renderers 0.1.8.dev55__tar.gz → 0.1.8.dev57__tar.gz

{renderers-0.1.8.dev55 → renderers-0.1.8.dev57}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: renderers
-Version: 0.1.8.dev55
+Version: 0.1.8.dev57
 Summary: Chat template renderers — deterministic message-to-token conversion for LLM training
 License-Expression: Apache-2.0
 License-File: LICENSE
@@ -56,7 +56,7 @@ next_prompt_ids = r.bridge_to_next_turn(
 )
 ```
 
-Hand-coded renderers ship for `qwen3`, `qwen3-vl`, `qwen3.5`, `qwen3.6`, `glm-5`, `glm-5.1`, `glm-4.5`, `minimax-m2`, `deepseek-v3`, `deepseek-r1`, `kimi-k2`, `kimi-k2.5` / `kimi-k2.6`, `nemotron-3`, `nemotron-3-ultra`, `llama-3`, and `gpt-oss`. Anything else falls back to `DefaultRenderer`, a generic `apply_chat_template` wrapper.
+Hand-coded renderers ship for `qwen3`, `qwen3-vl`, `qwen3.5`, `qwen3.6`, `glm-5`, `glm-5.1`, `glm-4.5`, `minimax-m2`, `deepseek-v3`, `deepseek-r1`, `kimi-k2`, `kimi-k2.5` / `kimi-k2.6`, `nemotron-3`, `nemotron-3-ultra`, `llama-3`, `gpt-oss`, `hy3`, and `prime-qwen3`. Anything else falls back to `DefaultRenderer`, a generic `apply_chat_template` wrapper.
 
 ## API
 

{renderers-0.1.8.dev55 → renderers-0.1.8.dev57}/README.md

@@ -40,7 +40,7 @@ next_prompt_ids = r.bridge_to_next_turn(
 )
 ```
 
-Hand-coded renderers ship for `qwen3`, `qwen3-vl`, `qwen3.5`, `qwen3.6`, `glm-5`, `glm-5.1`, `glm-4.5`, `minimax-m2`, `deepseek-v3`, `deepseek-r1`, `kimi-k2`, `kimi-k2.5` / `kimi-k2.6`, `nemotron-3`, `nemotron-3-ultra`, `llama-3`, and `gpt-oss`. Anything else falls back to `DefaultRenderer`, a generic `apply_chat_template` wrapper.
+Hand-coded renderers ship for `qwen3`, `qwen3-vl`, `qwen3.5`, `qwen3.6`, `glm-5`, `glm-5.1`, `glm-4.5`, `minimax-m2`, `deepseek-v3`, `deepseek-r1`, `kimi-k2`, `kimi-k2.5` / `kimi-k2.6`, `nemotron-3`, `nemotron-3-ultra`, `llama-3`, `gpt-oss`, `hy3`, and `prime-qwen3`. Anything else falls back to `DefaultRenderer`, a generic `apply_chat_template` wrapper.
 
 ## API
 

{renderers-0.1.8.dev55 → renderers-0.1.8.dev57}/docs/renderer-config.md

@@ -26,12 +26,14 @@ chat-template kwargs. Those fields are covered by parity tests against
 | Renderer | Config class | Template fields | Renderer-only fields |
 | --- | --- | --- | --- |
 | Qwen3 | `Qwen3RendererConfig` | `enable_thinking` | - |
+| PrimeIntellect Qwen3 | `PrimeQwen3RendererConfig` | - | - |
 | Qwen3.5 | `Qwen35RendererConfig` | `enable_thinking`, `add_vision_id` | `image_cache_max` |
 | Qwen3.6 | `Qwen36RendererConfig` | `enable_thinking`, `add_vision_id`, `preserve_thinking` | `image_cache_max` |
 | Qwen3-VL | `Qwen3VLRendererConfig` | `add_vision_id` | `image_cache_max` |
 | GLM-5 / 5.1 | `GLM5RendererConfig` / `GLM51RendererConfig` | `enable_thinking`, `clear_thinking` | - |
 | GLM-4.5 | `GLM45RendererConfig` | `enable_thinking` | - |
 | gpt-oss | `GptOssRendererConfig` | `reasoning_effort`, `conversation_start_date` | `use_system_prompt`, `knowledge_cutoff`, `model_identity`, `auto_drop_analysis` |
+| 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` |
 | Laguna XS.2 | `LagunaXS2RendererConfig` | `enable_thinking`, `render_assistant_messages_raw` | - |
@@ -129,11 +131,13 @@ the knobs its template actually exposes:
 | GLM-5 / 5.1 | `clear_thinking=False -> all`; else `enable_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` |
+| Hy3 | `preserved_thinking=True -> all`, else `tool_cycle` |
 | Kimi K2.5 / 2.6 | `thinking=False -> all`, else `tool_cycle` |
 | Nemotron-3 | `truncate_history_thinking=False -> all`; else `enable_thinking=False -> all`; else `tool_cycle` |
 | DeepSeek R1 | `template` |
 | MiniMax M2 | `tool_cycle` |
 | DeepSeek V3, Qwen3-VL, Kimi K2, Laguna XS.2 / XS-2.1, Llama 3 | `all` |
+| PrimeIntellect Qwen3 | `all` |
 
