jpmschweitzer/odysseus
1
0
Fork 0
mirror of https://github.com/pewdiepie-archdaemon/odysseus.git synced 2026-10-27 08:12:22 +01:00
Code Issues Packages Projects Releases Wiki Activity

Merge current dev and refresh PR #6739 from the 7011 integration

Browse Source
This commit is contained in:
pewdiepie-archdaemon
2026-10-26 16:21:17 +00:00
parent eb940f24a3 4d0610a1eb
commit 93da30c4af
157 changed files with 8685 additions and 1089 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/editor_agent_model.py
+13 -13
View File
@@ -1,8 +1,8 @@
"""Editor agent/planner model resolution.
The image-editor Quick Edit planner (`/api/editor/plan`, `/api/editor/agent`)
runs on a LOCAL Qwen3.5-9B vision model served on seneca when one is
reachable, instead of the hosted chat model. This module owns:
runs on an explicitly configured local vision model when one is reachable,
with the configured chat model as a fallback. This module owns:
* the local model base URLs / model id (settings-overridable),
* an ordered candidate list (per-request ``agent_model`` override, then the
@@ -30,18 +30,12 @@ import httpx
logger = logging.getLogger(__name__)
# The local Qwen3.5-9B vision server (seneca). :8097 (tmux AGENT_VLM) is the
# canonical editor-agent endpoint. Override with the ``editor_agent_base_url``
# setting (single URL wins) or the ``ODYSSEUS_EDITOR_AGENT_BASES`` env
# (comma-separated). NOTE: :8095 is a DIFFERENT server (Qwen3-VL-30B) used ONLY
# for the caption fallback below — never list it here as an agent model.
DEFAULT_LOCAL_BASES = [
"http://100.69.120.65:8097/v1",
]
# Local endpoints are opt-in through settings or ODYSSEUS_EDITOR_AGENT_BASES.
DEFAULT_LOCAL_BASES = []
DEFAULT_LOCAL_MODEL = "Qwen/Qwen3.5-9B"
# Vision caption model used when the editor-agent model cannot accept images.
DEFAULT_CAPTION_BASE = "http://100.69.120.65:8095/v1"
DEFAULT_CAPTION_BASE = ""
DEFAULT_CAPTION_MODEL = "Qwen/Qwen3.5-9B"
# Served-model-id cache per base (populated on first use). The local servers 400
@@ -278,6 +272,8 @@ async def caption_image(data_url: str, owner: Optional[str] = None, timeout: flo
}
]
base = caption_base()
if not base:
return None
model = await _resolve_model(base, caption_model(), local_model())
return await _post_chat(base, model, messages, None, timeout, max_tokens=200)
@@ -320,6 +316,9 @@ async def self_check(
return {"done": True, "issues": [], "retry_hint": "",
"reason": "unavailable", "text": "DONE"}
bases = local_bases() or list(DEFAULT_LOCAL_BASES)
if not bases:
return {"done": True, "issues": [], "retry_hint": "",
"reason": "unavailable", "text": "DONE"}
base = bases[0]
mdl = await _resolve_model(base, str(model or local_model()), local_model())
messages = [{
@@ -482,8 +481,9 @@ def _probe_worker() -> None:
_PROBE[base] = ok
logger.info("editor agent local base %s reachable=%s", base, ok)
cap = caption_base()
_PROBE[cap] = probe_models(cap)
logger.info("editor agent caption base %s reachable=%s", cap, _PROBE[cap])
if cap:
_PROBE[cap] = probe_models(cap)
logger.info("editor agent caption base %s reachable=%s", cap, _PROBE[cap])
def startup_probe() -> None: