diversify-text 0.2.0__tar.gz → 0.2.1__tar.gz

{diversify_text-0.2.0 → diversify_text-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: diversify-text
-Version: 0.2.0
+Version: 0.2.1
 Summary: Generate stylistic paraphrases of texts using local transformer models.
 Project-URL: Homepage, https://github.com/AnnaWegmann/diversify_text
 Project-URL: Documentation, https://annawegmann.github.io/diversify_text/
@@ -46,6 +46,7 @@ pip install diversify-text
 - [Usage](#usage)
   - [Single text](#single-text)
   - [Control number of paraphrases](#control-number-of-paraphrases)
+  - [Prompting method](#prompting-method)
   - [Caching](#caching)
   - [Using the class directly](#using-the-class-directly)
   - [List of texts](#list-of-texts)
@@ -92,6 +93,30 @@ results = diversify("Some text.", n=3)
 [{"original": "Some text.", "paraphrases": ["...", "...", "..."]}]
 ```
 
+### Prompting method
+
+Use the prompting method to generate paraphrases via a causal language model (default: [SmolLM3-3B](https://huggingface.co/HuggingFaceTB/SmolLM3-3B)):
+
+```python
+results = diversify("The experiment was conducted in a controlled lab setting.", methods=["prompting"])
+```
+
+Select specific prompt styles:
+
+```python
+results = diversify(
+    "The experiment was conducted in a controlled lab setting.",
+    methods=["prompting"],
+    method_kwargs={
+        "prompting": {
+            "prompt_keys": ["simple_kew", "complex_kew", "caps_reif"]
+        }
+    },
+)
+```
+
+Available prompt keys: `wikipedia_paraphrase`, `simple_kew`, `complex_kew`, `formal_reif`, `simple_reif`, `passive_reif`, `caps_reif`, `lowcaps_reif`, `text_emojis_reif`, `less_common_verbs_reif`, `humanize_llm-as-coauthor_original`, and all `finephrase_*` templates. See the [full prompt reference](https://annawegmann.github.io/diversify_text/prompts.html) for details.
+
 ### Caching
 
 The `diversify()` function automatically caches loaded models between calls.

{diversify_text-0.2.0 → diversify_text-0.2.1}/README.md

@@ -13,6 +13,7 @@ pip install diversify-text
 - [Usage](#usage)
   - [Single text](#single-text)
   - [Control number of paraphrases](#control-number-of-paraphrases)
+  - [Prompting method](#prompting-method)
   - [Caching](#caching)
   - [Using the class directly](#using-the-class-directly)
   - [List of texts](#list-of-texts)
@@ -59,6 +60,30 @@ results = diversify("Some text.", n=3)
 [{"original": "Some text.", "paraphrases": ["...", "...", "..."]}]
 ```
 
+### Prompting method
+
+Use the prompting method to generate paraphrases via a causal language model (default: [SmolLM3-3B](https://huggingface.co/HuggingFaceTB/SmolLM3-3B)):
+
+```python
+results = diversify("The experiment was conducted in a controlled lab setting.", methods=["prompting"])
+```
+
+Select specific prompt styles:
+
+```python
+results = diversify(
+    "The experiment was conducted in a controlled lab setting.",
+    methods=["prompting"],
+    method_kwargs={
+        "prompting": {
+            "prompt_keys": ["simple_kew", "complex_kew", "caps_reif"]
+        }
+    },
+)
+```
+
+Available prompt keys: `wikipedia_paraphrase`, `simple_kew`, `complex_kew`, `formal_reif`, `simple_reif`, `passive_reif`, `caps_reif`, `lowcaps_reif`, `text_emojis_reif`, `less_common_verbs_reif`, `humanize_llm-as-coauthor_original`, and all `finephrase_*` templates. See the [full prompt reference](https://annawegmann.github.io/diversify_text/prompts.html) for details.
+
 ### Caching
 
 The `diversify()` function automatically caches loaded models between calls.

{diversify_text-0.2.0 → diversify_text-0.2.1}/diversify_text/core.py

@@ -248,16 +248,40 @@ class Diversifier:
     ) -> int:
         """Infer *n* from per-method kwargs when only one method is used.
 
