| 63 | {"top_p", "value", "AR nucleus sampling threshold."}, |
| 64 | {"repetition_penalty", "value", "AR repetition penalty."}, |
| 65 | {"max_tokens", "n", "Maximum generated AR content/style tokens."}, |
| 66 | {"num_inference_steps", "n", "Flow-matching denoising steps."}, |
| 67 | {"seed", "n", "Request seed; omitted means random."}, |
| 68 | {"fm_noise_file", "path", "Optional flow-matching noise tensor file."}, |
| 69 | }; |
| 70 | out.session_options = { |
| 71 | {"vevo2.weight_type", "native|f32|f16|bf16|q8_0", "Default matmul weight storage type."}, |
| 72 | {"vevo2.conv_weight_type", "native|f32|f16", "Default convolution weight storage type."}, |
| 73 | {"vevo2.ar_weight_type", "native|f32|f16|bf16|q8_0", "AR matmul weight storage type."}, |
| 74 | {"vevo2.ar_weight_context_mb", "n", "AR weight context size."}, |
| 75 | {"vevo2.ar_prefill_graph_context_mb", "n", "AR prefill graph context size."}, |
| 76 | {"vevo2.ar_decode_graph_context_mb", "n", "AR decode graph context size."}, |
| 77 | {"vevo2.whisper_weight_type", "native|f32|f16|bf16|q8_0", "Whisper matmul weight storage type."}, |
| 78 | {"vevo2.whisper_conv_weight_type", "native|f32|f16", "Whisper convolution weight storage type."}, |
| 79 | {"vevo2.whisper_weight_context_mb", "n", "Whisper weight context size."}, |
no test coverage detected