renderers 0.1.13.dev5__tar.gz → 0.1.13.dev8__tar.gz

{renderers-0.1.13.dev5 → renderers-0.1.13.dev8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: renderers
-Version: 0.1.13.dev5
+Version: 0.1.13.dev8
 Summary: Chat template renderers — deterministic message-to-token conversion for LLM training
 License-Expression: Apache-2.0
 License-File: LICENSE
@@ -114,7 +114,7 @@ next_prompt_ids = r.bridge_to_next_turn(
 )
 ```
 
-Hand-coded renderers ship for `qwen3`, `qwen3-vl`, `qwen3.5`, `qwen3.6`, `qwen3.8`, `gemma4`, `glm-5`, `glm-5.1`, `glm-5.3`, `glm-5.3-flash`, `glm-4.5`, `minimax-m2`, `deepseek-v3`, `deepseek-r1`, `deepseek-v4` (V4 Flash 0731), `kimi-k2`, `kimi-k2.5` / `kimi-k2.6`, `laguna-xs.2`, `laguna-xs-2.1`, `laguna-s-2.1`, `laguna-m.1`, `nemotron-3`, `nemotron-3-ultra`, `nemotron-3.5`, `llama-3`, `gpt-oss`, `hy3`, `inkling` / `inkling-small`, and `prime-qwen3`. Anything else falls back to `DefaultRenderer`, a generic `apply_chat_template` wrapper. `qwen3-vl`, `qwen3.5`, `qwen3.6`, `qwen3.8`, `gemma4`, `glm-5.3-flash`, `kimi-k2.5` / `kimi-k2.6`, and the Inkling checkpoints are multimodal (Inkling handles both image **and** audio; `glm-5.3-flash` renders images only with `process_multimodal=False`).
+Hand-coded renderers ship for `qwen3`, `qwen3-vl`, `qwen3.5`, `qwen3.6`, `qwen3.8`, `gemma4`, `glm-5`, `glm-5.1`, `glm-5.3`, `glm-5.3-flash`, `glm-4.5`, `minimax-m2`, `deepseek-v3`, `deepseek-r1`, `deepseek-v4` (V4 Flash 0731), `kimi-k2`, `kimi-k2.5` / `kimi-k2.6`, `laguna-xs.2`, `laguna-xs-2.1`, `laguna-s-2.1`, `laguna-m.1`, `nemotron-3`, `nemotron-3-ultra`, `nemotron-3.5`, `llama-3`, `gpt-oss`, `muse-glimmer`, `hy3`, `inkling` / `inkling-small`, and `prime-qwen3`. Anything else falls back to `DefaultRenderer`, a generic `apply_chat_template` wrapper. `qwen3-vl`, `qwen3.5`, `qwen3.6`, `qwen3.8`, `gemma4`, `glm-5.3-flash`, `muse-glimmer`, `kimi-k2.5` / `kimi-k2.6`, and the Inkling checkpoints are multimodal (Inkling handles both image **and** audio; `glm-5.3-flash` and `muse-glimmer` render images only with `process_multimodal=False`).
 
 ## API
 

{renderers-0.1.13.dev5 → renderers-0.1.13.dev8}/README.md

@@ -94,7 +94,7 @@ next_prompt_ids = r.bridge_to_next_turn(
 )
 ```
 
-Hand-coded renderers ship for `qwen3`, `qwen3-vl`, `qwen3.5`, `qwen3.6`, `qwen3.8`, `gemma4`, `glm-5`, `glm-5.1`, `glm-5.3`, `glm-5.3-flash`, `glm-4.5`, `minimax-m2`, `deepseek-v3`, `deepseek-r1`, `deepseek-v4` (V4 Flash 0731), `kimi-k2`, `kimi-k2.5` / `kimi-k2.6`, `laguna-xs.2`, `laguna-xs-2.1`, `laguna-s-2.1`, `laguna-m.1`, `nemotron-3`, `nemotron-3-ultra`, `nemotron-3.5`, `llama-3`, `gpt-oss`, `hy3`, `inkling` / `inkling-small`, and `prime-qwen3`. Anything else falls back to `DefaultRenderer`, a generic `apply_chat_template` wrapper. `qwen3-vl`, `qwen3.5`, `qwen3.6`, `qwen3.8`, `gemma4`, `glm-5.3-flash`, `kimi-k2.5` / `kimi-k2.6`, and the Inkling checkpoints are multimodal (Inkling handles both image **and** audio; `glm-5.3-flash` renders images only with `process_multimodal=False`).
+Hand-coded renderers ship for `qwen3`, `qwen3-vl`, `qwen3.5`, `qwen3.6`, `qwen3.8`, `gemma4`, `glm-5`, `glm-5.1`, `glm-5.3`, `glm-5.3-flash`, `glm-4.5`, `minimax-m2`, `deepseek-v3`, `deepseek-r1`, `deepseek-v4` (V4 Flash 0731), `kimi-k2`, `kimi-k2.5` / `kimi-k2.6`, `laguna-xs.2`, `laguna-xs-2.1`, `laguna-s-2.1`, `laguna-m.1`, `nemotron-3`, `nemotron-3-ultra`, `nemotron-3.5`, `llama-3`, `gpt-oss`, `muse-glimmer`, `hy3`, `inkling` / `inkling-small`, and `prime-qwen3`. Anything else falls back to `DefaultRenderer`, a generic `apply_chat_template` wrapper. `qwen3-vl`, `qwen3.5`, `qwen3.6`, `qwen3.8`, `gemma4`, `glm-5.3-flash`, `muse-glimmer`, `kimi-k2.5` / `kimi-k2.6`, and the Inkling checkpoints are multimodal (Inkling handles both image **and** audio; `glm-5.3-flash` and `muse-glimmer` render images only with `process_multimodal=False`).
 
