webui: complete the shipped translations

model_specs/dramabox.json

@@ -57,39 +57,39 @@
       {
         "name": "num_inference_steps",
         "type": "int",
-        "description": "Diffusion sampling step count; default comes from config.json, 30 in the current package.",
+        "description": "Diffusion sampling step count; default 30.",
         "required": false,
         "min": 1,
         "default": 30
       },
       {
         "name": "guidance_scale",
         "type": "float",
-        "description": "Classifier-free guidance scale; default comes from config.json, 2.5 in the current package. Values greater than 1 enable CFG.",
+        "description": "Classifier-free guidance scale; default 2.5. Values greater than 1 enable CFG.",
         "required": false,
         "min": 0.0,
         "default": 2.5
       },
       {
         "name": "spatio_temporal_guidance_scale",
         "type": "float",
-        "description": "Spatio-temporal guidance scale; default comes from config.json, 1.5 in the current package. Values greater than 0 enable STG.",
+        "description": "Spatio-temporal guidance scale; default 1.5. Values greater than 0 enable STG.",
         "required": false,
         "min": 0.0,
         "default": 1.5
       },
       {
         "name": "duration_scale",
         "type": "float",
-        "description": "Multiplier applied to the estimated prompt duration when duration_sec is 0; default comes from config.json, 1.1 in the current package.",
+        "description": "Multiplier applied to the estimated prompt duration when duration_sec is 0; default 1.1.",
         "required": false,
         "min": 0.0,
         "default": 1.1
       },
       {
         "name": "reference_duration_sec",
         "type": "float",
-        "description": "Reference voice crop/repeat duration in seconds; default comes from config.json, 10.0 in the current package.",
+        "description": "Reference voice crop/repeat duration in seconds; default 10.0.",
         "required": false,
         "min": 0.0,
         "default": 10.0

model_specs/firered_audio.json

@@ -182,14 +182,6 @@
         "min": 1,
         "default": 1024
       },
-      {
-        "name": "helper_graph_arena_mb",
-        "type": "int",
-        "description": "Helper graph arena size in MiB; default 256.",
-        "required": false,
-        "min": 1,
-        "default": 256
-      },
       {
         "name": "weight_context_mb",
         "type": "int",

model_specs/fish_audio.json

@@ -29,7 +29,7 @@
       {
         "name": "reference_text",
         "type": "string",
-        "description": "Reference transcript used with speaker reference audio.",
+        "description": "Reference transcript used with speaker reference audio. Required whenever the request carries inline reference audio; the model rejects the request without it.",
         "required": false
       },
       {
@@ -57,10 +57,15 @@
       {
         "name": "text_chunk_mode",
         "type": "enum",
-        "description": "Framework text chunking mode; default word_budget.",
-        "preset": "text_chunk_mode_full",
+        "description": "Framework text chunking mode; default default.",
+        "values": [
+          "default",
+          "tag_aware",
+          "japanese",
+          "endline"
+        ],
         "required": false,
-        "default": "word_budget"
+        "default": "default"
       },
       {
         "name": "top_p",

model_specs/granite5asr.json

@@ -22,13 +22,6 @@
   },
   "options": {
     "request": [
-      {
-        "name": "language",
-        "type": "string",
-        "description": "Recognition language (currently English).",
-        "required": false,
-        "default": "en"
-      },
       {
         "name": "audio_chunk_mode",
         "type": "enum",

model_specs/heartmula.json

@@ -26,13 +26,13 @@
         "name": "lyrics",
         "type": "string",
         "description": "Lyrics text.",
-        "required": false
+        "required": true
       },
       {
         "name": "tags",
         "type": "string",
         "description": "Comma-separated music tags.",
-        "required": false
+        "required": true
       },
       {
         "name": "duration_sec",

model_specs/irodori_tts.json

@@ -74,6 +74,8 @@
         "type": "enum",
         "description": "Text chunking mode; default endline.",
         "values": [
+          "default",
+          "tag_aware",
           "japanese",
           "endline"
         ],

model_specs/midashenglm_gen.json

@@ -67,9 +67,9 @@
       {
         "name": "seed",
         "type": "int",
-        "description": "Generation seed; -1 selects a random seed.",
+        "description": "Generation seed; parsed as an unsigned 32-bit value, so negative seeds are rejected.",
         "required": false,
-        "min": -1,
+        "min": 0,
         "max": 2147483647,
         "default": 0
       }

model_specs/minimax_h3.json

@@ -192,6 +192,24 @@
         "min": 0.0,
         "default": 0.1
       },
+      {
+        "name": "first_block_cache_start_percent",
+        "type": "float",
+        "description": "Start of the denoise window where the first-block cache may be used, as a fraction of the schedule.",
+        "required": false,
+        "min": 0.0,
+        "max": 1.0,
+        "default": 0.1
+      },
+      {
+        "name": "first_block_cache_end_percent",
+        "type": "float",
+        "description": "End of the denoise window where the first-block cache may be used, as a fraction of the schedule; must be greater than first_block_cache_start_percent.",
+        "required": false,
+        "min": 0.0,
+        "max": 1.0,
+        "default": 0.95
+      },
       {
         "name": "first_block_cache_start_sigma",
         "type": "float",

