DFlash 2 is here! Qwen3.8-27B at 70 tok/s on an M5 Max MacBook Pro. ⚡ Up to 4.6× the speed of autoregressive decoding, with the same output. This is the next generation of DFlash, seeded at Z Lab an
The Hook
“(No spoken dialogue — silent terminal screencast)”
Product Reveal
“(No spoken dialogue — silent terminal screencast)”
Call to Action
“(No spoken dialogue — silent terminal screencast)”
Synthesizable in Remotion · key components of the split-screen-terminal-benchmark template.
Two synchronized terminal panels executing code and streaming Markdown side-by-side with distinct speed tiers.
Render dual absolute-positioned flex columns with fixed monospace font stacks, simulating token streaming via frame-interpolated text slicing.
Dynamic footer badge updating token throughput (TPS) upon completion.
Trigger badge color shift to vibrant green and scale pop using spring physics when the benchmark reaches final token count.
Purpose. Establish task context and model baseline.
Execution. User prompt entered into OpenCode CLI interface.
Purpose. Demonstrate clear speed superiority of speculative decoding.
Execution. Dual-column real-time streaming showing DFlash pulling ahead.
Purpose. Quantify speed gain with final TPS readout.
Execution. Green highlighted throughput badge showing 2x throughput.