 Config construction raises when an explicit template knob directly contradicts
 an explicit generic bridge policy. For example:

{renderers-0.1.8.dev55 → renderers-0.1.8.dev57}/renderers/__init__.py

@@ -51,6 +51,7 @@ from renderers.configs import (
     GLM51RendererConfig,
     GLM5RendererConfig,
     GptOssRendererConfig,
+    Hy3RendererConfig,
     KimiK25RendererConfig,
     KimiK2RendererConfig,
     LagunaXS2RendererConfig,
@@ -59,6 +60,7 @@ from renderers.configs import (
     MiniMaxM2RendererConfig,
     Nemotron3RendererConfig,
     Nemotron3UltraRendererConfig,
+    PrimeQwen3RendererConfig,
     Qwen35RendererConfig,
     Qwen36RendererConfig,
     Qwen3RendererConfig,
@@ -85,6 +87,7 @@ _LAZY_RENDERERS: dict[str, str] = {
     "GLM51Renderer": "renderers.glm5",
     "GLM5Renderer": "renderers.glm5",
     "GptOssRenderer": "renderers.gpt_oss",
+    "Hy3Renderer": "renderers.hy3",
     "KimiK25Renderer": "renderers.kimi_k25",
     "KimiK2Renderer": "renderers.kimi_k2",
     "LagunaXS21Renderer": "renderers.laguna_xs2",
@@ -93,6 +96,7 @@ _LAZY_RENDERERS: dict[str, str] = {
     "MiniMaxM2Renderer": "renderers.minimax_m2",
     "Nemotron3Renderer": "renderers.nemotron3",
     "Nemotron3UltraRenderer": "renderers.nemotron3",
+    "PrimeQwen3Renderer": "renderers.prime_qwen3",
     "Qwen35Renderer": "renderers.qwen35",
     "Qwen36Renderer": "renderers.qwen36",
     "Qwen3Renderer": "renderers.qwen3",
@@ -134,6 +138,8 @@ __all__ = [
     "GLM5RendererConfig",
     "GptOssRenderer",
     "GptOssRendererConfig",
+    "Hy3Renderer",
+    "Hy3RendererConfig",
     "ImagePart",
     "KimiK25Renderer",
     "KimiK25RendererConfig",
@@ -159,6 +165,8 @@ __all__ = [
     "ParsedResponse",
     "ParsedToolCall",
     "PlaceholderRange",
+    "PrimeQwen3Renderer",
+    "PrimeQwen3RendererConfig",
     "Qwen35Renderer",
     "Qwen35RendererConfig",
     "Qwen36Renderer",

{renderers-0.1.8.dev55 → renderers-0.1.8.dev57}/renderers/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.1.8.dev55'
-__version_tuple__ = version_tuple = (0, 1, 8, 'dev55')
+__version__ = version = '0.1.8.dev57'
+__version_tuple__ = version_tuple = (0, 1, 8, 'dev57')
 
 __commit_id__ = commit_id = None

{renderers-0.1.8.dev55 → renderers-0.1.8.dev57}/renderers/base.py

@@ -999,6 +999,10 @@ MODEL_RENDERER_MAP: dict[str, str] = {
     "Qwen/Qwen3-30B-A3B-Instruct-2507": "qwen3",
     "Qwen/Qwen3-30B-A3B-Thinking-2507": "qwen3",
     "Qwen/Qwen3-235B-A22B": "qwen3",
+    # PrimeIntellect Qwen3 — both sizes share the same Qwen3-Coder-style
+    # template with XML tool definitions and calls.
+    "PrimeIntellect/Qwen3-0.6B": "prime-qwen3",
+    "PrimeIntellect/Qwen3-1.7B": "prime-qwen3",
     # Qwen3.5. All seven sizes share the same renderer. The 4B / 9B /
     # 35B-A3B / 122B-A10B / 397B-A17B chat template defaults
     # ``enable_thinking=true`` (open ``<think>\n`` at the gen prompt);
@@ -1067,6 +1071,12 @@ MODEL_RENDERER_MAP: dict[str, str] = {
     # GPT-OSS.
     "openai/gpt-oss-20b": "gpt-oss",
     "openai/gpt-oss-120b": "gpt-oss",
+    # Tencent Hunyuan Hy3 (295B-A21B MoE). The FP8 checkpoint shares the same
+    # tokenizer and chat template. Hy3-preview is deliberately unmapped: it
+    # ships an older, incompatible template (un-suffixed special tokens,
+    # ``interleaved_thinking`` instead of ``preserved_thinking``).
+    "tencent/Hy3": "hy3",
+    "tencent/Hy3-FP8": "hy3",
 }
 
