runpod: a base and its adapter are two models, not one path (v0.3.39)

Seeds were de-duplicated by `path`. A base model and a LoRA of it share one path
by definition, so the adapter's entry was dropped and the pod was told about the
base alone — observed live, one line after the sibling seeding had correctly
found it:

    [runpod] pool 'pool:digesta-qwen38' is shared with: diogene-alpha
    [runpod] pod will be told about 1 model(s)

The dedup exists because a capability pod's list can overlap the one model a
per-model pool also seeds, which is a real case; it just has to key on what the
pod will SERVE the entry as — path plus alias, id and adapters — rather than on
the weights the entries happen to have in common.

Orchestrator-side only: the pod registers what it is handed, so no new image.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PxYHWbCAjjjD7A1rhFKtaW

codai/__init__.py

@@ -16,7 +16,7 @@
 
 # Canonical product version for CoderAI — single source of truth. Both the API
 # metadata and the admin web UI read from here.
-__version__ = "0.3.38"
+__version__ = "0.3.39"
 __tagline__ = "Complete orchestration, distribution and escalation of remotizable advanced inference."
 
 # Configure the CUDA caching allocator BEFORE torch is imported anywhere.

codai/api/runpod_worker.py

@@ -1104,13 +1104,18 @@ def _plan(engine, image, args, mcfg, api_key, entry, served,
         if one:
             seeds.extend(_json.loads(one))
     if seeds:
-        # De-duplicate by path: a capability pod's list can overlap the one model
-        # a per-model pool also seeds.
+        # De-duplicate by what the pod will SERVE it as, not by path: a capability
+        # pod's list can overlap the one model a per-model pool also seeds, which
+        # is what this guards against — but a base model and a LoRA of it share
+        # one path by definition, so keying on path silently dropped the adapter
+        # and the pod was "told about 1 model(s)" when it needed two.
         seen, unique = set(), []
         for s in seeds:
-            if s.get("path") in seen:
+            key = (s.get("path"), s.get("alias") or "", s.get("id") or "",
+                   _json.dumps(s.get("loras") or [], sort_keys=True))
+            if key in seen:
                 continue
-            seen.add(s.get("path"))
+            seen.add(key)
             unique.append(s)
         env["CODERAI_SEED_MODELS"] = _json.dumps(unique)
         print(f"[runpod] pod will be told about {len(unique)} model(s)", flush=True)

tests/test_text_lora_to_pod.py

@@ -382,3 +382,20 @@ def test_the_limiter_can_be_turned_off_by_env(monkeypatch):
     assert ratelimit._limits_enabled() is True
     monkeypatch.setenv("CODERAI_SERVER_RATE_LIMIT_ENABLED", "0")
     assert ratelimit._limits_enabled() is False
+
+
+def test_a_base_and_its_adapter_both_reach_the_pod(monkeypatch):
+    """Seeds were de-duplicated by PATH, and a base model and a LoRA of it share
+    one path by definition — so the adapter was dropped and the pod was told
+    about one model when it needed two. Observed live: "told about 1 model(s)"."""
+    from codai.api import runpod_worker as rw
+    base = {"id": "qwen38-awq", "path": "cyankiwi/Qwen3.8-27B-AWQ-INT4",
+            "alias": "qwen38"}
+    tuned = {"id": "diogene-alpha", "path": "cyankiwi/Qwen3.8-27B-AWQ-INT4",
+             "loras": [{"name": "diogene-alpha", "path": "sha256:" + "a" * 64}]}
+    mcfg = rw.parse_model_runpod({})
+    plan = rw._plan("vllm", "img", "", mcfg, "key", base, "qwen38-awq",
+                    seed_entries=[tuned])
+    seeded = json.loads((plan.get("env") or {}).get("CODERAI_SEED_MODELS") or "[]")
+    assert len(seeded) == 2, seeded
+    assert any(m.get("loras") for m in seeded), "the adapter's entry was dropped"