@@ -222,14 +222,14 @@ message ProposeBlockResponse {
222222// no numpy scalar, so it travels as the logical dtype "bfloat16" over a uint16
223223// bit buffer and is rebuilt by the torch/mlx bridge at the endpoint.
224224message Tensor {
225- string dtype = 1 ; // float32|float16|bfloat16|int32|int64|uint32|bool
225+ string dtype = 1 ; // float32|float16|bfloat16|int32|int64|uint32|bool
226226 repeated int64 shape = 2 ;
227227 bytes data = 3 ;
228228}
229229
230230// LayerKV is one verifier layer's restored K and V banks.
231231message LayerKV {
232- int32 layer = 1 ; // verifier layer index this K/V belongs to
232+ int32 layer = 1 ; // verifier layer index this K/V belongs to
233233 Tensor k = 2 ;
234234 Tensor v = 3 ;
235235}
@@ -289,7 +289,9 @@ message SeedContextRequest {
289289 repeated int32 positions = 3 ;
290290}
291291
292- message SeedContextResponse { uint32 context_len = 1 ; }
292+ message SeedContextResponse {
293+ uint32 context_len = 1 ;
294+ }
293295
294296message DraftBlockRequest {
295297 string session_id = 1 ;
@@ -301,7 +303,7 @@ message DraftBlockRequest {
301303}
302304
303305message DraftBlockResponse {
304- repeated uint32 draft_token_ids = 1 ; // exactly block_size drafts
306+ repeated uint32 draft_token_ids = 1 ; // exactly block_size drafts
305307 uint32 forward_passes = 2 ;
306308 uint64 peak_activation_bytes = 3 ;
307309}
@@ -313,8 +315,12 @@ message ExtendContextRequest {
313315 repeated int32 positions = 3 ;
314316}
315317
316- message ExtendContextResponse { uint32 context_len = 1 ; }
318+ message ExtendContextResponse {
319+ uint32 context_len = 1 ;
320+ }
317321
318- message DFlashProposerServiceCloseSessionRequest { string session_id = 1 ; }
322+ message DFlashProposerServiceCloseSessionRequest {
323+ string session_id = 1 ;
324+ }
319325
320326message DFlashProposerServiceCloseSessionResponse {}
0 commit comments