Downloads · 30 days
15
18% of all-time downloads
weili-0234/Qwen3.5-27B-NVFP4-QAD-W4A16-LR1e-5-s2640
Qwen3.5-27B-NVFP4-QAD-W4A16-LR1e-5-s2640 is a machine learning model from weili-0234. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
NVFP4 weight-only (W4A16-serving) export of the NV-W4A16 QAD arm final checkpoint from the stage-3 W4A4-vs-W4A16 training study on dense Qwen/Qwen3.5-27B.
Downloads · 30 days
15
18% of all-time downloads
All-time downloads
83
Public
Parameters
16.7B
19.7 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors19.7 GB · 100%
How the weights are stored.
U812.2B · 73%
From the Hugging Face model README
NVFP4 weight-only (W4A16-serving) export of the NV-W4A16 QAD arm final checkpoint from the stage-3 W4A4-vs-W4A16 training study on dense Qwen/Qwen3.5-27B.
8e7cf1a + the _save_checkpoint empty_cache+barrier fix +
the 2026-07-24 context patches (rank-local RNG state save/load;
torch.cuda.set_device(LOCAL_RANK) at entry).configs/llm/qwen3_5_27b_nvfp4_qad.yaml (W4A16 fake-quant
training), lr 1.0e-5 cosine (1% warmup), 2640 steps (~1 epoch),
global batch 32 (bs 4/GPU x 8 GPUs x accum 1), seq len 8192, seed 42,
BF16, FSDP2, save/250, eval/100.0406bb3a7a482352360716a1bc5e9e04; disjoint eval split).Final held-out eval (epoch 1.0, step 2640): fwd-KL 0.005815 vs the BF16 teacher (top-1 agreement 0.976). The matched NV-W4A4-trained twin plateaued at fwd-KL ~0.0135 from s1700 onward — for NVFP4 serving, W4A16 training reached 2.3x lower KL than W4A4 training at identical steps, lr, data, and seed.
python scripts/export_nvfp4_vllm.py \
--source /scratch/wxu/mxfp4/outputs/27b-nvfp4-w4a16-lr1.0e-5-2640 \
--model-assets /scratch/wxu/mxfp4/models/Qwen3.5-27B \
--output /scratch/wxu/mxfp4/exports/27b-nvw4a16-s2640-weightonly \
--device cuda:0 --weight-only
Load-smoked with vLLM 0.25.1 (vllm serve <dir> --max-model-len 4096):
loads clean and completes text correctly. First engine start pays a
FlashInfer fp4_gemm autotune (~10-25 min); subsequent starts reuse the cache.
7-benchmark protocol results (RULER-32k, AIME'25 avg@4, GPQA-Diamond, MMLU-Pro, HumanEval, IFEval, perplexity) will be added to this card when the matrix wave for arm finals completes.