model family

qwen-2-5

3 members have been measured on real machines. Every number below was signed by the machine that produced it — none of it is from a model card.

what each build proved
0.5b · Q4_K_M · Qwen
0.5b · Q4_K_M · Qwen
0.5b
0.5b · Q4_K_M · Qwen · Tool calling — proved0.5b · Q4_K_M · Qwen · Tool loop finished — proved0.5b · Q4_K_M · Qwen · Faithful to tool results — proved0.5b · Q4_K_M · Qwen · Structured output — proved0.5b · Q4_K_M · Qwen · Sustained output — proved0.5b · Q4_K_M · Qwen · Writes code — proved0.5b · Q4_K_M · Qwen · Follows instructions — proved0.5b · Q4_K_M · Qwen · Tool calling — proved0.5b · Q4_K_M · Qwen · Tool loop finished — proved0.5b · Q4_K_M · Qwen · Faithful to tool results — proved0.5b · Q4_K_M · Qwen · Structured output — proved0.5b · Q4_K_M · Qwen · Sustained output — proved0.5b · Q4_K_M · Qwen · Writes code — proved0.5b · Q4_K_M · Qwen · Follows instructions — proved0.5b · Tool calling — never asked0.5b · Tool loop finished — never asked0.5b · Faithful to tool results — never asked0.5b · Structured output — never asked0.5b · Sustained output — never asked0.5b · Writes code — never asked0.5b · Follows instructions — never asked
Tool calling
Tool loop finished
Faithful to tool results
Structured output
Sustained output
Writes code
Follows instructions
hf.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF:q4_k_m0.5b · Q4_K_M · 3 machines · 630M
8,192recalled
—served
124.8best tok/s
84.9 tok/s · unnamed machine39.7 tok/s · x86 · CPU · Hyderabad31.2 tok/s · unnamed machine
31.2 → 84.9 tok/s across 3 machines · ×2.7 spreadunnamed machinex86 · CPU
recalled 8k⌐ declared 32k · unsignedadvertises 4× what it proved
hf.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF:Q4_K_M0.5b · Q4_K_M · 3 machines · 630M
8,192recalled
—served
124.8best tok/s
84.9 tok/s · unnamed machine31.2 tok/s · unnamed machine25.5 tok/s · x86 · CPU · Hyderabad
25.5 → 84.9 tok/s across 3 machines · ×3.3 spreadunnamed machinex86 · CPU
recalled 8k⌐ declared 32k · unsignedadvertises 4× what it proved
qwen2.5:0.5b0.5b · 1 machine
—recalled
—served
4.8best tok/s
ask your own gram

Your concierge can answer this from the network rather than from this page. The ask carries the operator id, so it resolves to a capability and an engine — not to a name somebody has to recognise.

What has this network measured about the qwen-2-5 family (epn.family.qwen-2-5), and can my gram run any of it? See /inferences/qwen-2-5

open a gramx with this ask →epn.family.qwen-2-5