angrykirc/ThinkingCap-Qwen3.8-27B-INT8-W8A8-imatrix
1
ThinkingCap-Qwen3.8-27B — INT8 W8A8 (imatrix)
8-bit INT8 W8A8 quantization of bottlecapai/ThinkingCap-Qwen3.8-27B (55.6 GB BF16 → 30 GB).
Quantization recipe
Layout
model-0000X-of-00015.safetensors— quantized weights (400 int8 Linear weights + 400 per-channelweight_scale, everything else BF16)model_mtp.safetensors— MTP module (15 BF16 tensors) excluded from quantization; load withload_mtp=Truein vLLM or ignore if not using speculative decodingrecipe.yaml— exact llmcompressor recipe applied
Usage
transformers
from transformers import AutoModelForImageTextToText, AutoProcessor
processor = AutoProcessor.from_pretrained("angrykirc/ThinkingCap-Qwen3.8-27B-INT8-W8A8-imatrix")
model = AutoModelForImageTextToText.from_pretrained(
"angrykirc/ThinkingCap-Qwen3.8-27B-INT8-W8A8-imatrix",
dtype=torch.bfloat16, device_map="cuda:0",
)vLLM
vllm serve angrykirc/ThinkingCap-Qwen3.8-27B-INT8-W8A8-imatrixNote: for best performance with the linear-attention layers, install flash-linear-attention and causal_conv1d (transformers falls back to reference kernels without them).
Verification
- all 400 quantized Linear weights confirmed
I8in checkpoint, 400 per-channel scales - ignored modules confirmed BF16 (vision tower, inproja/b, lm_head, mtp)
- smoke test: 104-token greedy generation runs correctly
