qwen3-4b-gdn-wsd-ladder – AI Model by pinkskin | AlphaNeural AI

You can deploy this model and start earning money today!

pinkskin/
qwen3-4b-gdn-wsd-ladder

like

0

Views

No views yet

Qwen3-4B -> GDN hybrid: WSD OPD ladder (all final checkpoints)

One subfolder per arm; each README states that arm's decay window.

checkpoint	decay	horizon
`wsd-flat-200M`	160.8M -> 200.0M (39.2M, 2e-5 -> 1e-6)	16K
`wsd-split-200M`	160.8M -> 200.0M (39.2M, 2e-5 -> 1e-6 (GDN) / 2e-6 -> 1e-7 (inherited))	16K
`wsd-flat-ext50-250M`	210.7M -> 250.0M (39.3M, 2e-5 -> 1e-6)	16K
`wsd-flat-ext100-300M`	260.7M -> 300.0M (39.3M, 2e-5 -> 1e-6)	16K
`wsd-flat-ext150-350M`	290.0M -> 350.0M (60.0M, 2e-5 -> 1e-6)	16K
`wsd-flat-ext200-400M`	340.0M -> 400.0M (60.0M, 2e-5 -> 1e-6)	16K
`wsd-split-ext200-400M`	340.0M -> 400.0M (60.0M, 2e-5 -> 1e-6 (GDN) / 2e-6 -> 1e-7 (inherited))	16K
`wsd-flat-ext300-500M`	430.0M -> 500.0M (70.0M, 2e-5 -> 1e-6)	16K
`wsd-flat-ext400-600M`	520.0M -> 600.0M (80.0M, 2e-5 -> 1e-6)	16K
`wsd-flat-ext500-700M`	610.0M -> 700.0M (90.0M, 2e-5 -> 1e-6)	16K
`wsd-flat-ext600-800M-32k`	710.0M -> 800.0M (90.0M, 2e-5 -> 1e-6)	32K
`wsd-flat-ext700-900M-32k`	800.0M -> 900.0M (100.0M, 2e-5 -> 1e-6)	32K
`wsd-flat-ext800-1000M-32k`	890.0M -> 1000.0M (110.0M, 2e-5 -> 1e-6)	32K
`wsd-flat-3e5-32k-200M`	160.8M -> 200.0M (39.2M, 3e-5 -> 1.5e-6)	32K
`wsd-flat-ext200-decay0`	340.0M -> 403.2M (63.2M, 2e-5 -> 0)	16K
`wsd-flat-ext300-decay0`	430.0M -> 503.7M (73.7M, 2e-5 -> 0)	16K
`wsd-flat-ext400-decay0`	520.0M -> 604.2M (84.2M, 2e-5 -> 0)	16K
`wsd-flat-ext500-decay0`	610.0M -> 704.7M (94.7M, 2e-5 -> 0)	16K
`wsd-flat-ext600-decay0`	710.0M -> 804.7M (94.7M, 2e-5 -> 0)	32K
`wsd-flat-ext700-decay0`	800.0M -> 905.3M (105.3M, 2e-5 -> 0)	32K
`wsd-flat-ext800-decay0`	890.0M -> 1005.8M (115.8M, 2e-5 -> 0)	32K