Add Hugging Face support and enhance model resolution logic

- Introduced support for Hugging Face API integration, allowing fallback model resolution when Civitai fails.
- Updated configuration to include `HF_TOKEN` and `HF_TOKEN_PATH` for authentication.
- Enhanced model capture logic to differentiate between Civitai and Hugging Face sources.
- Improved error handling for model downloads, providing clearer messages for authentication issues.
- Updated documentation to reflect new environment variables and usage instructions for Hugging Face integration.
- Added tests to validate the new fallback mechanism and ensure robust model resolution.

src/gpu_rent/provision.py

@@ -234,16 +234,10 @@ def seed_swarmui_api_keys(cfg: Config, host: str, log: Log) -> None:
     Swarm stores them in Users.ldb GenericData — needed for Model Downloader in the UI.
     Call after SwarmUI HTTP is up (after wait_backend / verify).
     """
-    import os
-
     keys: dict[str, str] = {}
     if cfg.civitai_api_token:
         keys["civitai_api"] = cfg.civitai_api_token
-    hf = (
-        os.environ.get("HF_TOKEN")
-        or os.environ.get("HUGGING_FACE_HUB_TOKEN")
-        or ""
-    ).strip()
+    hf = (cfg.hf_token or "").strip()
     if hf:
         keys["huggingface_api"] = hf
     if not keys:
@@ -273,21 +267,52 @@ def seed_swarmui_api_keys(cfg: Config, host: str, log: Log) -> None:
 
