ThinkingCap Kearuga EXL3: TandemLLM-first DGX Spark serving, independent RTX 5090 receipts and a retained TensorFold reproduction recipe.
-
Updated
Oct 19, 2026 - Python
#
thinkingcap
Here are 4 public repositories matching this topic...
Qwen3.8-27B and the ThinkingCap fine-tune at 262K context on dual or quad RTX 3090s: Merlin engine v2.5.2 + 11 patches, FP8 KV pool 826,689 (dual) / 1,974,072 (quad), MTP K=4
-
Updated
Oct 22, 2026 - Python
Verified self-replay for shorter Qwen3 reasoning: ThinkingCap results, runnable experiments, and RunPod/Vast workflows.
-
Updated
Nov 7, 2026 - Python
Claude Code on a local 125B MoE (Qwen3.8-Flash-Next) on a 12 GB GPU: setup and measurements, a local agent that drives ComfyUI, and VS Code Remote-SSH fixes
windows vscode moe strata ai-agents mixture-of-experts remote-ssh comfyui local-llm llm-inference qwen anthropic-api abliteration claude-code consumer-gpu nvfp4 rtx-4070 minimax-h3 thinkingcap
-
Updated
Nov 6, 2026 - Python
Add this topic to your repo
To associate your repository with the thinkingcap topic, visit your repo's landing page and select "manage topics."