measured local AI

Measured tok/s — your machine, one size track at a time.

Personal bests by size — 7B / 14B / 27B / 70B+.

Board track All

Measure: 7B · qwen2.5:7b-instruct

Measure my machine Join Board How Trust

Personal bests by size track

Filters apply within each size.

Clear

7B

1 result · signed

qwen2.5-7b-instruct weights ollama q4_k_m ctx 4096

#	handle	PB tok/s	model	GPU	runs	delta	OS · CPU
1	test	274.60	qwen2.5-7b-instruct	NVIDIA GeForce RTX 5090	13	—	Linux · AMD

14B

1 result · signed

qwen2.5-14b weights ollama q4_k_m ctx 4096

#	handle	PB tok/s	model	GPU	runs	delta	OS · CPU
1	test	146.77	qwen2.5-14b	NVIDIA GeForce RTX 5090	11	—	Linux · AMD

27B

1 result · signed

qwen3.8-27b weights ollama q4_k_m ctx 4096

#	handle	PB tok/s	model	GPU	runs	delta	OS · CPU
1	test	148.25	qwen3.8-27b	NVIDIA GeForce RTX 5090	11	—	Linux · AMD

70B+

1 result · signed

qwen2.5-72b weights ollama q4_k_m ctx 4096

#	handle	PB tok/s	model	GPU	runs	delta	OS · CPU
1	test	2.69	qwen2.5-72b	NVIDIA GeForce RTX 5090	11	—	Linux · AMD

Measure

Copy the 7B pack for your bot.

Measure: 7B · qwen2.5:7b-instruct

MyColo measure on this machine.

Goal: one measured score on https://mycolo.ai for wl-smoke-qwen25-7b-ollama (7B board). No signup required to run this measure. Join only to keep a lasting member record.

Do only this, then stop and show me the board URL:
1. Install CLI once (portable): follow https://mycolo.ai/protocol — ensure mycolo-measure is on PATH. Do NOT use a machine-specific venv path.
2. Local signing key (stays on this box — never generated on the website):
   mycolo-measure keygen
3. ollama show qwen2.5:7b-instruct >/dev/null 2>&1 || ollama pull qwen2.5:7b-instruct   # skip if already local; MyColo does not host weights
4. Supervised measure (agents — preview only, no upload):
   mycolo-measure bench --model qwen2.5:7b-instruct --agent --show-payload --base https://mycolo.ai -o /tmp/mycolo-run.json
   Show me the redacted payload. Confirm workload_id is wl-smoke-qwen25-7b-ollama. Do NOT publish until I say accept.
5. Publish only after I accept:
   - Anonymous / no lasting row: use the human submit path on /protocol when eligible
   - Lasting member row: need handle + token from https://mycolo.ai/join — ask me once for HANDLE; never invent tokens
   mycolo-measure bench --model qwen2.5:7b-instruct --handle HANDLE --agent --submit-mode publish --base https://mycolo.ai -o /tmp/mycolo-run.json
6. Print tok/s, trust labels, 7B badge, bundle id, https://mycolo.ai/n/{id}, https://mycolo.ai/?track=smoke#board
7. Optional: https://mycolo.ai/signoff · Trust: https://mycolo.ai/api/trust

Rules: no deploy, no code changes, no secrets in chat. Coarsened OS/CPU only on public results.

Install once: /protocol. Local keygen for signing; site token only when claiming. Agents: --agent --show-payload, then publish after accept. /api/trust · /signoff

7B · wl-smoke-qwen25-7b-ollama · ollama show qwen2.5:7b-instruct >/dev/null 2>&1 || ollama pull qwen2.5:7b-instruct · weights.

Trust: /api/trust