 
 def seed_civitai(cfg: Config, host: str, log: Log) -> None:
+    from gpu_rent.huggingface import is_huggingface_url
+
     entries = parse_models(cfg.models_manifest)
-    if not cfg.civitai_api_token:
-        log("Civitai-seed пропущен: нет CIVITAI_API_TOKEN — дефолт SwarmUI")
-        return
     if not entries:
-        log("Civitai-seed пропущен: манифест пуст — дефолт SwarmUI")
+        log("model-seed пропущен: манифест пуст — дефолт SwarmUI")
         return
-    jobs = []
+
+    jobs: list[dict] = []
     for entry in entries:
+        url = (entry.url or "").strip()
+        if url and is_huggingface_url(url):
+            name = url.rstrip("/").rsplit("/", 1)[-1].split("?", 1)[0] or "model.safetensors"
+            folder = MODEL_DIRS.get(entry.kind, entry.kind)
+            dest = f"{DATA}/Models/{folder}/{name}"
+            stem = Path(name).stem
+            swarm = {
+                "name": stem,
+                "title": stem,
+                "description": f"Hugging Face: {url}",
+                "trigger_phrase": "",
+                "author": "",
+                "tags": ["huggingface"],
+            }
+            jobs.append(
+                {
+                    "dest": dest,
+                    "url": url,
+                    "sha256": "",
+                    "auth": "hf",
+                    "sidecars": {
+                        f"{stem}.swarm.json": json.dumps(swarm, ensure_ascii=False, indent=2),
+                    },
+                }
+            )
+            continue
+
+        if not cfg.civitai_api_token:
+            continue
         vid = entry.version_id or (extract_version_id(entry.url) if entry.url else None)
         if not vid:
-            log(f"пропуск {entry.kind}: нет version_id")
+            log(f"пропуск {entry.kind}: нет version_id / HF url")
             continue
         try:
-            api_host, version = fetch_model_version(cfg.civitai_api_token, cfg.civitai_api_host, vid)
+            api_host, version = fetch_model_version(
+                cfg.civitai_api_token, cfg.civitai_api_host, vid
+            )
         except CloudError as exc:
             log(str(exc))
             continue
@@ -309,7 +334,11 @@ def seed_civitai(cfg: Config, host: str, log: Log) -> None:
             "title": version.get("name") or stem,
             "description": (version.get("description") or "")[:2000],
             "trigger_phrase": phrase,
-            "author": ((version.get("model") or {}) if isinstance(version.get("model"), dict) else {}).get("name"),
+            "author": (
+                ((version.get("model") or {}) if isinstance(version.get("model"), dict) else {}).get(
+                    "name"
+                )
+            ),
             "tags": version.get("tags") or [],
         }
         jobs.append(
@@ -317,22 +346,43 @@ def seed_civitai(cfg: Config, host: str, log: Log) -> None:
                 "dest": dest,
                 "url": _download_url(api_host, vid, info),
                 "sha256": sha,
+                "auth": "civitai",
                 "sidecars": {
                     f"{stem}.civitai.json": civitai_json,
                     f"{stem}.swarm.json": json.dumps(swarm, ensure_ascii=False, indent=2),
                 },
             }
         )
+
     if not jobs:
-        log("Civitai-seed: ни одной скачиваемой строки")
+        if not cfg.civitai_api_token:
+            log("Civitai-seed пропущен: нет CIVITAI_API_TOKEN — дефолт SwarmUI")
+        else:
+            log("Civitai-seed: ни одной скачиваемой строки")
         return
+
+    hf_n = sum(1 for j in jobs if j.get("auth") == "hf")
+    civ_n = len(jobs) - hf_n
+    if hf_n and not cfg.hf_token:
+        log(
+            "⚠ в манифесте есть Hugging Face URL, но нет HF_TOKEN — "
+            "gated/abliterated файлы дадут 401. Токен: huggingface.co/settings/tokens"
+        )
+    if civ_n and not cfg.civitai_api_token:
+        log("⚠ Civitai-строки пропущены: нет CIVITAI_API_TOKEN")
+
     if not any(e.kind == "checkpoint" for e in entries):
         log("в манифесте нет checkpoint — генерация может не стартовать")
     put_text(cfg, host, "/tmp/gpu-rent-civitai-jobs.json", json.dumps(jobs, indent=2))
-    put_text(cfg, host, "/tmp/gpu-rent-civitai.token", cfg.civitai_api_token + "\n", mode=0o600)
+    if cfg.civitai_api_token:
+        put_text(
+            cfg, host, "/tmp/gpu-rent-civitai.token", cfg.civitai_api_token + "\n", mode=0o600
+        )
+    if cfg.hf_token:
+        put_text(cfg, host, "/tmp/gpu-rent-hf.token", cfg.hf_token + "\n", mode=0o600)
     log(
-        f"Civitai: {len(jobs)} в манифесте — на VM качаю отсутствующие "
-        f"(уже есть + sha → skip; прогресс [N/{len(jobs)}])"
+        f"model-seed: {len(jobs)} файл(ов) "
+        f"(civitai={civ_n}, huggingface={hf_n}) — прогресс [N/{len(jobs)}]"
     )
     run_python(
         cfg,
@@ -419,7 +469,6 @@ def provision_llm(cfg: Config, host: str, log: Log) -> None:
             gguf_filename_from_url,
             parse_llamacpp_models,
         )
-        import os
 
         entries = parse_llamacpp_models(cfg.llamacpp_models_manifest)
         defaults = [e for e in entries if e.default]
@@ -439,13 +488,14 @@ def provision_llm(cfg: Config, host: str, log: Log) -> None:
             put_text(
                 cfg, host, "/tmp/gpu-rent-llamacpp-models.json", json.dumps(jobs, indent=2)
             )
-            hf = (
-                os.environ.get("HF_TOKEN")
-                or os.environ.get("HUGGING_FACE_HUB_TOKEN")
-                or ""
-            ).strip()
+            hf = (cfg.hf_token or "").strip()
             if hf:
                 put_text(cfg, host, "/tmp/gpu-rent-hf.token", hf + "\n", mode=0o600)
+            else:
+                log(
+                    "⚠ HF_TOKEN не задан — gated GGUF (abliterated и др.) часто дают 401. "
+                    "Добавь HF_TOKEN=hf_… в .env → https://huggingface.co/settings/tokens"
+                )
             log(f"llama.cpp: скачиваю {len(jobs)} GGUF из манифеста")
             run_python(
                 cfg,