db036920c91d1e2092399c76297b97483fa72e1
Qwen3.8-Flash-Next-WV-v5-3.5bpw.usage-seed.json Qwen3.8-Flash-Next WV-v5 3.5bpw (H=96) 305,786 b9abe45c456c1ae57940ea1f5883322927f5bd8bf7cfa9a69489facfd3a3adc0
Qwen3.8-Flash-Next-WV-v5-Sub2.5bpw.vram-experts.json Qwen3.8-Flash-Next WV-v5 Sub2.5bpw (H=96) 368,442 bf2c4ca9f7e2aa0689d966490264bc6ee31b2bb4ce4a34fe34b15f03f71c1bb4
Qwen3.8-Flash-Next-WV-v5-Sub2.5bpw.experts.layer-order.json Qwen3.8-Flash-Next WV-v5 Sub2.5bpw (H=96) 528 22fcc9221f39df6e62549ef01db036920c91d1e2092399c76297b97483fa72e1
Qwen3.8-Flash-Next-WV-v5-Sub2.5bpw.usage-seed.json Qwen3.8-Flash-Next WV-v5 Sub2.5bpw (H=96) 305,786 b9abe45c456c1ae57940ea1f5883322927f5bd8bf7cfa9a69489facfd3a3adc0
Qwen3.8-Flash-Next-WV-v5-2.26bpw.vram-experts.json Qwen3.8-Flash-Next WV-v5 2.26bpw (H=96) 368,442 bf2c4ca9f7e2aa0689d966490264bc6ee31b2bb4ce4a34fe34b15f03f71c1bb4
Qwen3.8-Flash-Next-WV-v5-2.26bpw.experts.layer-order.json Qwen3.8-Flash-Next WV-v5 2.26bpw (H=96) 528 22fcc9221f39df6e62549ef01db036920c91d1e2092399c76297b97483fa72e1
Qwen3.8-Flash-Next-WV-v5-2.26bpw.usage-seed.json Qwen3.8-Flash-Next WV-v5 2.26bpw (H=96) 305,786 b9abe45c456c1ae57940ea1f5883322927f5bd8bf7cfa9a69489facfd3a3adc0
Qwen3.8-Flash-Next-WV-v5-MTP-Draft.gguf Qwen3.8-Flash-Next Native-GGUF blk.48 MTP Sidecar (draft-mtp) 2,561,310,720 ac0744c1696fc849a5dfed01efc10cfc12c374c69dce4cfbaf5159c5fcd99965
Downloads last month
211
GGUF
Model size
4B params
Architecture
qwen4exp
Hardware compatibility
Log In to add your hardware

We're not able to determine the quantization variants.

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support