witcheer

witcheer's activity

• All
• Models
• Datasets
• Spaces
• Buckets
• Papers
• Collections
• Community
• Posts
• Upvotes
• Likes
• Articles

New activity in witcheer/microduck-roulade 24 days ago

Metadata for Microduck policy

1

#1 opened 24 days ago by

kramp

New activity in witcheer/microduck-kick-right 24 days ago

Metadata for Microduck policy

#1 opened 24 days ago by

kramp

New activity in witcheer/microduck-standup 24 days ago

Metadata for Microduck policy

#1 opened 24 days ago by

kramp

New activity in witcheer/microduck-walk-recover 24 days ago

Metadata for Microduck policy

#1 opened 24 days ago by

kramp

New activity in witcheer/microduck-walk-rough 24 days ago

Metadata for Microduck policy

#1 opened 24 days ago by

kramp

New activity in witcheer/microduck-walk-flat 24 days ago

Metadata for Microduck policy

#1 opened 24 days ago by

kramp

New activity in witcheer/windows-rtx-4060ti-8gb-moe-offload-bench-2026-05 5 months ago

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

#11 opened 5 months ago by

witcheer

New activity in witcheer/local-agentic-coding-bench-8gb-vram-2026-05 5 months ago

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

#5 opened 5 months ago by

witcheer

New activity in witcheer/windows-rtx-4060ti-8gb-moe-offload-bench-2026-05 5 months ago

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

#10 opened 5 months ago by

witcheer

New activity in witcheer/local-agentic-coding-bench-8gb-vram-2026-05 5 months ago

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

#4 opened 5 months ago by

witcheer

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

#3 opened 5 months ago by

witcheer

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

#1 opened 5 months ago by

witcheer

New activity in witcheer/windows-rtx-4060ti-8gb-moe-offload-bench-2026-05 5 months ago

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

#9 opened 5 months ago by

witcheer

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

#8 opened 5 months ago by

witcheer

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

#7 opened 5 months ago by

witcheer

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

#6 opened 5 months ago by

witcheer

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

#5 opened 5 months ago by

witcheer

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

#4 opened 5 months ago by

witcheer

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

#3 opened 5 months ago by

witcheer

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

#2 opened 5 months ago by

witcheer