 ## API
 

{renderers-0.1.13.dev5 → renderers-0.1.13.dev8}/docs/renderer-config.md

@@ -37,6 +37,7 @@ product in `tests/test_parity.py`.
 | GLM-5.3 / 5.3 Flash | `GLM53RendererConfig` / `GLM53FlashRendererConfig` | `clear_thinking`, `reasoning_effort` | - |
 | GLM-4.5 | `GLM45RendererConfig` | `enable_thinking` | - |
 | gpt-oss | `GptOssRendererConfig` | `reasoning_effort`, `conversation_start_date` | `use_system_prompt`, `knowledge_cutoff`, `model_identity`, `auto_drop_analysis` |
+| Muse Glimmer | `MuseGlimmerRendererConfig` | `reasoning_strength`, `knowledge_cutoff`, `current_date` | - |
 | Hy3 | `Hy3RendererConfig` | `reasoning_effort`, `preserved_thinking`, `is_training`, `raw_last_assistant`, `fallback_strategy` | - |
 | Kimi K2 | `KimiK2RendererConfig` | - | `enable_thinking` |
 | Kimi K2.5 / 2.6 | `KimiK25RendererConfig` | `thinking` | `image_cache_max` |
@@ -144,6 +145,7 @@ the knobs its template actually exposes:
 | GLM-5.3 / 5.3 Flash | `clear_thinking=False -> all`, else `tool_cycle` |
 | GLM-4.5 | `enable_thinking=False -> all`, else `tool_cycle` |
 | gpt-oss | `auto_drop_analysis=False -> all`, else `tool_cycle` |
+| Muse Glimmer | `all` (the template keeps every assistant's reasoning) |
 | Hy3 | `preserved_thinking=True -> all`, else `tool_cycle` |
 | Kimi K2.5 / 2.6 | `thinking=False -> all`, else `tool_cycle` |
 | Nemotron-3 / 3.5 | `truncate_history_thinking=False -> all`; else `enable_thinking=False -> all`; else `tool_cycle` |

{renderers-0.1.13.dev5 → renderers-0.1.13.dev8}/renderers/__init__.py

@@ -68,6 +68,7 @@ from renderers.configs import (
     LagunaXS21RendererConfig,
     Llama3RendererConfig,
     MiniMaxM2RendererConfig,
+    MuseGlimmerRendererConfig,
     Nemotron3RendererConfig,
     Nemotron3UltraRendererConfig,
     Nemotron35RendererConfig,
@@ -111,6 +112,7 @@ _LAZY_RENDERERS: dict[str, str] = {
     "LagunaXS2Renderer": "renderers.laguna_xs2",
     "Llama3Renderer": "renderers.llama_3",
     "MiniMaxM2Renderer": "renderers.minimax_m2",
+    "MuseGlimmerRenderer": "renderers.muse_glimmer",
     "Nemotron35Renderer": "renderers.nemotron3",
     "Nemotron3Renderer": "renderers.nemotron3",
     "Nemotron3UltraRenderer": "renderers.nemotron3",
@@ -193,6 +195,8 @@ __all__ = [
     "MiniMaxM2RendererConfig",
     "MultiModalData",
     "MultimodalRenderer",
+    "MuseGlimmerRenderer",
+    "MuseGlimmerRendererConfig",
     "Nemotron35Renderer",
     "Nemotron35RendererConfig",
     "Nemotron3Renderer",

{renderers-0.1.13.dev5 → renderers-0.1.13.dev8}/renderers/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.1.13.dev5'
-__version_tuple__ = version_tuple = (0, 1, 13, 'dev5')
+__version__ = version = '0.1.13.dev8'
+__version_tuple__ = version_tuple = (0, 1, 13, 'dev8')
 
