ablate-llm 0.3.0__tar.gz → 0.3.1__tar.gz

{ablate_llm-0.3.0/src/ablate_llm.egg-info → ablate_llm-0.3.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ablate-llm
-Version: 0.3.0
+Version: 0.3.1
 Summary: Directional ablation (abliteration) toolkit for open-source language models: extract, ablate, and optimize the removal of refusal / behavioral directions in the residual stream.
 Author: AI Anytime
 License: Proprietary
@@ -50,7 +50,7 @@ Dynamic: license-file
 
 <h1 align="center">Ablate</h1>
 
-<p align="center"><b>Directional ablation (abliteration) toolkit for automatic censorship removal of open-source language models.</b></p>
+<p align="center"><b>Directional ablation (abliteration) toolkit for automatic censorship removal of open-source language models (SLMs/LLMs).</b></p>
 
 <p align="center">
   <a href="https://pypi.org/project/ablate-llm/"><img alt="PyPI" src="https://img.shields.io/pypi/v/ablate-llm.svg?color=7c5cff"></a>
@@ -59,19 +59,26 @@ Dynamic: license-file
 </p>
 
 <p align="center">
+  <a href="https://aianytime.github.io/ablate/">🌐 Website</a> &nbsp;•&nbsp;
   <a href="https://pypi.org/project/ablate-llm/">📦 PyPI</a> &nbsp;•&nbsp;
   <a href="https://huggingface.co/ai-anytime/qwen-1.5b-abliterated">🤗 Model</a> &nbsp;•&nbsp;
   <a href="https://youtu.be/1-1bM3Ynql4">▶️ Video walkthrough</a> &nbsp;•&nbsp;
+  <a href="https://dubai.aitinkerers.org/talks/rsvp_rDbfcoxeQ5Y">🎤 Dubai AI Tinkerers talk</a> &nbsp;•&nbsp;
   <a href="#citation">📚 Cite</a>
 </p>
 
 <p align="center"><i>Created with 💜 by <b>AI Anytime</b></i></p>
 
 <p align="center">
-  <a href="https://youtu.be/1-1bM3Ynql4">
-    <img alt="Watch the video walkthrough" src="https://img.youtube.com/vi/1-1bM3Ynql4/maxresdefault.jpg" width="560">
+  <a href="https://youtu.be/_pLKCRgzu6A">
+    <img
+      src="https://img.youtube.com/vi/_pLKCRgzu6A/maxresdefault.jpg"
+      alt="Ablate — Full YouTube Walkthrough"
+      width="800"
+    />
   </a>
-  <br><i>▶️ Watch the full walkthrough on YouTube</i>
+  <br />
+  <i>▶️ Watch the full walkthrough on YouTube</i>
 </p>
 
 ---
@@ -93,6 +100,23 @@ iterate on locally on a laptop, or on a free Colab T4.
 > direction, not requests to fulfill. Use it on models and in contexts you're
 > authorized to.
 
+> **🎤 Presented at Dubai AI Tinkerers (October 2026).** Ablate was demoed live,
+> no slides, using the browser UI shown below:
+> [dubai.aitinkerers.org/talks/rsvp_rDbfcoxeQ5Y](https://dubai.aitinkerers.org/talks/rsvp_rDbfcoxeQ5Y)
+
+## Architecture
+
+<p align="center">
+  <a href="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-architecture.png"><img alt="Ablate architecture: three interfaces, one pipeline, and the three places it touches the model" src="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-architecture.png" width="900"></a>
+</p>
+
+Three ways in (browser UI, CLI, Python API), one `Ablator` pipeline, no training.
+The coloured arrows are the only three places `ablate` touches the model: it
+**reads** the residual stream on two piles of prompts, **subtracts** the refusal
+direction at every layer's output while the model runs (a hook), or **rewrites**
+the weights for good (bake). A one-page printable version is in
+[`media/ablate-architecture.pdf`](media/ablate-architecture.pdf).
+
 ## 🤗 Released model
 
