fish_drama3: increase default text chunk size to 4000

docs/models/fish_drama3.md

@@ -239,7 +239,7 @@ checkpoint or session setting, or leaves the optional value unset as described.
 | `previous_takes` | `string` | Model/session default | Continuation/edit history as a JSON array of {text, vq_codes or audio_path}; a string generates that earlier take first. |
 | `previous_takes_file` | `string` | Model/session default | JSON history file containing a take object or an ordered array of earlier takes. Avoids the Windows command-line length limit for VQ grids. |
 | `max_tokens` | `int` | `1024` | Maximum generated acoustic frames per text chunk (about 21.5 frames/second). Declared range: minimum 1. |
-| `text_chunk_size` | `int` | `128` | Maximum characters per long-form chunk. Declared range: minimum 1. |
+| `text_chunk_size` | `int` | `4000` | Maximum characters per long-form chunk. Declared range: minimum 1. |
 | `skip_empty_chunks` | `bool` | `false` | Log and skip chunks that emit EOS before producing audio. The request fails if all chunks are empty. |
 | `text_chunk_mode` | `word_budget`, `tag_aware`, `japanese`, `endline` | `tag_aware` | Framework text chunking mode. |
 | `temperature` | `float` | `1.0` | Main and residual sampling temperature. Zero makes the initial draws greedy; semantic repeat redraws retain the fixed Drama 3 RAS settings. Declared range: minimum 0. |
@@ -462,7 +462,7 @@ synchronously and calls the event sink while audio is generated; `finalize()`
 returns the complete audio and the reusable take artifact. An application can
 invoke generation on its worker thread to keep its UI responsive.
 
-Long-form text uses the framework chunker, with `text_chunk_size=128` and
+Long-form text uses the framework chunker, with `text_chunk_size=4000` and
 `text_chunk_mode=tag_aware` by default. Each new chunk includes the first generated
 chunk as an anchor and the latest chunk as recent context, without duplicating
 the first chunk. If that prompt does not fit, automatic history is reduced to the first chunk only, then removed.

model_specs/fish_drama3.json

@@ -81,7 +81,7 @@
         "description": "Maximum characters per long-form chunk.",
         "required": false,
         "min": 1,
-        "default": 128
+        "default": 4000
       },
       {
         "name": "skip_empty_chunks",

src/community_models/fish_drama3/session.cpp

@@ -204,7 +204,7 @@ runtime::TaskResult Drama3Session::synthesize(const runtime::TaskRequest & reque
     }
     const auto skip_empty = runtime::find_option(effective.options, {"skip_empty_chunks"});
     const bool skip_empty_chunks = skip_empty && runtime::parse_bool_option(*skip_empty, "skip_empty_chunks");
-    const int64_t chunk_size = runtime::parse_positive_i64_option(effective.options, {"text_chunk_size"}, 128);
+    const int64_t chunk_size = runtime::parse_positive_i64_option(effective.options, {"text_chunk_size"}, 4000);
     const auto mode = engine::text::parse_text_chunk_mode_override(effective.options).value_or(engine::text::TextChunkMode::TagAware);
     const auto chunks = runtime::chunk_text_request(effective, chunk_size, mode);
     runtime::TaskResult result;

webui/configs/model_params.json

@@ -130,7 +130,7 @@
     {"name":"top_k","type":"number","label":"Top classes","default":10,"minimum":1,"maximum":527,"step":1}
   ],
   "fish_drama3": [
-    {"name":"text_chunk_size","type":"number","label":"Characters per chunk","default":128,"minimum":1,"step":1,"info":"Smaller chunks reduce waveform-decoder peak memory; larger chunks reduce repeated prefills."},
+    {"name":"text_chunk_size","type":"number","label":"Characters per chunk","default":4000,"minimum":1,"step":1,"info":"Smaller chunks reduce waveform-decoder peak memory; larger chunks reduce repeated prefills."},
     {"name":"temperature","type":"slider","label":"Temperature","default":1,"minimum":0,"maximum":2,"step":0.05},
     {"name":"top_p","type":"slider","label":"Top-p","default":0.9,"minimum":0.01,"maximum":1,"step":0.01},
     {"name":"top_k","type":"number","label":"Top-k","default":4097,"minimum":1,"step":1},