Base Compute

company

Verified

https://basecompute.co

AI & ML interests

local AI, GPU kernels, performance optimisation, agentic research pipelines, automated research

Recent Activity

isuramanchi new activity 5 days ago

basecompute/gemma-4-E4B-it:gemma-4-E4B-it-Q8: Google's MTP head embedded (gemma4-mtp, head_version 1)

isuramanchi new activity 5 days ago

basecompute/gemma-4-E2B-it:gemma-4-E2B-it-Q8: Google's MTP head embedded (gemma4-mtp, head_version 1)

prabodbc new activity 9 days ago

basecompute/Qwen3.8-27B-BaseQuant-V1:Add XL tier (25.30 GB, size-matched to unsloth UD-Q6_K_XL)

View all activity

Articles

15% More Accurate Than Unsloth: Qwen-3.8-27B Quants for Local Agents

Introducing Superfluid: the local-AI native LLM server for multi-agent inference

Compute:Arena / Measuring Local Inference Across Models, Quants, Chips, and Runtimes

Base Optimization Stack: From Open Weights to Frontier On-Device Inference Speed

BaseRT: Best-in-Class LLM Inference on Apple Silicon via Native Metal

View all articles

in basecompute/gemma-4-E4B-it 5 days ago

gemma-4-E4B-it-Q8: Google's MTP head embedded (gemma4-mtp, head_version 1)

#1 opened 5 days ago by

in basecompute/gemma-4-E2B-it 5 days ago

gemma-4-E2B-it-Q8: Google's MTP head embedded (gemma4-mtp, head_version 1)

#1 opened 5 days ago by

in basecompute/Qwen3.8-27B-BaseQuant-V1 9 days ago

Add XL tier (25.30 GB, size-matched to unsloth UD-Q6_K_XL)

#3 opened 9 days ago by

Add XS tier (8.36 GB, size-matched to unsloth UD-IQ2_S)

#2 opened 10 days ago by

in basecompute/Qwen3.8-27B-BaseQuant-V1 10 days ago

Memory needed for Small / Medium / Large at 8K context

#1 opened 10 days ago by

updated a model 11 days ago

basecompute/Qwen3.8-27B-BaseQuant-V1

Text Generation • 27B • Updated 9 days ago • 6.81k • 28

updated a model 12 days ago

basecompute/gemma-4-E4B-it-BaseQuant-V1

Updated 12 days ago • 39 • 6

published an article 13 days ago

Article

15% More Accurate Than Unsloth: Qwen-3.8-27B Quants for Local Agents

basecompute

•

13 days ago

• 5

published a dataset 13 days ago

basecompute/quant-fidelity-eval

Viewer • Updated 13 days ago • 240 • 49

updated a dataset 13 days ago

basecompute/quant-fidelity-eval

Viewer • Updated 13 days ago • 240 • 49

published 2 models 13 days ago

basecompute/gemma-4-E4B-it-BaseQuant-V1

Updated 12 days ago • 39 • 6

basecompute/Qwen3.8-27B-BaseQuant-V1

Text Generation • 27B • Updated 9 days ago • 6.81k • 28

in basecompute/Qwen3.8-27B 15 days ago

Q4: add the MTP drafting-vocabulary ranking (mtp.draft_rank), weights unchanged

#3 opened 15 days ago by

in basecompute/Qwen3.6-35B-A3B 15 days ago

Q4: add the MTP drafting-vocabulary ranking (mtp.draft_rank), weights unchanged

#2 opened 15 days ago by

in basecompute/Qwen3.8-27B 21 days ago

Re-convert with the Qwen MTP head (header.speculator): basertd --speculate mtp-head

#2 opened about 2 months ago by

in basecompute/Qwen3.6-35B-A3B 21 days ago

Re-convert with the Qwen MTP head (header.speculator): basertd --speculate mtp-head

#1 opened about 2 months ago by

in basecompute/Qwen3.6-27B 21 days ago

Re-convert with the Qwen MTP head (header.speculator): basertd --speculate mtp-head

#1 opened about 2 months ago by

in basecompute/Qwen3.5-35B-A3B 21 days ago

Re-convert with the Qwen MTP head (header.speculator): basertd --speculate mtp-head

#1 opened about 2 months ago by

in basecompute/Qwen3.5-2B-Base 21 days ago

Re-convert with the Qwen MTP head (header.speculator): basertd --speculate mtp-head

#1 opened about 2 months ago by

in basecompute/Qwen3.5-2B 21 days ago

Re-convert with the Qwen MTP head (header.speculator): basertd --speculate mtp-head

#1 opened about 2 months ago by