 
@@ -1302,12 +1312,14 @@ def _populate_registry():
     from renderers.glm5 import GLM5Renderer, GLM51Renderer
     from renderers.glm45 import GLM45Renderer
     from renderers.gpt_oss import GptOssRenderer
+    from renderers.hy3 import Hy3Renderer
     from renderers.kimi_k2 import KimiK2Renderer
     from renderers.kimi_k25 import KimiK25Renderer
     from renderers.laguna_xs2 import LagunaXS2Renderer, LagunaXS21Renderer
     from renderers.llama_3 import Llama3Renderer
     from renderers.minimax_m2 import MiniMaxM2Renderer
     from renderers.nemotron3 import Nemotron3Renderer, Nemotron3UltraRenderer
+    from renderers.prime_qwen3 import PrimeQwen3Renderer
     from renderers.qwen3 import Qwen3Renderer
     from renderers.qwen3_vl import Qwen3VLRenderer
     from renderers.qwen35 import Qwen35Renderer
@@ -1317,6 +1329,7 @@ def _populate_registry():
         {
             "default": DefaultRenderer,
             "qwen3": Qwen3Renderer,
+            "prime-qwen3": PrimeQwen3Renderer,
             "qwen3-vl": Qwen3VLRenderer,
             "qwen3.5": Qwen35Renderer,
             "qwen3.6": Qwen36Renderer,
@@ -1326,6 +1339,7 @@ def _populate_registry():
             "minimax-m2": MiniMaxM2Renderer,
             "deepseek-v3": DeepSeekV3Renderer,
             "deepseek-r1": DeepSeekR1Renderer,
+            "hy3": Hy3Renderer,
             "kimi-k2": KimiK2Renderer,
             "kimi-k2.5": KimiK25Renderer,
             "laguna-xs.2": LagunaXS2Renderer,

{renderers-0.1.8.dev55 → renderers-0.1.8.dev57}/renderers/configs.py

@@ -181,6 +181,12 @@ class Qwen3RendererConfig(BaseRendererConfig):
     ``enable_thinking`` kwarg."""
 
 
+class PrimeQwen3RendererConfig(BaseRendererConfig):
+    """PrimeIntellect Qwen3 renderer config."""
+
+    name: Literal["prime-qwen3"] = "prime-qwen3"
+
+
 class Qwen35RendererConfig(BaseRendererConfig):
     """Qwen3.5 renderer config."""
 
@@ -310,6 +316,70 @@ class GLM45RendererConfig(BaseRendererConfig):
     the chat template's ``enable_thinking`` kwarg."""
 
