jpmschweitzer/odysseus
1
0
Fork 0
mirror of https://github.com/pewdiepie-archdaemon/odysseus.git synced 2026-10-31 10:12:21 +01:00
Code Issues Packages Projects Releases Wiki Activity

fix(hwfit): keep the MoE active size for HF collection models (#6729)

Browse Source 
_entry_from_collection_item parsed the repo name (the only source of the
"-A3B" active size) only when HF's numParameters was missing, which is
almost never. Every MoE model from the HF collection catalog was scored
as dense, with speed estimates 3-4x too low.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
dxrpl
2026-10-15 01:11:46 +02:00
co-authored by Claude Opus 5.5
parent c3819e3cbf
commit bfa514f6e8
2 changed files with 37 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
services/hwfit/hf_discovery.py
+4 -2
View File
@@ -210,9 +210,11 @@ def _entry_from_collection_item(collection, item, source):
if repo_prefix and not repo_id.startswith(repo_prefix):
return None
raw_params = item.get("numParameters") or 0
active = None
# Read the name even when HF gives a total: it is the only source of the
# MoE active size (e.g. "-35B-A3B"), which the speed estimate depends on.
name_params, active = _parse_params_from_name(repo_id)
if not raw_params:
raw_params, active = _parse_params_from_name(repo_id)
raw_params = name_params
param_label, raw_params = _format_params(raw_params)
if not raw_params:
return None