 A worked example of the full pipeline is published at
@@ -285,6 +309,16 @@ ablate ui --model Qwen/Qwen2.5-0.5B-Instruct  # preload a model
 ablate ui --share                           # let others on the same Wi-Fi explore
 ```
 
+<p align="center">
+  <a href="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-playground.png"><img alt="Playground: the original model refuses a gardening question, the edited model answers it" src="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-playground.png" width="900"></a>
+  <br><i>Playground: same prompt, same weights. The original refuses, the edited model answers.</i>
+</p>
+
+<p align="center">
+  <a href="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-find-direction.png"><img alt="Find direction: per-layer separation chart and every prompt plotted along the refusal direction" src="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-find-direction.png" width="900"></a>
+  <br><i>Find direction: which layer holds the refusal signal, and every prompt plotted along it.</i>
+</p>
+
 A local, dependency-free web app that walks through the whole pipeline:
 
 1. **Model** — search the HuggingFace Hub and load any chat model.
@@ -423,4 +457,4 @@ mechanistic-interpretability work of Arditi et al. and the open-source
 
 ## License
 
-Proprietary — all rights reserved. This code is published for viewing and evaluation only; no use, copying, modification, redistribution, commercial use, or use as AI/ML training data without written permission. See [LICENSE](LICENSE). Commercial licensing: aianytime07@gmail.com · sonu@aianytime.net.
+Proprietary — all rights reserved. This code is published for viewing and evaluation only; no use, copying, modification, redistribution, commercial use, or use as AI/ML training data without written permission. See [LICENSE](LICENSE). Commercial licensing: sonu@aianytime.net.

{ablate_llm-0.3.0 → ablate_llm-0.3.1}/README.md

@@ -8,7 +8,7 @@
 
 <h1 align="center">Ablate</h1>
 
-<p align="center"><b>Directional ablation (abliteration) toolkit for automatic censorship removal of open-source language models.</b></p>
+<p align="center"><b>Directional ablation (abliteration) toolkit for automatic censorship removal of open-source language models (SLMs/LLMs).</b></p>
 
 <p align="center">
   <a href="https://pypi.org/project/ablate-llm/"><img alt="PyPI" src="https://img.shields.io/pypi/v/ablate-llm.svg?color=7c5cff"></a>
@@ -17,19 +17,26 @@
 </p>
 
 <p align="center">
+  <a href="https://aianytime.github.io/ablate/">🌐 Website</a> &nbsp;•&nbsp;
   <a href="https://pypi.org/project/ablate-llm/">📦 PyPI</a> &nbsp;•&nbsp;
   <a href="https://huggingface.co/ai-anytime/qwen-1.5b-abliterated">🤗 Model</a> &nbsp;•&nbsp;
   <a href="https://youtu.be/1-1bM3Ynql4">▶️ Video walkthrough</a> &nbsp;•&nbsp;
+  <a href="https://dubai.aitinkerers.org/talks/rsvp_rDbfcoxeQ5Y">🎤 Dubai AI Tinkerers talk</a> &nbsp;•&nbsp;
   <a href="#citation">📚 Cite</a>
 </p>
 
 <p align="center"><i>Created with 💜 by <b>AI Anytime</b></i></p>
 
 <p align="center">
-  <a href="https://youtu.be/1-1bM3Ynql4">
-    <img alt="Watch the video walkthrough" src="https://img.youtube.com/vi/1-1bM3Ynql4/maxresdefault.jpg" width="560">
+  <a href="https://youtu.be/_pLKCRgzu6A">
+    <img
+      src="https://img.youtube.com/vi/_pLKCRgzu6A/maxresdefault.jpg"
+      alt="Ablate — Full YouTube Walkthrough"
+      width="800"
+    />
   </a>
-  <br><i>▶️ Watch the full walkthrough on YouTube</i>
+  <br />
+  <i>▶️ Watch the full walkthrough on YouTube</i>
 </p>
 
 ---
@@ -51,6 +58,23 @@ iterate on locally on a laptop, or on a free Colab T4.
 > direction, not requests to fulfill. Use it on models and in contexts you're
 > authorized to.
 
+> **🎤 Presented at Dubai AI Tinkerers (October 2026).** Ablate was demoed live,
+> no slides, using the browser UI shown below:
+> [dubai.aitinkerers.org/talks/rsvp_rDbfcoxeQ5Y](https://dubai.aitinkerers.org/talks/rsvp_rDbfcoxeQ5Y)
+
+## Architecture
+
+<p align="center">
+  <a href="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-architecture.png"><img alt="Ablate architecture: three interfaces, one pipeline, and the three places it touches the model" src="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-architecture.png" width="900"></a>
+</p>
+
+Three ways in (browser UI, CLI, Python API), one `Ablator` pipeline, no training.
+The coloured arrows are the only three places `ablate` touches the model: it
+**reads** the residual stream on two piles of prompts, **subtracts** the refusal
+direction at every layer's output while the model runs (a hook), or **rewrites**
+the weights for good (bake). A one-page printable version is in
+[`media/ablate-architecture.pdf`](media/ablate-architecture.pdf).
+
 ## 🤗 Released model
 
 A worked example of the full pipeline is published at
@@ -243,6 +267,16 @@ ablate ui --model Qwen/Qwen2.5-0.5B-Instruct  # preload a model
 ablate ui --share                           # let others on the same Wi-Fi explore
 ```
 
