Ricky
SkyIsNotGreen
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
updated a model about 1 hour ago
SkyIsNotGreen/Scion-FlashNext-176B-A6B new activity about 2 hours ago
Qwen/Qwen3.8-Flash-Next:Qwen3.8-Flash-Next at 57.6 GB: 80.5 tok/s on 2× 7900 XT, with only 32 GB of RAM updated a model 5 days ago
SkyIsNotGreen/splinter-memoryOrganizations
Qwen3.8-Flash-Next at 57.6 GB: 80.5 tok/s on 2× 7900 XT, with only 32 GB of RAM
#79 opened about 2 hours ago
by SkyIsNotGreen
Request to bonsai
👍 4
8
#95 opened 11 days ago
by 
Banaxi-Tech
Whaaaaat? hmm, why A6B vs A3B?
2
#2 opened 11 days ago
by 
MultivariatePerspective777
next version of the model?
7
#4 opened 14 days ago
by MultivariatePerspective777
Bonsai 2 in Strata: 64.4 tok/s on one RX 7900 XT with no MTP.
#94 opened 12 days ago
by SkyIsNotGreen
Ternary Training
6
#2 opened 28 days ago
by MultivariatePerspective777
Independent results for Bonsai-2 PQ2_0 (GSM8K / MATH / HumanEval) with caveats
#87 opened 18 days ago
by SkyIsNotGreen
What level of performance does it achieve?
12
#3 opened 24 days ago
by 
jian2023
mtp on A3B ternary model
1
#1 opened 26 days ago
by MultivariatePerspective777
maybe I miss something ...
2
#1 opened 29 days ago
by 
RegisteredWednesday
Ternary-based 35B-A3B I made, using this distill.
👍 1
1
#8 opened 29 days ago
by SkyIsNotGreen
I took Bonsai 2 27B apart, part 2: the format is public, the last 8% is ordinary ternary QAT.
👀 1
6
#62 opened about 1 month ago
by SkyIsNotGreen
I took Bonsai 2 27B apart: the format is public, the last 8% is trained
👀❤️ 7
10
#44 opened about 1 month ago
by SkyIsNotGreen
I took Bonsai 2 27B apart: the format is public, the last 8% is trained
👀❤️ 7
10
#44 opened about 1 month ago
by SkyIsNotGreen
I took Bonsai 2 27B apart: the format is public, the last 8% is trained
👀❤️ 7
10
#44 opened about 1 month ago
by SkyIsNotGreen