-        When a single method is active and the caller provided method-
-        specific keys (``prompt_keys`` for prompting, ``styles`` for
-        tinystyler), returns the length of those keys so each is used
-        exactly once.  Otherwise returns :attr:`_DEFAULT_N`.
+        When a single method is active (currently outof tinystyler and prompting)
+        and the caller provided method-specific keys, infers the number of
+        paraphrases from number of keys.  Otherwise returns :attr:`_DEFAULT_N`.
+
+        For the prompting method, the inference depends on what is provided:
+            * ``prompt_keys`` only → ``len(prompt_keys)`` (one per template).
+            * ``styles`` only → ``len(styles)`` (style transfer, one per style).
+            * Both → each style-dependent template (in
+            :data:`EXAMPLE_BASED_PROMPT_BANK` or :data:`NAME_BASED_PROMPT_BANK`)
+            contributes ``len(styles)``, each zero-shot template contributes 1.
         """
         if len(self._methods) == 1 and method_kwargs:
             method = self._methods[0]
             kw = method_kwargs.get(method.name, {})
-            if method.name == "prompting" and "prompt_keys" in kw:
-                return len(kw["prompt_keys"])
+            if method.name == "prompting":
+                from diversify_text.method.prompting.prompts import STYLE_DEP_PROMPTS
+                from diversify_text.styles import DEFAULT_STYLES
+
+                prompt_keys = kw.get("prompt_keys")
+                styles = kw.get("styles")
+                # When styles are not provided but style-dependent prompts
+                # are selected, default to DEFAULT_STYLES.
+                if not styles and prompt_keys and any(k in STYLE_DEP_PROMPTS for k in prompt_keys):
+                    styles = DEFAULT_STYLES
+                if prompt_keys:
+                    n = 0
+                    for key in prompt_keys:
+                        if key in STYLE_DEP_PROMPTS and styles:
+                            n += len(styles)
+                        else:
+                            n += 1
+                    return n
+                if styles:
+                    return len(styles)
             if method.name == "tinystyler" and "styles" in kw:
                 return len(kw["styles"])
         return self._DEFAULT_N

{diversify_text-0.2.0 → diversify_text-0.2.1}/diversify_text/method/prompting/method.py

@@ -14,6 +14,7 @@ from diversify_text.method.prompting.prompts import (
     PLACEHOLDER_STYLE_NAME,
     PLACEHOLDER_TEXT,
     PROMPT_BANK,
+    STYLE_DEP_PROMPTS,
 )
 from diversify_text.styles import resolve_style_sets
 
@@ -75,7 +76,7 @@ class PromptingMethod(DiversificationMethod):
     def _resolve_prompts(
         prompt_bank: dict[str, str] | None = None,
         prompt_keys: list[str] | None = None,
-        style_example_keys: list[str] | None = None,
+        styles: list[str] | None = None,
         custom_style_bank: dict[str, list[str]] | None = None,
     ) -> list[tuple[str, str]]:
         """Resolve prompt configuration into an ordered list of (key, template) pairs.
@@ -93,7 +94,7 @@ class PromptingMethod(DiversificationMethod):
             zero-shot + few-shot bank).
         prompt_keys : list[str] or None
             Select only these keys from the bank.  Order is preserved.
-        style_example_keys : list[str] or None
+        styles : list[str] or None
             Names of style sets for few-shot examples.  If provided
             without *prompt_keys*, the method automatically selects
             the ``"style_transfer"`` prompt template.  When combined
@@ -101,7 +102,7 @@ class PromptingMethod(DiversificationMethod):
             the ``[STYLE EXAMPLES]`` placeholder.
         custom_style_bank : dict or None
             Custom style bank — same trigger behavior as
-            *style_example_keys*.
+            *styles*.
 
         Returns
         -------
@@ -113,13 +114,13 @@ class PromptingMethod(DiversificationMethod):
         Raises
         ------
         ValueError
-            If *prompt_keys* contains unknown keys, or if *style_example_keys*
+            If *prompt_keys* contains unknown keys, or if *styles*
             / *custom_style_bank* are provided but the selected
             templates do not contain a ``[STYLE EXAMPLES]``
             placeholder.
         """
         bank = prompt_bank if prompt_bank is not None else PROMPT_BANK
-        has_styles = style_example_keys is not None or custom_style_bank is not None
+        has_styles = styles is not None or custom_style_bank is not None
 
         # --- Select templates (four mutually exclusive cases) ---
 
@@ -153,12 +154,12 @@ class PromptingMethod(DiversificationMethod):
             for _k, t in templates
         ):
             raise ValueError(
-                "style_example_keys or custom_style_bank were provided, but the "
+                "styles or custom_style_bank were provided, but the "
                 "selected prompt template(s) do not contain the "
                 f"{PLACEHOLDER_STYLE_EXAMPLES} or {PLACEHOLDER_STYLE_NAME} "
                 f"placeholder. Use a style-aware template "
                 f"(e.g. prompt_keys=['style_transfer'] or prompt_keys=['reif']) "
-                f"or remove style_example_keys. See "
+                f"or remove styles. See "
                 f"https://annawegmann.github.io/diversify_text/prompts.html"
             )
 
@@ -244,12 +245,41 @@ class PromptingMethod(DiversificationMethod):
         Returns a dict mapping style names to example sentences, or
         an empty dict when no style kwargs are provided.
         """
-        style_keys = kwargs.get("style_example_keys")
+        style_keys = kwargs.get("styles")
         custom_bank = kwargs.get("custom_style_bank")
         if style_keys is not None or custom_bank is not None:
             return resolve_style_sets(custom_bank, style_keys)
         return {}
 
+    @staticmethod
+    def _build_schedule(
+        prompt_templates: list[tuple[str, str]],
+        fs_style_examples: dict[str, list[str]],
+    ) -> list[tuple[str, str, int | None]]:
+        """Build a generation schedule from templates and style examples.
+
+        Style-dependent templates (in :data:`EXAMPLE_BASED_PROMPT_BANK` or
+        :data:`NAME_BASED_PROMPT_BANK`) expand to one entry per style;
+        zero-shot templates get a single entry.
+
+        The caller iterates the schedule with modulo to fill ``n`` slots,
+        so the schedule represents one full "natural" cycle.
+
+        Each entry is ``(key, template, style_idx)`` where *style_idx* is
+        ``None`` for zero-shot templates.
+        """
+        n_styles = len(fs_style_examples) if fs_style_examples else 0
+
+
+        schedule: list[tuple[str, str, int | None]] = []
+        for key, tmpl in prompt_templates:
+            if n_styles and key in STYLE_DEP_PROMPTS:
+                for style_idx in range(n_styles):
+                    schedule.append((key, tmpl, style_idx))
+            else:
+                schedule.append((key, tmpl, None))
+        return schedule
+
     def _fill_template(
         self,
         template: str,
@@ -346,7 +376,7 @@ class PromptingMethod(DiversificationMethod):
             Sampling parameters. ``None`` uses defaults.
         **kwargs
             Extra options forwarded from ``Diversifier``, including
-            ``prompt_keys``, ``prompt_bank``, ``style_example_keys``,
+            ``prompt_keys``, ``prompt_bank``, ``styles``,
             ``custom_style_bank``, and ``n_style_examples``.
         """
         model = self._ensure_model()
@@ -356,7 +386,7 @@ class PromptingMethod(DiversificationMethod):
         prompt_templates = self._resolve_prompts(
             prompt_bank=kwargs.get("prompt_bank"),
             prompt_keys=kwargs.get("prompt_keys"),
-            style_example_keys=kwargs.get("style_example_keys"),
+            styles=kwargs.get("styles"),
             custom_style_bank=kwargs.get("custom_style_bank"),
         )
         all_max_new_tokens = self._compute_max_new_tokens(
@@ -369,24 +399,32 @@ class PromptingMethod(DiversificationMethod):
         )
 
         fs_style_examples = self._resolve_few_shot_examples(**kwargs)
+        # Default to DEFAULT_STYLES when style-dependent prompts are
+        # selected but no explicit styles were provided.
+        if not fs_style_examples:
+    
+            if any(k in STYLE_DEP_PROMPTS for k, _ in prompt_templates):
+                from diversify_text.styles import DEFAULT_STYLES
+                fs_style_examples = resolve_style_sets(None, DEFAULT_STYLES)
         if fs_style_examples:
             logger.info("Style sets: %s", ", ".join(fs_style_examples.keys()))
 
         n_ex = kwargs.get("n_style_examples", _DEFAULT_N_STYLE_EXAMPLES)
 
-        # Build prompts in the same order as all_max_new_tokens.
+        schedule = self._build_schedule(prompt_templates, fs_style_examples)
+
         # TODO: accept texts as an Iterable (not just list) to support
         #       streaming from large files without materialising everything
         #       in memory.
         all_prompts: list[str] = []
         for i in range(n):
-            _key, template = prompt_templates[i % len(prompt_templates)]
+            _key, template, style_idx = schedule[i % len(schedule)]
             for t in texts:
                 all_prompts.append(
                     self._fill_template(
                         template=template,
                         text=t,
-                        style_idx=i,
+                        style_idx=style_idx,
                         fs_style_examples=fs_style_examples,
                         n_style_examples=n_ex,
                     )

{diversify_text-0.2.0 → diversify_text-0.2.1}/diversify_text/method/prompting/prompts.py

@@ -249,6 +249,9 @@ PLACEHOLDER_STYLE_NAME = "[STYLE NAME]"
 
 PROMPT_BANK: dict[str, str] = {**ZS_PROMPT_BANK, **EXAMPLE_BASED_PROMPT_BANK, **NAME_BASED_PROMPT_BANK}
 
+#: Prompt keys whose templates depend on style examples or style names.
+STYLE_DEP_PROMPTS: set[str] = {*EXAMPLE_BASED_PROMPT_BANK, *NAME_BASED_PROMPT_BANK}
+
 DEFAULT_PROMPTS: list[str] = [
     HUMANIZE_LLM_AS_COAUTHOR_ORIGINAL,
     COMPLEX_KEW,

{diversify_text-0.2.0 → diversify_text-0.2.1}/diversify_text/method/tinystyler/method.py

@@ -118,7 +118,7 @@ class TinyStylerMethod(DiversificationMethod):
             styles_arg,
         )
         # When explicit style keys are given, they determine the count.
-        effective_n = len(styles_arg) if styles_arg is not None else n
+        effective_n = n
         if effective_n > len(style_bank):
             logger.warning(
                 "n=%d exceeds the number of style bank entries (%d). "

{diversify_text-0.2.0 → diversify_text-0.2.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "diversify-text"
-version = "0.2.0"
+version = "0.2.1"
 description = "Generate stylistic paraphrases of texts using local transformer models."
 readme = "README.md"
 requires-python = ">=3.10"