+<p align="center">
+  <a href="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-playground.png"><img alt="Playground: the original model refuses a gardening question, the edited model answers it" src="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-playground.png" width="900"></a>
+  <br><i>Playground: same prompt, same weights. The original refuses, the edited model answers.</i>
+</p>
+
+<p align="center">
+  <a href="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-find-direction.png"><img alt="Find direction: per-layer separation chart and every prompt plotted along the refusal direction" src="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-find-direction.png" width="900"></a>
+  <br><i>Find direction: which layer holds the refusal signal, and every prompt plotted along it.</i>
+</p>
+
 A local, dependency-free web app that walks through the whole pipeline:
 
 1. **Model** — search the HuggingFace Hub and load any chat model.
@@ -381,4 +415,4 @@ mechanistic-interpretability work of Arditi et al. and the open-source
 
 ## License
 
-Proprietary — all rights reserved. This code is published for viewing and evaluation only; no use, copying, modification, redistribution, commercial use, or use as AI/ML training data without written permission. See [LICENSE](LICENSE). Commercial licensing: aianytime07@gmail.com · sonu@aianytime.net.
+Proprietary — all rights reserved. This code is published for viewing and evaluation only; no use, copying, modification, redistribution, commercial use, or use as AI/ML training data without written permission. See [LICENSE](LICENSE). Commercial licensing: sonu@aianytime.net.

{ablate_llm-0.3.0 → ablate_llm-0.3.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "ablate-llm"
-version = "0.3.0"
+version = "0.3.1"
 description = "Directional ablation (abliteration) toolkit for open-source language models: extract, ablate, and optimize the removal of refusal / behavioral directions in the residual stream."
 readme = "README.md"
 requires-python = ">=3.9"

{ablate_llm-0.3.0 → ablate_llm-0.3.1}/src/ablate/__init__.py

@@ -23,7 +23,7 @@ from .pipeline import Ablator, run
 from .weights import bake_direction, bake_subspace
 from .hub import build_model_card, push_to_hub
 
-__version__ = "0.3.0"
+__version__ = "0.3.1"
 
 __all__ = [
     "Ablator",

{ablate_llm-0.3.0 → ablate_llm-0.3.1}/src/ablate/hub.py

@@ -19,10 +19,14 @@ def build_model_card(
     method: str,
     config: dict,
     metrics: Optional[dict] = None,
-    license: str = "apache-2.0",
+    license: Optional[str] = "apache-2.0",
     extra_notes: str = "",
+    license_name: Optional[str] = None,
+    license_link: Optional[str] = None,
 ) -> str:
     """Return the full README.md (model card) string, YAML frontmatter included."""
+    license = license or "other"
+    built_with_llama = _is_llama(base_model, license)
     tags = [
         "abliteration",
         "directional-ablation",
@@ -31,10 +35,14 @@ def build_model_card(
         "safety-research",
         "ablate",
     ]
+    if built_with_llama:
+        tags.append("llama")
     fm = [
         "---",
         f"base_model: {base_model}",
         f"license: {license}",
+        *([f"license_name: {license_name}"] if license_name else []),
+        *([f"license_link: {license_link}"] if license_link else []),
         "library_name: transformers",
         "pipeline_tag: text-generation",
         "tags:",
@@ -55,8 +63,16 @@ def build_model_card(
             "judged attack-success rate on a harmful benchmark.\n\n"
         )
 
+    llama_md = (
+        "**Built with Llama.** This model is a derivative of a Meta Llama model and is "
+        "distributed under the base model's Llama Community License. A copy of the "
+        "license and acceptable use policy from the base model is included in this repo.\n\n"
+        if built_with_llama else ""
+    )
+
     body = f"""# {repo_id.split('/')[-1]}
 
+{llama_md}
 This is a **directionally-ablated ("abliterated")** version of
 [`{base_model}`](https://huggingface.co/{base_model}), produced with
 [**ablate**](https://github.com/AIAnytime/ablate) — an activation-engineering toolkit for
@@ -115,6 +131,46 @@ If you use this model or `ablate`, please cite Arditi et al. (2024),
     return "\n".join(fm) + body
 
 
+def _is_llama(base_model: str, license: Optional[str]) -> bool:
+    return base_model.lower().startswith("meta-llama/") or str(license or "").lower().startswith("llama")
+
+
+def base_model_license(base_model: str, token: Optional[str] = None) -> dict:
+    """Read ``license`` / ``license_name`` / ``license_link`` from the base model's card.
+
+    A derivative must carry its base model's license, so the card copies it instead
+    of assuming one. Returns ``{}`` when the base card cannot be read.
+    """
+    try:
+        from huggingface_hub import HfApi
+        cd = HfApi(token=token).model_info(base_model).card_data
+    except Exception:
+        return {}
+    if cd is None:
+        return {}
+    out = {k: cd.get(k) for k in ("license", "license_name", "license_link") if cd.get(k)}
+    if isinstance(out.get("license"), list):
+        out["license"] = out["license"][0]
+    return out
+
+
+def _copy_license_files(api, base_model: str, dest: str) -> list:
+    """Copy LICENSE / NOTICE / USE_POLICY files from the base repo into ``dest``."""
+    copied = []
+    try:
+        from huggingface_hub import hf_hub_download
+        names = [f for f in api.list_repo_files(base_model)
+                 if "/" not in f and f.upper().startswith(("LICENSE", "NOTICE", "USE_POLICY"))]
+        for f in names:
+            src = hf_hub_download(base_model, f, token=api.token)
+            with open(src, "rb") as a, open(os.path.join(dest, f), "wb") as b:
+                b.write(a.read())
+            copied.append(f)
+    except Exception:
+        pass
+    return copied
+
+
 def push_to_hub(
     model,
     tokenizer,
@@ -125,12 +181,13 @@ def push_to_hub(
     config: Optional[dict] = None,
     metrics: Optional[dict] = None,
     private: bool = True,
-    license: str = "apache-2.0",
+    license: Optional[str] = None,
     extra_notes: str = "",
     commit_message: str = "Upload abliterated model via ablate",
 ) -> str:
     """Save, card, and upload a model to ``repo_id``. Returns the repo URL.
 
