deploy: update Space from deploy_preflight --push

.codeboarding/logs/wrapper-server.log

@@ -112,3 +112,13 @@ INFO:     127.0.0.1:44498 - "GET /health HTTP/1.1" 200 OK
 [stderr] 2026-06-14 04:55:32 INFO     [watchfiles.main:308] 1 change detected
 [stderr] 2026-06-14 04:55:32 WARNING  [codeboarding_pro.analysis_controller:506] Skipping incremental update: full analysis artifacts not ready yet
 [stderr] 2026-06-14 04:55:32 INFO     [codeboarding_pro.analysis_controller:314] No cached analysis results; skipping health report
+[stderr] 2026-06-14 04:56:51 INFO     [watchfiles.main:308] 1 change detected
+[stderr] 2026-06-14 04:56:52 WARNING  [codeboarding_pro.analysis_controller:506] Skipping incremental update: full analysis artifacts not ready yet
+[stderr] 2026-06-14 04:56:52 INFO     [codeboarding_pro.analysis_controller:314] No cached analysis results; skipping health report
+[stderr] 2026-06-14 04:57:07 INFO     [watchfiles.main:308] 1 change detected
+[stderr] 2026-06-14 04:57:07 WARNING  [codeboarding_pro.analysis_controller:506] Skipping incremental update: full analysis artifacts not ready yet
+[stderr] 2026-06-14 04:57:07 INFO     [codeboarding_pro.analysis_controller:314] No cached analysis results; skipping health report
+[stderr] 2026-06-14 04:57:18 INFO     [watchfiles.main:308] 1 change detected
+[stderr] 2026-06-14 04:57:18 INFO     [watchfiles.main:308] 2 changes detected
+[stderr] 2026-06-14 04:57:18 WARNING  [codeboarding_pro.analysis_controller:506] Skipping incremental update: full analysis artifacts not ready yet
+[stderr] 2026-06-14 04:57:18 INFO     [codeboarding_pro.analysis_controller:314] No cached analysis results; skipping health report

app.py

@@ -90,6 +90,7 @@ MODEL_OPTIONS = [
     "LoRA v2 (Standard E4B)",
     "Base (Gemma 4 E4B)",
     "Modal API (remote)",
+    # Future: "Ollama v3 (local GGUF)", "Ollama v2 (local GGUF)" — once GGUF files are pushed to Ollama.com
 ]
 
 MODEL_LORA_MAP = {
@@ -250,18 +251,17 @@ def warm_up_cb() -> str:
     return f"<div class='ce-sub' style='font-size:12px;'>MODEL · {llm.warm_up()}</div>"
 
 
-def pick_model(choice: str) -> str:
-    """Model switcher. Live serves now; the trained LoRA / QAT variants are tracked
-    in learn/finetune and are NOT swapped into the live runtime until a held-out eval
-    earns it — surfaced honestly rather than faked (the project's honesty rule)."""
-    if choice == "Live":
-        return f"<div class='ce-sub' style='font-size:12px;'>MODEL · {llm.backend_status()}</div>"
-    note = ("LoRA on the accumulated ledger" if choice == "LoRA"
-            else "QAT (quantization-aware) variant")
-    return ("<div class='ce-sub' style='font-size:12px;'>MODEL · "
-            f"<span style='color:var(--ao-yellow);'>●</span> {choice} selected · {note} is "
-            "<b>training on Modal</b>, not yet serving — Live model is active. "
-            "See learn/finetune.</div>")
+def select_model(choice: str) -> str:
+    """Model switcher (backend by learn/finetune via _apply_model_choice). Applies the
+    chosen backend, then reports the model actually serving. The LoRA v2/v3 adapters serve
+    through ZeroGPU on the Space (and fall back gracefully if the GPU/deps aren't present);
+    Base is the stock Gemma 4 E4B; Modal API is the remote endpoint."""
+    choice = choice or MODEL_OPTIONS[0]
+    try:
+        _apply_model_choice(choice)
+    except Exception:
+        pass
+    return f"<div class='ce-sub' style='font-size:12px;'>MODEL · {choice} · {llm.backend_status()}</div>"
 
