feat(us-016): auto-detect shard range from model config

Layer count is now fetched from the curated catalog (zero network calls
for known models) or via AutoConfig.from_pretrained() (~1 KB config.json
only) when model_id is given without --shard-start/--shard-end.

- model_catalog: add detect_num_layers(), two small Qwen models at top
- startup: _detect_num_layers() helper; shard range auto-derived
- wizard: show detected layer count for custom HF repos
- tests: 3 new tests for auto-shard; fix catalog-order assumptions

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

tests/test_mining_cli.py

@@ -21,7 +21,7 @@ def test_curated_models_list_is_non_empty():
 def test_model_preset_vram_for_quant():
     from meshnet_node.model_catalog import CURATED_MODELS
 
-    m = CURATED_MODELS[0]  # Llama-3-70B
+    m = next(m for m in CURATED_MODELS if "Llama-3-70B" in m.name)
     assert m.vram_for_quant("nf4") == m.vram_nf4
     assert m.vram_for_quant("int8") == m.vram_int8
     assert m.vram_for_quant("bf16") == m.vram_bf16
@@ -39,7 +39,8 @@ def test_model_preset_fits_vram():
 def test_recommended_quant_respects_vram():
     from meshnet_node.model_catalog import CURATED_MODELS
 
-    m = CURATED_MODELS[0]  # Llama-3-70B: nf4=18, int8=40, bf16=140
+    m = next(m for m in CURATED_MODELS if "Llama-3-70B" in m.name)
+    # nf4=18, int8=40, bf16=140
     assert m.recommended_quant(200) == "bf16"
     assert m.recommended_quant(50) == "int8"
     assert m.recommended_quant(20) == "nf4"
@@ -125,9 +126,9 @@ def test_wizard_writes_config_on_happy_path(tmp_path, monkeypatch):
     # Tracker not reachable (stub)
     monkeypatch.setattr(wiz, "_ping_tracker", lambda url: False)
 
-    # Simulate user selecting model 3 (Mixtral), quant 1 (nf4), default dir, default tracker, default wallet
+    # Simulate user selecting model 1 (Qwen2.5-0.5B), quant 1 (nf4), default dir, default tracker, default wallet
     inputs = iter([
-        "3",          # pick Mixtral (index 3 in CURATED_MODELS)
+        "1",          # pick Qwen2.5-0.5B-Instruct (index 1 in CURATED_MODELS)
         "1",          # quant NF4
         str(tmp_path / "models"),  # download dir
         "http://localhost:8080",   # tracker
@@ -136,7 +137,7 @@ def test_wizard_writes_config_on_happy_path(tmp_path, monkeypatch):
     monkeypatch.setattr("builtins.input", lambda prompt="": next(inputs))
 
     cfg = wiz.run_wizard(config_path_override=tmp_path / "config.json")
-    assert cfg["model_hf_repo"] == "mistralai/Mixtral-8x7B-Instruct-v0.1"
+    assert cfg["model_hf_repo"] == "Qwen/Qwen2.5-0.5B-Instruct"
     assert cfg["quantization"] == "nf4"
     assert "download_dir" in cfg
     assert cfg["tracker_url"] == "http://localhost:8080"
@@ -265,6 +266,63 @@ def test_config_command_prints_saved_config(tmp_path, monkeypatch, capsys):
     assert data["model_hf_repo"] == saved["model_hf_repo"]
 
 
+def test_detect_num_layers_returns_catalog_value_without_network(monkeypatch):
+    """detect_num_layers uses the curated catalog first — no network call."""
+    from meshnet_node.model_catalog import detect_num_layers
+
+    # Qwen2.5-0.5B is in the catalog with 24 layers
+    layers = detect_num_layers("Qwen/Qwen2.5-0.5B-Instruct")
+    assert layers == 24
+
+
+def test_detect_num_layers_returns_none_on_error(monkeypatch):
+    from meshnet_node.model_catalog import detect_num_layers
+
+    # Monkeypatch AutoConfig to raise
+    import meshnet_node.model_catalog as cat
+    monkeypatch.setattr(cat, "detect_num_layers", lambda repo: None if "bad" in repo else detect_num_layers(repo))
+    assert cat.detect_num_layers("bad/repo") is None
+
+
+def test_startup_auto_detects_shard_range(monkeypatch, tmp_path):
+    """When shard_start/end are None, startup reads layer count from catalog."""
+    from meshnet_node import startup as su
+    from meshnet_node.model_catalog import detect_num_layers
+
+    calls = []
+
+    def fake_detect(repo):
+        calls.append(repo)
+        return 24  # Qwen2.5-0.5B
+
+    monkeypatch.setattr(su, "_detect_num_layers", fake_detect)
+
+    # Fake hardware detection
+    monkeypatch.setattr(su, "detect_hardware", lambda: {"device": "cpu", "gpu_name": None, "vram_mb": 0})
+
+    # Fake wallet
+    monkeypatch.setattr(su, "load_or_create_wallet", lambda **kw: (None, None, "fake-wallet"))
+
+    # Fake TorchNodeServer
+    class FakeNode:
+        chat_completion_count = 0
+        def start(self): return 9999
+        def stop(self): pass
+
+    import meshnet_node.startup as su2
+    monkeypatch.setattr(su2, "TorchNodeServer", lambda **kw: FakeNode())
+
+    node = su.run_startup(
+        tracker_url="http://localhost:8080",
+        model_id="Qwen/Qwen2.5-0.5B-Instruct",
+        # shard_start and shard_end intentionally omitted
+        quantization="bfloat16",
+        host="127.0.0.1",
+    )
+    assert calls == ["Qwen/Qwen2.5-0.5B-Instruct"]
+    assert isinstance(node, FakeNode)
+
+
 def test_legacy_start_subcommand_accepted(monkeypatch):
     """meshnet-node start --tracker http://... does not crash on arg parsing."""
     from meshnet_node.cli import main