+    ``license`` defaults to the base model's own license, read from its Hub card.
     ``token`` falls back to the ``HF_TOKEN`` / ``HUGGINGFACE_TOKEN`` env vars or a
     cached ``huggingface-cli login``.
     """
@@ -143,14 +200,19 @@ def push_to_hub(
     api = HfApi(token=token)
     api.create_repo(repo_id=repo_id, private=private, exist_ok=True, repo_type="model")
 
+    lic = {"license": license} if license else base_model_license(base_model, token)
     card = build_model_card(
         repo_id=repo_id, base_model=base_model, method=method,
-        config=config or {}, metrics=metrics, license=license, extra_notes=extra_notes,
+        config=config or {}, metrics=metrics, license=lic.get("license"),
+        license_name=lic.get("license_name"), license_link=lic.get("license_link"),
+        extra_notes=extra_notes,
     )
 
     with tempfile.TemporaryDirectory() as d:
         model.save_pretrained(d)
         tokenizer.save_pretrained(d)
+        if base_model != "unknown":
+            _copy_license_files(api, base_model, d)
         with open(os.path.join(d, "README.md"), "w", encoding="utf-8") as f:
             f.write(card)
         with open(os.path.join(d, "ablate_meta.json"), "w") as f:

{ablate_llm-0.3.0 → ablate_llm-0.3.1}/src/ablate/pipeline.py

@@ -234,7 +234,7 @@ class Ablator:
         token: Optional[str] = None,
         private: bool = True,
         metrics: Optional[dict] = None,
-        license: str = "apache-2.0",
+        license: Optional[str] = None,
         bake_if_needed: bool = True,
     ) -> str:
         """Bake (if needed) and upload the abliterated model + a generated model

{ablate_llm-0.3.0 → ablate_llm-0.3.1}/src/ablate/ui/server.py

@@ -35,7 +35,7 @@ from ..config import AblationConfig, SubspaceConfig
 from ..evaluate import _coherence, is_refusal, mean_kl_divergence
 from ..extract import collect_activations
 from ..hooks import AblationHooks
-from ..hub import build_model_card
+from ..hub import base_model_license, build_model_card
 from ..model import LM
 from ..optimize import optimize, optimize_subspace
 from ..pipeline import Ablator
@@ -599,7 +599,10 @@ def _card_metrics() -> Optional[dict]:
 def card_preview(body: dict) -> dict:
     basis, conf = _resolve(body.get("config") or {})
     repo = (body.get("repo_id") or "you/model-abliterated").strip()
-    return {"card": build_model_card(repo, S.lm.name, _method_of(conf), conf.to_dict(), _card_metrics())}
+    lic = base_model_license(S.lm.name, S.token)
+    return {"card": build_model_card(repo, S.lm.name, _method_of(conf), conf.to_dict(), _card_metrics(),
+                                     license=lic.get("license"), license_name=lic.get("license_name"),
+                                     license_link=lic.get("license_link"))}
 
 
 def _bake_then(body: dict, emit: Emit, ship: Callable[[], str]) -> None:

{ablate_llm-0.3.0 → ablate_llm-0.3.1/src/ablate_llm.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ablate-llm
-Version: 0.3.0
+Version: 0.3.1
 Summary: Directional ablation (abliteration) toolkit for open-source language models: extract, ablate, and optimize the removal of refusal / behavioral directions in the residual stream.
 Author: AI Anytime
 License: Proprietary
@@ -50,7 +50,7 @@ Dynamic: license-file
 
 <h1 align="center">Ablate</h1>
 
-<p align="center"><b>Directional ablation (abliteration) toolkit for automatic censorship removal of open-source language models.</b></p>
+<p align="center"><b>Directional ablation (abliteration) toolkit for automatic censorship removal of open-source language models (SLMs/LLMs).</b></p>
 
 <p align="center">
   <a href="https://pypi.org/project/ablate-llm/"><img alt="PyPI" src="https://img.shields.io/pypi/v/ablate-llm.svg?color=7c5cff"></a>
@@ -59,19 +59,26 @@ Dynamic: license-file
 </p>
 
 <p align="center">
+  <a href="https://aianytime.github.io/ablate/">🌐 Website</a> &nbsp;•&nbsp;
   <a href="https://pypi.org/project/ablate-llm/">📦 PyPI</a> &nbsp;•&nbsp;
   <a href="https://huggingface.co/ai-anytime/qwen-1.5b-abliterated">🤗 Model</a> &nbsp;•&nbsp;
   <a href="https://youtu.be/1-1bM3Ynql4">▶️ Video walkthrough</a> &nbsp;•&nbsp;
+  <a href="https://dubai.aitinkerers.org/talks/rsvp_rDbfcoxeQ5Y">🎤 Dubai AI Tinkerers talk</a> &nbsp;•&nbsp;
   <a href="#citation">📚 Cite</a>
 </p>
 
