witcheer's picture

Open to Work

witcheer PRO

witcheer

·

https://x.com/witcheer

AI & ML interests

Local AI Maxxing.

Recent Activity

updated a dataset about 19 hours ago

witcheer/rtx-5090-benchmarks

updated a dataset 1 day ago

witcheer/agentic-score-leaderboard

updated a bucket 3 days ago

gemma-challenge/gemma-witcheer

View all activity

Organizations

New activity in witcheer/windows-rtx-4060ti-8gb-moe-offload-bench-2026-05 23 days ago

Gemma 4-E4B agentic results: fastest portscout (30s), PARTIAL logpulse

#11 opened 23 days ago by

New activity in witcheer/local-agentic-coding-bench-8gb-vram-2026-05 23 days ago

Gemma 4-E4B: fastest portscout ever (30s), logpulse PARTIAL

#5 opened 23 days ago by

New activity in witcheer/windows-rtx-4060ti-8gb-moe-offload-bench-2026-05 24 days ago

Gemma 4-E4B: 64 tok/s, 6/6 quality (best on leaderboard)

#10 opened 24 days ago by

New activity in witcheer/local-agentic-coding-bench-8gb-vram-2026-05 24 days ago

OmniCoder-9B: fastest code gen, same agent loop problems (2 new rows)

#4 opened 24 days ago by

GBNF structured CoT: fixing Qwen3.6 rumination with a 4-rule grammar (12 new rows)

#3 opened 24 days ago by

New activity in witcheer/local-agentic-coding-bench-8gb-vram-2026-05 25 days ago

deep-dive benchmark: 8 prompts used to test gpt-oss-20b

#1 opened 25 days ago by

New activity in witcheer/windows-rtx-4060ti-8gb-moe-offload-bench-2026-05 27 days ago

Qwen3.6 27B dense: MoE vs dense control experiment (10.8x speed gap)

#9 opened 27 days ago by

New activity in witcheer/windows-rtx-4060ti-8gb-moe-offload-bench-2026-05 29 days ago

Llama 3.2 1B — 228 tok/s speed ceiling, 771 MB

#8 opened 29 days ago by

Qwen3 8B — dense 8.2B, first to pass hallucination test

#7 opened 29 days ago by

New activity in witcheer/windows-rtx-4060ti-8gb-moe-offload-bench-2026-05 30 days ago

GLM 4.7 Flash — 30B MoE + MLA, ncmoe sweep, quality tests

#6 opened 30 days ago by

New activity in witcheer/windows-rtx-4060ti-8gb-moe-offload-bench-2026-05 about 1 month ago

added Mistral 7B v0.3: 56.4 tok/s, best quality (5/6), VRAM-hungry

#5 opened about 1 month ago by

added Gemma 4 E2B: 117.8 tok/s, dense, full GPU, quality traces

#4 opened about 1 month ago by

LFM2 24B A2B — ncmoe sweep + deep stress test results (RTX 4060 Ti 8GB)

#3 opened about 1 month ago by

Gemma 4 26B A4B — ncmoe sweep results added (RTX 4060 Ti 8GB)

#2 opened about 1 month ago by

New activity in witcheer/rtx-4060ti-8gb-turboquant-bench-2026-05 about 1 month ago

new dataset: turboquant KV cache benchmarks

#1 opened about 1 month ago by