Qwen3.6-35B-A3B-INT8
INT8 weight-only quantization of Qwen/Qwen3.6-35B-A3B using llmcompressor.
Quantization details
- Scheme: INT8 (weight-only, no calibration data)
- Tool: llmcompressor
QuantizationModifier - Excluded:
lm_head, vision tower, MoE router gates, shared expert gate - Size: ~35GB (vs ~70GB BF16 original)
Usage
from transformers import AutoModelForCausalLM, AutoProcessor
model = AutoModelForCausalLM.from_pretrained("lokeshe09/Qwen3.6-35B-A3B-INT8", trust_remote_code=True)
processor = AutoProcessor.from_pretrained("lokeshe09/Qwen3.6-35B-A3B-INT8", trust_remote_code=True)