 
+class Hy3RendererConfig(BaseRendererConfig):
+    """Tencent Hunyuan Hy3 renderer config.
+
+    Hy3 reasons via a ``reasoning_effort`` gate rather than a boolean
+    ``enable_thinking``. ``"no_think"`` (the template default) prefills an
+    empty ``<think></think>`` at the generation prompt so the model answers
+    directly; ``"low"`` / ``"high"`` prefill only the ``<think>`` opener so
+    the model streams reasoning up to a ``</think>`` it emits itself.
+
+    ``preserved_thinking`` mirrors the template kwarg of the same name:
+    ``True`` keeps ``<think>{reasoning}</think>`` on every historical
+    assistant turn; ``False`` collapses past-cycle reasoning to
+    ``<think></think>``, keeping it only on the in-flight turn (after the
+    last user query). ``None`` (default) follows the template's own
+    default — ``True`` when ``tools`` are supplied at render time, ``False``
+    otherwise. Bridge policy tracks the same resolution: ``"all"`` whenever
+    ``preserved_thinking`` resolves to ``True`` for the tools at hand,
+    ``"tool_cycle"`` otherwise.
+    """
+
+    name: Literal["hy3"] = "hy3"
+
+    reasoning_effort: Literal["no_think", "low", "high"] = "no_think"
+    """Reasoning gate. Mirrors the chat template's ``reasoning_effort`` kwarg.
+    ``"no_think"`` prefills ``<think></think>`` at the generation prompt;
+    ``"low"`` / ``"high"`` prefill just ``<think>``."""
+
+    preserved_thinking: bool | None = None
+    """Keep historical assistant reasoning. Mirrors the template's
+    ``preserved_thinking`` kwarg. ``None`` defers to the template default
+    (``True`` with tools, ``False`` without). Bridge policy is ``"all"``
+    whenever this resolves to ``True`` for the tools at hand."""
+
+    is_training: bool = False
+    """Mirrors the template's ``is_training`` kwarg. ``True`` renders SFT
+    targets: reasoning is kept on every assistant turn (regardless of
+    ``preserved_thinking`` / position) and the final assistant is terminated
+    with ``<｜hy_eos｜>``. Leave ``False`` for inference-faithful renders;
+    the training loss mask is normally derived via ``build_training_sample``
+    rather than this flag."""
+
+    raw_last_assistant: bool = False
+    """Mirrors the template's ``raw_last_assistant`` kwarg. When ``True`` a
+    trailing non-tool assistant message is emitted as raw visible content —
+    no ``<think>`` wrap, no ``<｜hy_eos｜>`` — for prefill / continuation."""
+
+    fallback_strategy: Literal["reasoning_toolcall_retry"] | None = None
+    """Mirrors the template's ``fallback_strategy`` kwarg. The sole active
+    value, ``"reasoning_toolcall_retry"``, forces ``reasoning_effort="high"``
+    and suppresses the generation prompt (``add_generation_prompt=False``)."""
+
+    @model_validator(mode="after")
+    def _check_thinking_retention(self):
+        if self.preserved_thinking is not None and self.thinking_retention is not None:
+            implied = "all" if self.preserved_thinking else "tool_cycle"
+            if self.thinking_retention != implied:
+                raise ValueError(
+                    f"preserved_thinking={self.preserved_thinking!r} implies "
+                    f"thinking_retention={implied!r}, which conflicts with "
+                    f"explicit thinking_retention={self.thinking_retention!r}."
+                )
+        return self
+
+
 class GptOssRendererConfig(BaseRendererConfig):
     """OpenAI gpt-oss (harmony) renderer config.
 
@@ -585,6 +655,7 @@ RendererConfig = Annotated[
         AutoRendererConfig,
         DefaultRendererConfig,
         Qwen3RendererConfig,
+        PrimeQwen3RendererConfig,
         Qwen35RendererConfig,
         Qwen36RendererConfig,
         Qwen3VLRendererConfig,
@@ -592,6 +663,7 @@ RendererConfig = Annotated[
         GLM51RendererConfig,
         GLM45RendererConfig,
         GptOssRendererConfig,
+        Hy3RendererConfig,
         KimiK2RendererConfig,
         KimiK25RendererConfig,
         LagunaXS2RendererConfig,
@@ -623,6 +695,7 @@ _CONFIG_BY_NAME: dict[str, type[BaseRendererConfig]] = {
     "auto": AutoRendererConfig,
     "default": DefaultRendererConfig,
     "qwen3": Qwen3RendererConfig,
+    "prime-qwen3": PrimeQwen3RendererConfig,
     "qwen3.5": Qwen35RendererConfig,
     "qwen3.6": Qwen36RendererConfig,
     "qwen3-vl": Qwen3VLRendererConfig,
@@ -630,6 +703,7 @@ _CONFIG_BY_NAME: dict[str, type[BaseRendererConfig]] = {
     "glm-5.1": GLM51RendererConfig,
     "glm-4.5": GLM45RendererConfig,
     "gpt-oss": GptOssRendererConfig,
+    "hy3": Hy3RendererConfig,
     "kimi-k2": KimiK2RendererConfig,
     "kimi-k2.5": KimiK25RendererConfig,
     "laguna-xs.2": LagunaXS2RendererConfig,
@@ -677,6 +751,7 @@ __all__ = [
     "GLM51RendererConfig",
     "GLM5RendererConfig",
     "GptOssRendererConfig",
+    "Hy3RendererConfig",
     "KimiK25RendererConfig",
     "KimiK2RendererConfig",
     "LagunaXS2RendererConfig",
@@ -685,6 +760,7 @@ __all__ = [
     "MiniMaxM2RendererConfig",
     "Nemotron3RendererConfig",
     "Nemotron3UltraRendererConfig",
+    "PrimeQwen3RendererConfig",
     "Qwen35RendererConfig",
     "Qwen36RendererConfig",
     "Qwen3RendererConfig",