 
 # ── simulated environment (this is a sim lab — conditions are generated, overridable) ──
@@ -540,14 +540,18 @@ def build() -> gr.Blocks:
         gr.HTML(command_bar(llm.backend_status()))
         # header row: model switcher + warm-up + live status (studio-16/18)
         with gr.Row(elem_id="ce-modelswitch"):
-            model_switch = gr.Radio(["Live", "LoRA", "QAT"], value="Live", show_label=False,
-                                    elem_classes=["ce-seg"], scale=0)
+            model_select = gr.Dropdown(MODEL_OPTIONS, value=MODEL_OPTIONS[0],
+                                       show_label=False, container=False,
+                                       elem_classes=["ce-modeldd"], scale=0)
             warm_btn = gr.Button("WARM UP MODEL", elem_classes=["ce-pillbtn"], scale=0)
             model_status = gr.HTML(_status_html())
+        gr.HTML("<div class='ce-sub' style='font-size:11px;'>Model · the LoRA v2/v3 adapters serve "
+                "via ZeroGPU on the Space; Base is the stock Gemma 4 E4B; Modal API is the remote "
+                "endpoint. Local users: pull the LoRA from HF Hub "
+                "(kylebrodeur/microfactory-node-lora-v2 · -v3-qat) or via ollama.</div>")
 
         state = gr.State()
         part = gr.State({"geometry": None, "mesh": None, "label": None, "read": None})
-        model_choice_state = gr.State("LoRA v3 (QAT E4B)")  # hidden, UI agent will replace with dropdown
 
         with gr.Tabs() as tabs:
             # ───────────────────────── STUDIO · define the job ───────────────────────
@@ -712,7 +716,7 @@ def build() -> gr.Blocks:
 
         # Model warm-up + switcher
         warm_btn.click(warm_up_pending, None, [model_status]).then(warm_up_cb, None, [model_status])
-        model_switch.change(pick_model, [model_switch], [model_status])
+        model_select.change(select_model, [model_select], [model_status])
 
         # Simulated environment: roll on load, re-roll on demand, keep readout + footer in sync.
         sensor_outs = [temp, humidity, bed_position, sensors_readout]
@@ -731,10 +735,9 @@ def build() -> gr.Blocks:
         build_outs = [backend, precedent, reasoning, risks, settings_html, spine_notes,
                       gcode_html, confirm_btn, node_cards, state, vprint, vp_slider, vp_layer,
                       build_loader, build_results]
-        # NOTE: model_choice dropdown will be added by UI agent. Currently defaults to "LoRA v3 (QAT E4B)".
         run_btn.click(build_start, [part], build_start_outs).then(
             None, None, None, js=_SCROLL_TOP).then(
-            build_job, [part, material, description, temp, humidity, bed_position, model_choice_state],
+            build_job, [part, material, description, temp, humidity, bed_position, model_select],
             build_outs).then(job_readout, [state], [p_job])
         vp_slider.change(scrub_layer, [vp_slider, part], [vp_layer])
         read_toggle.change(toggle_read, [read_toggle, state],

learn/finetune/SERVING.md

@@ -60,6 +60,43 @@ Path B (Ollama ADAPTER command) because:
 - `convert_lora_to_gguf.py` compatibility with Gemma 4 not tested
 - Merge→GGUF is the most battle-tested path
 
+### Quantization
+
+The GGUF pipeline uses `--outtype q4_k_m` by default:
+- **Does NOT reduce parameter count**: 8B params stay 8B. Quantization reduces
+  weight precision from 16-bit (bf16) to 4-bit per weight.
+- **File size**: ~5-6GB for 8B model (vs ~16GB for bf16 safetensors)
+- **Quality**: q4_k_m is the recommended balance of size vs quality
+- **v3 advantage**: QAT-trained model retains more quality after quantization
+  because it was trained with simulated quantization during fine-tuning
+
+Other quantization options (pass `--outtype` to override):
+| Format | Size (8B) | Quality | Use Case |
+|--------|-----------|---------|----------|
+| q8_0 | ~8.5GB | Near-lossless | Max quality, local GPU |
+| q4_k_m | ~5.5GB | Good | Balanced (default) |
+| q4_k_s | ~5.0GB | Slightly lower | Tight storage |
+| q4_0 | ~4.5GB | Lower | QAT-optimized (Google's QAT Q4_0 target) |
+
+### Parallel GGUF runs
+Both tracks can run simultaneously — the pipeline uses separate Modal functions
+(merge on GPU, convert on CPU) and writes distinct output filenames via `--name`:
+
+```bash
+# Terminal 1 — v2:
+modal run learn/finetune/gguf_pipeline_modal.py \
+  --adapter kylebrodeur/microfactory-node-lora-v2 \
+  --name microfactory-node-v2
+
+# Terminal 2 — v3 (simultaneously):
+modal run learn/finetune/gguf_pipeline_modal.py \
+  --base google/gemma-4-E4B-it-qat-q4_0-unquantized \
+  --adapter kylebrodeur/microfactory-node-lora-v3-qat \
+  --name microfactory-node-v3-qat
+```
+
+Download both: `modal volume get microfactory-node-finetune gguf/ --force`
+
 ---
 
