Upload 130 files

app.py

@@ -278,6 +278,12 @@ METHODS = {
     "nuclear (maximum force combo)": "nuclear",
 }
 
+# ── Community Hub push ────────────────────────────────────────────────
+# Shared org + token so users can auto-push without their own HF_TOKEN.
+# Set OBLITERATUS_HUB_TOKEN as a Space secret with write access to the org.
+_HUB_COMMUNITY_ORG = os.environ.get("OBLITERATUS_HUB_ORG", "OBLITERATUS-community")
+_HUB_COMMUNITY_TOKEN = os.environ.get("OBLITERATUS_HUB_TOKEN")
+
 # Import preset configs for Advanced Settings defaults
 from obliteratus.abliterate import METHODS as _PRESET_CONFIGS  # noqa: E402
 from obliteratus.prompts import (  # noqa: E402
@@ -383,10 +389,10 @@ def _validate_hub_repo(hub_repo: str) -> str:
             "Invalid repo format — use `username/model-name` "
             "(letters, numbers, hyphens, dots only)"
         )
-    if not os.environ.get("HF_TOKEN"):
+    if not os.environ.get("HF_TOKEN") and not _HUB_COMMUNITY_TOKEN:
         warnings.append(
-            "HF_TOKEN not set — push to Hub will fail. "
-            "Set it via: `export HF_TOKEN=hf_...`"
+            "No Hub token available — push will fail. "
+            "Set HF_TOKEN or OBLITERATUS_HUB_TOKEN."
         )
     if warnings:
         return "**Warning:** " + " | ".join(warnings)
@@ -1600,7 +1606,11 @@ def obliterate(model_choice: str, method_choice: str,
         )
         return
 
-    # Early validation: Hub repo format + HF_TOKEN
+    # Early validation: Hub repo format + token availability
+    # Resolve which token to use: user's own HF_TOKEN, or the shared community token.
+    _user_token = os.environ.get("HF_TOKEN")
+    _hub_token = _user_token or _HUB_COMMUNITY_TOKEN
+    _hub_org = None if _user_token else _HUB_COMMUNITY_ORG  # community org only when using shared token
     if push_to_hub:
         if push_to_hub != "auto" and not re.match(r'^[a-zA-Z0-9_-]+/[a-zA-Z0-9_.-]+$', push_to_hub):
             yield (
@@ -1608,10 +1618,10 @@ def obliterate(model_choice: str, method_choice: str,
                 "", gr.update(), gr.update(), gr.update(), gr.update(),
             )
             return
-        if not os.environ.get("HF_TOKEN"):
+        if not _hub_token:
             yield (
-                "**Error:** HF_TOKEN not set. Push to Hub requires a write token. "
-                "Set it via `export HF_TOKEN=hf_...` or in your Space secrets.",
+                "**Error:** No Hub token available. Set HF_TOKEN or OBLITERATUS_HUB_TOKEN "
+                "as an environment variable or Space secret.",
                 "", gr.update(), gr.update(), gr.update(), gr.update(),
             )
             return
@@ -1690,6 +1700,8 @@ def obliterate(model_choice: str, method_choice: str,
                     device="auto",
                     dtype="float16",
                     push_to_hub=push_to_hub,
+                    hub_token=_hub_token,
+                    hub_community_org=_hub_org,
                     quantization=quantization,
                     trust_remote_code=is_preset,
                     harmful_prompts=harmful_all[:n],
@@ -1708,6 +1720,8 @@ def obliterate(model_choice: str, method_choice: str,
                     dtype="float16",
                     method=method,
                     push_to_hub=push_to_hub,
+                    hub_token=_hub_token,
+                    hub_community_org=_hub_org,
                     quantization=quantization,
                     trust_remote_code=is_preset,
                     harmful_prompts=harmful_all[:n],
@@ -1762,7 +1776,8 @@ def obliterate(model_choice: str, method_choice: str,
     log_lines.append(f"Prompt volume: {vol_label} pairs")
     if push_to_hub:
         if push_to_hub == "auto":
-            log_lines.append("Push to Hub: auto ({you}/{model}-OBLITERATED)")
+            _ns = _hub_org or "{you}"
+            log_lines.append(f"Push to Hub: auto ({_ns}/{{model}}-OBLITERATED)")
         else:
             log_lines.append(f"Push to Hub: {push_to_hub}")
     if quantization:
@@ -3512,8 +3527,8 @@ with gr.Blocks(theme=THEME, css=CSS, js=_JS, title="OBLITERATUS", fill_height=Tr
                 hub_auto_push = gr.Checkbox(
                     label="Auto-push to Hub",
                     value=False,
-                    info="Auto-names your model {you}/{model}-OBLITERATED and pushes to HF Hub. "
-                         "Requires HF_TOKEN env var with write access.",
+                    info=f"Pushes your model to {_HUB_COMMUNITY_ORG}/{{model}}-OBLITERATED on HF Hub. "
+                         "No token needed — works out of the box!",
                 )
                 hub_repo = gr.Textbox(
                     label="Push to Hub (optional override)",
@@ -4237,11 +4252,13 @@ The winner gets auto-pushed to HuggingFace Hub.
                 def on_round(rnd):
                     pass  # logged via on_log
 
