feat(local-models): pin Flash Next's downloads to the revisions the catalog describes

Downloads resolved files from main, so a re-upload on the publisher's side would change the bytes
behind the catalog's exact sizes and the pricing measured from them. AssetFile carries a revision
and the download plan resolves each file at it. Flash Next pins unsloth's weights and projector to
38bb39ee and ggml-org's head to 052beea; entries without a revision still follow main.

Co-authored-by: Ryan Hill <38674843+RyanUnderhill@users.noreply.github.com>

hermes_cli/local_runtime/catalog.json

@@ -50,15 +50,18 @@
           "files": [
             {
               "path": "UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00003.gguf",
-              "size_bytes": 10946624
+              "size_bytes": 10946624,
+              "revision": "38bb39ee97821de2c9009abb7e93950eec396e66"
             },
             {
               "path": "UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00002-of-00003.gguf",
-              "size_bytes": 49835229856
+              "size_bytes": 49835229856,
+              "revision": "38bb39ee97821de2c9009abb7e93950eec396e66"
             },
             {
               "path": "UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00003-of-00003.gguf",
-              "size_bytes": 43836407744
+              "size_bytes": 43836407744,
+              "revision": "38bb39ee97821de2c9009abb7e93950eec396e66"
             }
           ],
           "lazy_bytes": 28800138240
@@ -75,12 +78,14 @@
       "mmproj": {
         "path": "mmproj-BF16.gguf",
         "size_bytes": 907542944,
-        "local": "mmproj-Qwen3.8-Flash-Next-BF16.gguf"
+        "local": "mmproj-Qwen3.8-Flash-Next-BF16.gguf",
+        "revision": "38bb39ee97821de2c9009abb7e93950eec396e66"
       },
       "mtp_head": {
         "repo": "ggml-org/Qwen3.8-Flash-Next-GGUF",
         "path": "mtp-Qwen3.8-Flash-Next-Q8_0.gguf",
-        "size_bytes": 4137429280
+        "size_bytes": 4137429280,
+        "revision": "052beeaca7bec4a303e59cc7bc630c4f3a1b845d"
       },
       "min_engine": "b11370",
       "quality": 95,

hermes_cli/local_runtime/catalog.py

@@ -39,6 +39,7 @@ class AssetFile:
     size_bytes: int
     local: str | None = None
     repo: str | None = None     # HF repo when it is not the entry's own (a companion published elsewhere)
+    revision: str = "main"      # commit whose bytes size_bytes and the entry's pricing describe
 
     @property
     def local_name(self) -> str:
@@ -301,7 +302,7 @@ def _asset_from(d: "dict | None") -> "AssetFile | None":
     if not d:
         return None
     return AssetFile(path=d["path"], size_bytes=int(d["size_bytes"]), local=d.get("local"),
-                     repo=d.get("repo"))
+                     repo=d.get("repo"), revision=d.get("revision", "main"))
 
 
 # Scalar CatalogEntry fields parsed from JSON: key -> (coerce, default); None default = required.

hermes_cli/web_routers/local_models.py

@@ -397,8 +397,8 @@ def _assign_default(job: Dict[str, Any], model_id: str) -> None:
 
 
 # ── downloads: ranged parallel streams ───────────────────────
-def _hf_url(repo: str, path: str) -> str:
-    return f"https://huggingface.co/{repo}/resolve/main/{path}"
+def _hf_url(repo: str, path: str, revision: str = "main") -> str:
+    return f"https://huggingface.co/{repo}/resolve/{revision}/{path}"
 
 
 def _model_id_for(gguf: Path) -> str:
@@ -418,8 +418,9 @@ def _variant_files_on_disk(model_id: str) -> "list[Path]":
 
 def _download_plan(entry, variant) -> list:
     """Everything a variant needs: split parts + mmproj/draft/MTP head assets, as (url, dest, bytes) tuples."""
-    plan = [(_hf_url(entry.repo, a.path), bootstrap.models_dir() / a.local_name, a.size_bytes) for a in variant.files]
-    plan += [(_hf_url(a.repo or entry.repo, a.path), bootstrap.assets_dir() / a.local_name, a.size_bytes)
+    plan = [(_hf_url(entry.repo, a.path, a.revision), bootstrap.models_dir() / a.local_name, a.size_bytes)
+            for a in variant.files]
+    plan += [(_hf_url(a.repo or entry.repo, a.path, a.revision), bootstrap.assets_dir() / a.local_name, a.size_bytes)
              for a in (entry.mmproj, entry.draft, entry.mtp_head) if a is not None]
     return plan
 

tests/hermes_cli/test_local_mtp_head.py

@@ -27,15 +27,15 @@ ENTRY = CatalogEntry(
     id="head-model", display_name="Head Model", description="", repo="org/Head-Model-GGUF",
     variants=(BUILD,), n_ctx_train=262144, full_layers=12, recurrent_layers=36, per_layer_f16=2048,
     moe=True, n_vocab=248320, mtp_draft_depth=3,
-    mtp_head=AssetFile("mtp-Head-Model-Q8_0.gguf", 4 * GIB, repo="other/Head-Model-GGUF"))
+    mtp_head=AssetFile("mtp-Head-Model-Q8_0.gguf", 4 * GIB, repo="other/Head-Model-GGUF", revision="0123abc"))
 
 
 def test_the_head_downloads_with_the_model_from_its_own_repo_and_is_priced_before_download():
     without_head = replace(ENTRY, mtp_head=None)
 
     urls = [url for url, _, _ in local_models._download_plan(ENTRY, BUILD)]
-    assert "https://huggingface.co/other/Head-Model-GGUF/resolve/main/mtp-Head-Model-Q8_0.gguf" in urls
-    assert all("/org/Head-Model-GGUF/" in url for url in urls[:2])
+    assert "https://huggingface.co/other/Head-Model-GGUF/resolve/0123abc/mtp-Head-Model-Q8_0.gguf" in urls
+    assert all("/org/Head-Model-GGUF/resolve/main/" in url for url in urls[:2])
     assert ENTRY.mtp_capable and not without_head.mtp_capable
     with_plan = ENTRY.launch_plan(BUILD, UNIFIED_128)
     without_plan = without_head.launch_plan(BUILD, UNIFIED_128)