 __commit_id__ = commit_id = None

{renderers-0.1.13.dev5 → renderers-0.1.13.dev8}/renderers/base.py

@@ -1063,6 +1063,8 @@ MODEL_RENDERER_MAP: dict[str, str] = {
     # GPT-OSS.
     "openai/gpt-oss-20b": "gpt-oss",
     "openai/gpt-oss-120b": "gpt-oss",
+    # Muse Glimmer (ATEM channel format; images as one <|patch|> each).
+    "meta-models/Muse-Glimmer-30B": "muse-glimmer",
     # Thinking Machines Inkling checkpoints share byte-identical tokenizer,
     # chat-template, and processor assets (vision + audio; transformers >= 5.14).
     "thinkingmachines/Inkling": "inkling",
@@ -1375,6 +1377,7 @@ def _populate_registry():
     )
     from renderers.llama_3 import Llama3Renderer
     from renderers.minimax_m2 import MiniMaxM2Renderer
+    from renderers.muse_glimmer import MuseGlimmerRenderer
     from renderers.nemotron3 import (
         Nemotron3Renderer,
         Nemotron3UltraRenderer,
@@ -1419,6 +1422,7 @@ def _populate_registry():
             "nemotron-3-ultra": Nemotron3UltraRenderer,
             "nemotron-3.5": Nemotron35Renderer,
             "gpt-oss": GptOssRenderer,
+            "muse-glimmer": MuseGlimmerRenderer,
         }
     )
 

{renderers-0.1.13.dev5 → renderers-0.1.13.dev8}/renderers/client.py

@@ -261,9 +261,11 @@ async def generate(
     Returns a dict with: request_id, prompt_ids, renderer_prompt_ids,
     mm_placeholders, completion_ids, completion_logprobs, content,
     reasoning_content, tool_calls, finish_reason, routed_experts,
-    multi_modal_data, prompt_attribution. ``renderer_prompt_ids`` is the
-    unexpanded logical prompt when ``process_multimodal=False`` and ``None``
-    otherwise.
+    weight_versions, multi_modal_data, prompt_attribution.
+    ``renderer_prompt_ids`` is the unexpanded logical prompt when
+    ``process_multimodal=False`` and ``None`` otherwise. ``weight_versions``
+    is vLLM's (>= 0.32) list of ``{version, start, end}`` spans over
+    completion-token indexes, or ``None`` when the engine labels none.
 
     ``prompt_attribution`` is the renderer's :class:`RenderedTokens` for
     the prompt — either the one this call computed via
@@ -431,6 +433,7 @@ async def generate(
         "reasoning_complete": parsed.reasoning_complete,
         "routed_experts": routed_experts,
         "sampling_mask": sampling_mask,
+        "weight_versions": choice.get("weight_versions"),
         # The mm sidecar consumed on the request side, surfaced back so
         # callers can persist it on the trajectory step for downstream
         # multi-turn bridging and training-sample construction.

{renderers-0.1.13.dev5 → renderers-0.1.13.dev8}/renderers/configs.py

@@ -647,6 +647,34 @@ class GptOssRendererConfig(BaseRendererConfig):
         return self
 
 
+class MuseGlimmerRendererConfig(BaseRendererConfig):
+    """Muse Glimmer (ATEM) renderer config.
+
+    The chat template never drops historical reasoning, so bridges keep all
+    reasoning by default (``thinking_retention`` resolves to ``"all"``).
+    """
+
+    name: Literal["muse-glimmer"] = "muse-glimmer"
+    _template_fields = frozenset(
+        {"reasoning_strength", "knowledge_cutoff", "current_date"}
+    )
+
+    reasoning_strength: Literal["low", "medium", "high", "xhigh"] = "high"
+    """``Reasoning strength: <value>.`` line in the system preamble. Mirrors
+    the template's ``reasoning_strength`` kwarg (template default ``high``).
+    A system message that already states a reasoning strength (or effort)
+    keeps its own line instead."""
+
+    knowledge_cutoff: str = "2026-01-04"
+    """Knowledge-cutoff date in the default preamble (used only when the
+    conversation has no system message). Mirrors ``knowledge_cutoff``."""
+
+    current_date: str | None = None
+    """``Current date`` in the default preamble, as ``YYYY-MM-DD``. ``None``
+    uses today's date (the template's ``strftime_now`` fallback), fixed when
+    the renderer is constructed. Pin it for reproducible training renders."""
+
+
 class KimiK2RendererConfig(BaseRendererConfig):
     """Kimi K2 renderer config.
 
@@ -1051,6 +1079,7 @@ RendererConfig = Annotated[
         GLM53FlashRendererConfig,
         GLM45RendererConfig,
         GptOssRendererConfig,
+        MuseGlimmerRendererConfig,
         Hy3RendererConfig,
         InklingRendererConfig,
         KimiK2RendererConfig,
@@ -1101,6 +1130,7 @@ _CONFIG_BY_NAME: dict[str, type[BaseRendererConfig]] = {
     "glm-5.3-flash": GLM53FlashRendererConfig,
     "glm-4.5": GLM45RendererConfig,
     "gpt-oss": GptOssRendererConfig,
+    "muse-glimmer": MuseGlimmerRendererConfig,
     "hy3": Hy3RendererConfig,
     "inkling": InklingRendererConfig,
     "kimi-k2": KimiK2RendererConfig,
@@ -1168,6 +1198,7 @@ __all__ = [
     "LagunaXS21RendererConfig",
     "Llama3RendererConfig",
     "MiniMaxM2RendererConfig",
+    "MuseGlimmerRendererConfig",
     "Nemotron35RendererConfig",
     "Nemotron3RendererConfig",
     "Nemotron3UltraRendererConfig",