# tinker.types.ClientConfigResponse

## *class* [**tinker.types.ClientConfigResponse**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L8)(*[BaseModel](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/_models.py#L88)*)

Server-side feature flags resolved for this caller.

Uses BaseModel (extra="ignore") so new flags from the server are silently dropped until the SDK adds fields for them.

**Fields:**

- [**pjwt_auth_enabled**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L15) (*bool*)
- [**credential_default_source**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L16) (*str*)
- [**sample_dispatch_bytes_semaphore_size**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L17) (*int*)
- [**inflight_response_bytes_semaphore_size**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L18) (*int*)
- [**parallel_fwdbwd_chunks**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L19) (*bool*)
- [**proto_compress_fwdbwd**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L20) (*bool*) – When true, the SDK zstd-compresses the proto fwd/bwd request body and sets Content-Encoding: zstd. Real fwd/bwd payloads compress >10× — the API server decompresses transparently via an ASGI middleware.
- [**fwdbwd_max_chunk_len**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L24) (*int*) – Maximum number of datums per fwd/bwd chunk request.
- [**fwdbwd_max_chunk_bytes_count**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L26) (*int*) – Maximum estimated request payload bytes (serialized proto, before zstd compression) per fwd/bwd chunk request.
- [**fwdbwd_dispatch_bytes_semaphore_size**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L29) (*int*) – Maximum estimated fwd/bwd request payload bytes being dispatched at the same time (same pre-compression estimate as fwdbwd_max_chunk_bytes_count). Bounds client memory for serialized bodies and smooths submission bursts.
- [**billing_exception_max_pause_duration_sec**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L33) (*int*)
- [**sample_no_retries**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L34) (*bool*)
- [**sample_enable_stuck_detection**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L35) (*bool*) – When true, the SDK runs the retry handler's progress timeout check that raises `APIConnectionError("...Requests appear to be stuck.")` when no progress is made within `RetryConfig.progress_timeout`.
- [**sample_max_concurrent_requests**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L39) (*int*) – Maximum number of in-flight sampling requests a SamplingClient will allow (the size of its retry handler's semaphore). Always applied as the sampling `RetryConfig.max_connections`, overriding any value in a caller-provided `retry_config`.
- [**use_pyqwest_transport**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L44) (*bool*) – When true, the SDK builds its default httpx async client on top of the pyqwest (reqwest/hyper-based) transport adapter. Set to false server-side to force every client to fall back to httpx's default transport.
- [**create_model_via_load_weights**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/client_config_response.py#L48) (*bool*) – When true, create_training_client_from_state\* skips the weights_info and create_model round-trips and sends a single LoadWeightsRequest with session addressing (session_id + model_seq_id) that creates the model server-side, configured from the checkpoint's owning model. Requires a server with load-weights-first support.