Ricky

SkyIsNotGreen

sky-is-green

AI & ML interests

Independent software engineer focused on; long-horizon memory, ternary quantization, LLM forensics & analysis, local-hosting on consumer hardware, and AI portability.

Recent Activity

new activity about 3 hours ago

prism-ml/Ternary-Bonsai-2-27B-gguf:Independent test: 98% retention on easier tasks, 87% on harder reasoning/coding (vs 4-bit Qwen3.8 27B)

upvoted an article 2 days ago

Build specialized multimodal models with LFM3

new activity 2 days ago

deepseek-ai/DeepSeek-V4.1-Flash-1017:DeepSeek-V4.1-Flash-GSQ on one RX 7900 XT vRAM 20 GB + sysRAM 31.7 GiB: 121–127 tok/s prefill, 16 tok/s decode at 6K, 23 s for a 648-token turn

View all activity

Organizations

New activity in prism-ml/Ternary-Bonsai-2-27B-gguf about 3 hours ago

Independent test: 98% retention on easier tasks, 87% on harder reasoning/coding (vs 4-bit Qwen3.8 27B)

🔥❤️ 5

#102 opened 6 days ago by

heldoutlab

upvoted an article 2 days ago

Article

Build specialized multimodal models with LFM3

LiquidAI

•

2 days ago

• 74

New activity in deepseek-ai/DeepSeek-V4.1-Flash-1017 2 days ago

DeepSeek-V4.1-Flash-GSQ on one RX 7900 XT vRAM 20 GB + sysRAM 31.7 GiB: 121–127 tok/s prefill, 16 tok/s decode at 6K, 23 s for a 648-token turn

#38 opened 2 days ago by

SkyIsNotGreen

updated 3 models 3 days ago

New activity in Qwen/Qwen3.8-Flash-Next 3 days ago

Qwen3.8-Flash-Next at 57.6 GB: 80.5 tok/s on 2× 7900 XT, with only 32 GB of RAM

#79 opened 3 days ago by

SkyIsNotGreen

updated a model 8 days ago

SkyIsNotGreen/splinter-memory

updated a dataset 8 days ago

SkyIsNotGreen/hive-cascade

Updated 8 days ago • 25

updated a Space 8 days ago

The Hive Cascade: detect -> repair -> escalate

🐝

Verify and route LLM answers with interactive memory control

published a Space 8 days ago

The Hive Cascade: detect -> repair -> escalate

🐝

Verify and route LLM answers with interactive memory control

published a dataset 8 days ago

SkyIsNotGreen/hive-cascade

Updated 8 days ago • 25

published a model 8 days ago

SkyIsNotGreen/splinter-encoder

updated a model 8 days ago

SkyIsNotGreen/splinter-encoder

published a model 10 days ago

SkyIsNotGreen/splinter-memory

New activity in prism-ml/Ternary-Bonsai-2-27B-gguf 12 days ago

Request to bonsai

👍 4

#95 opened 14 days ago by

Banaxi-Tech

updated a dataset 14 days ago

SkyIsNotGreen/decision-heads

Viewer • Updated 14 days ago • 6.23k • 1.01k

liked a model 14 days ago

DavidAU/NVIDIA-Nemotron-3.5-Bolt-Turbo-Brilliance-50x-74B-A9B-24GB-GGUF

Image-Text-to-Text • 74B • Updated 9 days ago • 12.6k • 27

New activity in SkyIsNotGreen/Scion-FlashNext-176B-A6B 14 days ago

Whaaaaat? hmm, why A6B vs A3B?

#2 opened 14 days ago by

MultivariatePerspective777

New activity in SkyIsNotGreen/Scion-35B-A3B 15 days ago

next version of the model?

#4 opened 17 days ago by

MultivariatePerspective777

Ricky

AI & ML interests

Recent Activity

Organizations

SkyIsNotGreen's activity

Independent test: 98% retention on easier tasks, 87% on harder reasoning/coding (vs 4-bit Qwen3.8 27B)

Build specialized multimodal models with LFM3

DeepSeek-V4.1-Flash-GSQ on one RX 7900 XT vRAM 20 GB + sysRAM 31.7 GiB: 121–127 tok/s prefill, 16 tok/s decode at 6K, 23 s for a 648-token turn

Qwen3.8-Flash-Next at 57.6 GB: 80.5 tok/s on 2× 7900 XT, with only 32 GB of RAM

The Hive Cascade: detect -> repair -> escalate

The Hive Cascade: detect -> repair -> escalate

Request to bonsai

Whaaaaat? hmm, why A6B vs A3B?

next version of the model?