Unverified50% confidenceFactExact time
The actual parameter count of the full pipeline is 16.22B, comprising 7.115B for the Transformer, 8.767B for the text encoder (Qwen3-VL 8B), and 0.338B for the VAE.
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedModels such as Llama-3.1-8B are built on a stacked Transformer decoder architecture, typically containing 32 decoder layers, with approximately 250 million parameters per layer and approximately 500MB of VRAM per layer at FP16 precision.67% similarUnverifiedUnoptimized baseline token decoding speeds for running Qwen 3 (27B): RTX 5090 ~75, RTX 4090 ~46, RTX 3090 ~40, M5 Max 31.4, DGX Spark 12–17, Ryzen AI Max 11.5 tokens/s.65% similarUnverifiedWhen running Qwen3 27B (MXFP4) at 16 concurrent sessions, aggregate prefill reaches 6,300–6,800 tok/s, and aggregate decode throughput peaks at approximately 900 tok/s.63% similarUnverifiedQwen3-235B-A22B adopts a MoE architecture, activating only 22B out of 235B total parameters at a time.63% similarVerifiedA typical EnCodec configuration uses 8 layers of RVQ with a codebook size of 1024 per layer, producing 75 frames per second at a 24kHz sampling rate, compressing 1 second of audio into 600 tokens.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/996805API
curl https://kongchang.com/api/v1/knowledge/claims/996805MCP
get_claim(id=996805)