+                dataset_key = get_source_key_from_label(dataset) if dataset else "builtin"
+
                 runner = TourneyRunner(
                     model_name=model_id,
                     hub_org=hub_org_val,
                     hub_repo=hub_repo_val,
-                    dataset_key=dataset or "builtin",
+                    dataset_key=dataset_key,
                     quantization=quant,
                     on_log=on_log,
                     on_round=on_round,

obliteratus/abliterate.py

@@ -504,21 +504,22 @@ class StageResult:
     details: dict[str, Any] = field(default_factory=dict)
 
 
-def auto_hub_repo_id(model_name: str, *, api=None) -> str:
-    """Generate a Hub repo ID like ``{hf_user}/{short_model}-OBLITERATED``.
+def auto_hub_repo_id(model_name: str, *, api=None, org: str | None = None) -> str:
+    """Generate a Hub repo ID like ``{namespace}/{short_model}-OBLITERATED``.
 
-    Resolves the authenticated HF username via the API, and extracts a
-    short model name (e.g. ``meta-llama/Llama-3-8B`` → ``Llama-3-8B``).
+    If *org* is given, uses that as the namespace (e.g. a shared community org).
+    Otherwise resolves the authenticated HF username via the API.
     """
     import re
 
-    if api is None:
-        from huggingface_hub import HfApi
-        api = HfApi()
-
-    # Resolve HF username from token
-    user_info = api.whoami()
-    username = user_info.get("name") or user_info.get("user", "unknown")
+    if org:
+        namespace = org
+    else:
+        if api is None:
+            from huggingface_hub import HfApi
+            api = HfApi()
+        user_info = api.whoami()
+        namespace = user_info.get("name") or user_info.get("user", "unknown")
 
     # Extract short model name (part after '/')
     short = model_name.split("/")[-1] if "/" in model_name else model_name
@@ -526,7 +527,7 @@ def auto_hub_repo_id(model_name: str, *, api=None) -> str:
     short = re.sub(r"[^a-zA-Z0-9\-.]", "-", short)
     short = re.sub(r"-+", "-", short).strip("-")
 
-    return f"{username}/{short}-OBLITERATED"
+    return f"{namespace}/{short}-OBLITERATED"
 
 
 # ── Main pipeline ───────────────────────────────────────────────────────
@@ -559,6 +560,8 @@ class AbliterationPipeline:
         trust_remote_code: bool = False,
         method: str = "advanced",
         push_to_hub: str | None = None,
+        hub_token: str | None = None,
+        hub_community_org: str | None = None,
         n_directions: int | None = None,
         norm_preserve: bool | None = None,
         regularization: float | None = None,
@@ -618,6 +621,8 @@ class AbliterationPipeline:
         self.trust_remote_code = trust_remote_code
         self.large_model_mode = large_model_mode
         self.push_to_hub = push_to_hub
+        self.hub_token = hub_token
+        self.hub_community_org = hub_community_org
         self.harmful_prompts = list(harmful_prompts) if harmful_prompts is not None else list(HARMFUL_PROMPTS)
         self.harmless_prompts = list(harmless_prompts) if harmless_prompts is not None else list(HARMLESS_PROMPTS)
         if not self.harmful_prompts:
@@ -5948,11 +5953,13 @@ class AbliterationPipeline:
         if self.push_to_hub:
             from huggingface_hub import HfApi
 
-            api = HfApi()
+            api = HfApi(token=self.hub_token) if self.hub_token else HfApi()
 
-            # Resolve "auto" → {hf_user}/{short_model}-OBLITERATED
+            # Resolve "auto" → {namespace}/{short_model}-OBLITERATED
             if self.push_to_hub == "auto":
-                repo_id = auto_hub_repo_id(self.model_name, api=api)
+                repo_id = auto_hub_repo_id(
+                    self.model_name, api=api, org=self.hub_community_org,
+                )
                 self.log(f"Auto-named Hub repo: {repo_id}")
             else:
                 repo_id = self.push_to_hub

obliteratus/informed_pipeline.py

@@ -181,6 +181,8 @@ class InformedAbliterationPipeline(AbliterationPipeline):
         on_log: Callable[[str], None] | None = None,
         # Base pipeline kwargs forwarded to AbliterationPipeline
         push_to_hub: str | None = None,
+        hub_token: str | None = None,
+        hub_community_org: str | None = None,
         quantization: str | None = None,
         # Analysis configuration
         run_cone_analysis: bool = True,
@@ -212,6 +214,8 @@ class InformedAbliterationPipeline(AbliterationPipeline):
             on_stage=on_stage,
             on_log=on_log,
             push_to_hub=push_to_hub,
+            hub_token=hub_token,
+            hub_community_org=hub_community_org,
             quantization=quantization,
             # Set informed defaults
             norm_preserve=True,