Skip to main content
POST
Generate bounded reviewed configuration candidates without provider mutation or performance claims

Authorizations

​
Authorization
string
header
required

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

Body

application/json
​
gpu
string
required
Required string length: 1 - 128
​
model_identity
string
required
Required string length: 1 - 256
​
provider
string
required
Required string length: 1 - 128
​
compute_connection_id
string

Verified tenant compute connection used to derive and immutably bind the executable provider adapter.

Maximum string length: 255
​
gpu_count
integer
default:1
Required range: 1 <= x <= 1024
​
include_simulated
boolean
default:false
​
max_candidates
integer
default:10
Required range: 1 <= x <= 100
​
max_error_rate
number
Required range: 0 <= x <= 1
​
max_hourly_cost
number
Required range: x >= 0
​
max_tpot_p95_ms
number
Required range: x >= 0
​
max_ttft_p95_ms
number
Required range: x >= 0
​
min_goodput
number
Required range: x >= 0
​
min_output_tokens_second
number
Required range: x >= 0
​
model_revision
string
Pattern: ^[a-fA-F0-9]{40,64}$
​
objective
enum<string>
default:interactive
Available options:
interactive,
latency,
throughput,
cost-efficiency
​
region
string
Maximum string length: 128
​
runtimes
string[]
Maximum array length: 16
Required string length: 1 - 64
​
target_concurrency
number
​
workload_fingerprint
string
Maximum string length: 256
​
workload_profile
enum<string>
Available options:
balanced,
buffered,
interactive,
long-context,
long-generation,
overload,
public-decode-heavy,
public-interactive,
public-long-prefill,
throughput
​
workload_source
enum<string>
Available options:
reviewed_default,
public_prior,
customer_observed

Response

Success

The response is of type object.