Unverified50% confidenceBenchmarkExact time
One B70 user running vLLM + GPT-Q INT4 quantization with MTP depth 4 achieved 84.6 tokens/s on a 27B model, but this only supports single-stream and uses different quantization, making direct comparison impossible.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/5/2026
First Seen
Valid until: 1/3/2027
Sources
Related Entities
Related Claims
UnverifiedThe GGUF quantized version Q4_K_M of MiniCPM 2B has a model size of only 1.56G and can run smoothly on devices with 4G VRAM.72% similarVerifiedQLoRA introduces 4-bit quantization technology, making it possible to fine-tune models with 65 billion parameters on a single RTX 3090/4090 GPU with 24 GB of VRAM.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/977981API
curl https://kongchang.com/api/v1/knowledge/claims/977981MCP
get_claim(id=977981)