Description
Return one expert's routing load and overflow.
Inference budgets depend on the number of stored values, sequence lengths, routing assignments and the chosen accounting basis. Logical cache bytes and padded token counts differ from actual device allocation, latency and billed usage. Caller-supplied prices, cache fractions and capacity assumptions must match the intended system.
Return one expert's routing load and overflow.
When to use this tool
Use Mixture Router Load One Expert Term when reconciling the stated cache, padding, routing or price basis.
How to use the tool and how it works
- Enter expertAssignments, totalAssignments, totalTokens, capacity. Use the units and conventions below.
- Run the tool. Correct any format, range, or cross-field input error before interpreting the output.
- Read loadShare, assignmentsPerToken, overflow. Use the output convention below.
Return one expert's routing load and overflow. Require positive whole totalAssignments/totalTokens, bounded expertAssignments and nonnegative capacity; loadShare = expert/totalAssignments, assignmentsPerToken = expert/totalTokens, overflow floors expert-capacity at zero.1
| Input field | Meaning and units | Accepted input |
|---|---|---|
| expertAssignments | Expert Assignments | Required. Safe integer; minimum 0. |
| totalAssignments | Total Assignments | Required. Safe integer; minimum 1. |
| totalTokens | Total Tokens | Required. Safe integer; minimum 1. |
| capacity | Capacity | Required. Safe integer; minimum 0. |
| Output field | Interpretation | Returned shape |
|---|---|---|
| loadShare | expertAssignments/totalAssignments. | Required. Finite number. |
| assignmentsPerToken | expertAssignments/totalTokens; a different denominator from loadShare. | Required. Finite number. |
| overflow | Overflow | Required. Safe integer. |
{
"input": {
"expertAssignments": 40,
"totalAssignments": 100,
"totalTokens": 50,
"capacity": 30
},
"output": {
"loadShare": 0.4,
"assignmentsPerToken": 0.8,
"overflow": 10
}
}Limitations and assumptions
- Assignments and tokens use different denominators; no routing, reassignment, load balancing or actual model execution occurs.
- Inputs must satisfy both the field types and the cross-field conditions above. Invalid input causes an error; a successful false flag or documented placeholder is a different result.
- Numeric calculations use binary floating point unless exact decimal-integer arithmetic is explicitly stated. No hidden measurement tolerance or decimal rounding is added; nonfinite or out-of-contract results are rejected. Numeric signed-zero results are exported as zero.
Alternative or Complementary methods
Use actual model/runtime layouts, allocator measurements and current provider billing records for real memory, performance or cost planning.
References
Similar or alternative tools
- Grouped Query Head Replication Factor
Return grouped-query head replication and mapping.
- Attention Head Masked Share
Return masked share for a selected attention head.