model_specs/minimax_music3.json

@@ -80,6 +80,47 @@
         "required": false,
         "min": 0,
         "default": 0
+      },
+      {
+        "name": "flow_uncond_interval",
+        "type": "int",
+        "description": "Evaluate the flow unconditional CFG branch only every N-th step and reuse the cached guidance delta in between; 1 keeps the exact reference trajectory.",
+        "required": false,
+        "min": 1,
+        "default": 1
+      },
+      {
+        "name": "flow_uncond_warmup",
+        "type": "int",
+        "description": "Number of initial flow steps that always evaluate both CFG branches when delta reuse is enabled.",
+        "required": false,
+        "min": 0,
+        "default": 2
+      },
+      {
+        "name": "ensemble_takes",
+        "type": "int",
+        "description": "Decode N independent takes of the same prompt in one batched AR pass (seeds seed..seed+N-1); outputs are returned as named audio take_01..take_NN.",
+        "required": false,
+        "min": 1,
+        "max": 16,
+        "default": 1
+      },
+      {
+        "name": "ensemble_prefix_frames",
+        "type": "int",
+        "description": "Intro-lock for ensembles: decode the first N AR frames once as a shared master trajectory (~25 frames per second), then fork the takes; 0 disables.",
+        "required": false,
+        "min": 0,
+        "default": 0
+      },
+      {
+        "name": "flow_chunk_hop_frames",
+        "type": "int",
+        "description": "Flow chunk hop in AR frames (~25/sec); 0 keeps the model config (100, 50% chunk overlap).",
+        "required": false,
+        "min": 0,
+        "default": 0
       }
     ],
     "session": [
@@ -141,6 +182,13 @@
         "description": "Load large generation stages only while they are needed to reduce peak VRAM.",
         "required": false,
         "default": true
+      },
+      {
+        "name": "pipeline_overlap",
+        "type": "bool",
+        "description": "Overlap AR decoding with per-chunk condition/flow/vocoder work on a second backend stream. Requires mem_saver=false; falls back to the sequential pipeline otherwise.",
+        "required": false,
+        "default": false
       }
     ],
     "load": []

model_specs/miocodec.json

@@ -26,6 +26,60 @@
       "speaker_reference"
     ]
   },
+  "options": {
+    "request": [],
+    "session": [
+      {
+        "name": "weight_type",
+        "type": "enum",
+        "description": "Codec weight storage type; default f32.",
+        "preset": "weight_type_full",
+        "required": false,
+        "default": "f32"
+      },
+      {
+        "name": "weight_context_mb",
+        "type": "int",
+        "description": "Codec weight descriptor context size in MiB; default 256.",
+        "required": false,
+        "min": 1,
+        "default": 256
+      },
+      {
+        "name": "constant_context_mb",
+        "type": "int",
+        "description": "Reusable constant context size in MiB; default 256.",
+        "required": false,
+        "min": 1,
+        "default": 256
+      },
+      {
+        "name": "content_graph_arena_mb",
+        "type": "int",
+        "description": "Content encoder graph arena size in MiB; default 512.",
+        "required": false,
+        "min": 1,
+        "default": 512
+      },
+      {
+        "name": "global_graph_arena_mb",
+        "type": "int",
+        "description": "Global encoder graph arena size in MiB; default 256.",
+        "required": false,
+        "min": 1,
+        "default": 256
+      },
+      {
+        "name": "wave_graph_arena_mb",
+        "type": "int",
+        "description": "Wave decoder graph arena size in MiB; default 512.",
+        "required": false,
+        "min": 1,
+        "default": 512
+      }
+    ],
+    "load": []
+  },
   "runtime": {
     "tags": [
       "gguf"

model_specs/mms_forced_aligner.json

@@ -61,7 +61,7 @@
       {
         "name": "return_timestamps",
         "type": "bool",
-        "description": "Request word timestamps in the result; set automatically by --words-out.",
+        "description": "Accepted for cross-model compatibility (--words-out sets it) and ignored: the forced aligner always returns word timestamps.",
         "required": false,
         "default": true
       }

model_specs/outetts.json

@@ -107,7 +107,7 @@
       {
         "name": "reference_text",
         "type": "string",
-        "description": "Transcript matching the reference voice audio for voice cloning.",
+        "description": "Transcript matching the reference voice audio for voice cloning. Required whenever reference audio is supplied; the model rejects voice cloning without it.",
         "required": false
       },
       {
@@ -128,10 +128,15 @@
       {
         "name": "text_chunk_mode",
         "type": "enum",
-        "description": "Framework long-form text chunking mode; default word_budget.",
-        "preset": "text_chunk_mode_full",
+        "description": "Framework long-form text chunking mode; default default.",
+        "values": [
+          "default",
+          "tag_aware",
+          "japanese",
+          "endline"
+        ],
         "required": false,
-        "default": "word_budget"
+        "default": "default"
       }
     ],
     "session": [

model_specs/parakeet_tdt.json

@@ -54,6 +54,12 @@
   },
   "options": {
     "request": [
+      {
+        "name": "language",
+        "type": "string",
+        "description": "Transcription language from the OpenAI-compatible request field. Accepted for API compatibility; Parakeet TDT detects the language itself and does not use this value (see keep_language_tags).",
+        "required": false
+      },
       {
         "name": "max_tokens",
         "type": "int",