 ## 2. Modal Model Hosting — Implemented

learn/finetune/gguf_pipeline_modal.py

@@ -87,11 +87,11 @@ if modal is not None:
     @app.function(image=cpu_image, timeout=3600,
                   volumes={"/out": vol},
                   secrets=[modal.Secret.from_name("chief-engineer-secrets")])
-    def convert_to_gguf(merged_path: str, outtype: str = "q4_k_m") -> str:
+    def convert_to_gguf(merged_path: str, outtype: str = "q4_k_m", name: str = "microfactory-node") -> str:
         """Convert merged HF model to GGUF using llama.cpp."""
         import subprocess
 
-        out_name = "microfactory-node.gguf"
+        out_name = f"{name}.gguf"
         out_path = f"/out/gguf/{out_name}"
         os.makedirs("/out/gguf", exist_ok=True)
 
@@ -115,14 +115,14 @@ if modal is not None:
 
     @app.local_entrypoint()
     def main(base: str = BASE_MODEL, adapter: str = "",
-             outtype: str = "q4_k_m"):
+             outtype: str = "q4_k_m", name: str = "microfactory-node"):
         if not adapter:
             print("ERROR: --adapter required. Example:")
-            print("  modal run learn/finetune/gguf_pipeline_modal.py --adapter kylebrodeur/microfactory-node-lora-v2")
+            print("  modal run learn/finetune/gguf_pipeline_modal.py --adapter kylebrodeur/microfactory-node-lora-v2 --name microfactory-node-v2")
             return
 
         print(f"=== GGUF Pipeline: {adapter} ===")
-        print(f"Base: {base} | Outtype: {outtype}")
+        print(f"Base: {base} | Outtype: {outtype} | Name: {name}")
 
         # Step 1: Merge on GPU
         print("\n--- Step 1: Merge LoRA (GPU) ---")
@@ -131,14 +131,14 @@ if modal is not None:
 
         # Step 2: Convert to GGUF on CPU
         print("\n--- Step 2: Convert to GGUF (CPU) ---")
-        gguf_path = convert_to_gguf.remote(merged_path, outtype)
+        gguf_path = convert_to_gguf.remote(merged_path, outtype, name)
         print(f"\n=== PIPELINE COMPLETE ===")
         print(f"GGUF file: {gguf_path}")
         print(f"\nDownload:")
         print(f"  modal volume get microfactory-node-finetune gguf/ --force")
         print(f"\nOllama import:")
         print(f"  cat > Modelfile << 'EOF'")
-        print(f"  FROM ./microfactory-node.gguf")
+        print(f"  FROM ./{name}.gguf")
         print(f'  TEMPLATE """{{{{ if .System }}}}<start_of_turn>system')
         print(f'  {{{{ .System }}}}<end_of_turn>')
         print(f'  {{{{ end }}}}<start_of_turn>user')
@@ -148,8 +148,8 @@ if modal is not None:
         print(f'  PARAMETER stop "<start_of_turn>user"')
         print(f'  PARAMETER stop "<end_of_turn>"')
         print(f'  EOF')
-        print(f"  ollama create microfactory-node -f Modelfile")
-        print(f"  ollama run microfactory-node")
+        print(f"  ollama create {name} -f Modelfile")
+        print(f"  ollama run {name}")
 
 
 if __name__ == "__main__":