 <p align="center"><i>Created with 💜 by <b>AI Anytime</b></i></p>
 
 <p align="center">
-  <a href="https://youtu.be/1-1bM3Ynql4">
-    <img alt="Watch the video walkthrough" src="https://img.youtube.com/vi/1-1bM3Ynql4/maxresdefault.jpg" width="560">
+  <a href="https://youtu.be/_pLKCRgzu6A">
+    <img
+      src="https://img.youtube.com/vi/_pLKCRgzu6A/maxresdefault.jpg"
+      alt="Ablate — Full YouTube Walkthrough"
+      width="800"
+    />
   </a>
-  <br><i>▶️ Watch the full walkthrough on YouTube</i>
+  <br />
+  <i>▶️ Watch the full walkthrough on YouTube</i>
 </p>
 
 ---
@@ -93,6 +100,23 @@ iterate on locally on a laptop, or on a free Colab T4.
 > direction, not requests to fulfill. Use it on models and in contexts you're
 > authorized to.
 
+> **🎤 Presented at Dubai AI Tinkerers (October 2026).** Ablate was demoed live,
+> no slides, using the browser UI shown below:
+> [dubai.aitinkerers.org/talks/rsvp_rDbfcoxeQ5Y](https://dubai.aitinkerers.org/talks/rsvp_rDbfcoxeQ5Y)
+
+## Architecture
+
+<p align="center">
+  <a href="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-architecture.png"><img alt="Ablate architecture: three interfaces, one pipeline, and the three places it touches the model" src="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-architecture.png" width="900"></a>
+</p>
+
+Three ways in (browser UI, CLI, Python API), one `Ablator` pipeline, no training.
+The coloured arrows are the only three places `ablate` touches the model: it
+**reads** the residual stream on two piles of prompts, **subtracts** the refusal
+direction at every layer's output while the model runs (a hook), or **rewrites**
+the weights for good (bake). A one-page printable version is in
+[`media/ablate-architecture.pdf`](media/ablate-architecture.pdf).
+
 ## 🤗 Released model
 
 A worked example of the full pipeline is published at
@@ -285,6 +309,16 @@ ablate ui --model Qwen/Qwen2.5-0.5B-Instruct  # preload a model
 ablate ui --share                           # let others on the same Wi-Fi explore
 ```
 
+<p align="center">
+  <a href="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-playground.png"><img alt="Playground: the original model refuses a gardening question, the edited model answers it" src="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-playground.png" width="900"></a>
+  <br><i>Playground: same prompt, same weights. The original refuses, the edited model answers.</i>
+</p>
+
+<p align="center">
+  <a href="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-find-direction.png"><img alt="Find direction: per-layer separation chart and every prompt plotted along the refusal direction" src="https://raw.githubusercontent.com/AIAnytime/ablate/main/media/ui-find-direction.png" width="900"></a>
+  <br><i>Find direction: which layer holds the refusal signal, and every prompt plotted along it.</i>
+</p>
+
 A local, dependency-free web app that walks through the whole pipeline:
 
 1. **Model** — search the HuggingFace Hub and load any chat model.
@@ -423,4 +457,4 @@ mechanistic-interpretability work of Arditi et al. and the open-source
 
 ## License
 
-Proprietary — all rights reserved. This code is published for viewing and evaluation only; no use, copying, modification, redistribution, commercial use, or use as AI/ML training data without written permission. See [LICENSE](LICENSE). Commercial licensing: aianytime07@gmail.com · sonu@aianytime.net.
+Proprietary — all rights reserved. This code is published for viewing and evaluation only; no use, copying, modification, redistribution, commercial use, or use as AI/ML training data without written permission. See [LICENSE](LICENSE). Commercial licensing: sonu@aianytime.net.