Add auto-deploy feature to StreamWise dashboard

[James-QiuHaoran]

Add an Auto Deploy button to the web dashboard that automatically optimizes GPU resource allocation across workflow components using the model provisioner's greedy allocator.

Depends on: #312 (refactor: move policies to model_provisioner)

Changes

• streamwise/allocator_bridge.py (new): Maps allocator output to K8s deployment parameters (Model enum → container names, GPU specs)
• streamwise/streamwise.py: Add /api/auto_deploy and /api/auto_deploy/confirm routes
• streamwise/templates/add_pod.html: Auto-deploy UI section with GPU budget inputs, workflow selector, and deployment plan preview
• tests/streamwise/test_allocator_bridge.py (new): 15 tests for allocator bridge
• tests/streamwise/test_streamwise_auto_deploy.py (new): 10 tests for auto-deploy API

Testing

• 108 streamwise tests pass (1 pre-existing Windows failure: \ est_list_files)
• flake8 clean
• mypy clean

[Copilot]

Pull request overview

Adds a new "Auto Deploy" feature to the StreamWise dashboard: an HTTP API and UI that runs the greedy AutoModelAllocator over a user-supplied GPU budget + workflow, previews the resulting per-container deployment plan, and (on confirm) creates the corresponding pods via pod_manager.add_pod. A new bridge module translates the simulator's abstract Model/GPUType allocations into concrete K8s container specs, and the Dockerfile/setup script now bundle model_provisioner and simulator into the streamwise image.

Changes:

• New streamwise/allocator_bridge.py mapping Model/GPUType allocations to deployment specs and exposing run_allocator + JSON serialization.
• New /api/auto_deploy, /api/auto_deploy/confirm, /api/auto_deploy/workflows routes in streamwise.py, plus an Auto-Deploy UI section in add_pod.html with optimize/confirm flow.
• Packaging changes (Dockerfile, setup_image.sh, requirements.txt) to ship simulator + model_provisioner and add pandas/tabulate; new tests for the bridge and the API routes.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 14 comments.

Show a summary per file

File	Description
streamwise/allocator_bridge.py	New bridge: Model→container/MIG/resource mapping and run_allocator/serialization helpers.
streamwise/streamwise.py	Three new auto-deploy routes invoking the bridge and pod_manager.add_pod.
streamwise/templates/add_pod.html	Auto-deploy form, results table, and fetch logic for optimize/confirm.
streamwise/requirements.txt	Adds pandas and tabulate dependencies.
deployment/streamwise/Dockerfile	Copies model_provisioner and simulator into the image.
deployment/setup_image.sh	Stages model_provisioner and simulator for the streamwise image build.
tests/streamwise/test_allocator_bridge.py	Unit/integration tests for the bridge (mappings, specs, real allocator runs).
tests/streamwise/test_streamwise_auto_deploy.py	API tests for the three new routes, with K8s mocks and allocator patches.

[Copilot]

Just as a heads up, I was blocked by some firewall rules while working on your feedback. Expand below for details.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• example.com
  • Triggering command: /home/REDACTED/.local/bin/pytest /home/REDACTED/.local/bin/pytest --ignore=tests/simulator --ignore=tests/streamwise --ignore=tests/streamwise_app -vv -main/dist/indexgpg.program (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

[github-actions]

Test Results

1 174 tests  +29   1 171 ✅ +26   14m 56s ⏱️ -5s
   13 suites ± 0       0 💤 ± 0 
   13 files   ± 0       3 ❌ + 3 

For more details on these failures, see this check.

Results for commit 19b539d. ± Comparison against base commit 9bcbc4e.

♻️ This comment has been updated with latest results.

[goiri]

Try to make it generic for GPUType

[goiri]

Make a row per GPU type and make the number a slider

[goiri]

This should show the friendly name

[goiri]

This should show the friendly name.

[goiri]

Next version may want to allow customizing the plan.

[goiri]

Pass the disaggregation or leave a comment that is not disaggregated.

[goiri]

Make this output nicer.

[github-actions]

Lint Results

Check	Status
Python	✅
Shell	✅
YAML	✅
JSON	✅
Markdown	✅
Bicep	✅

[github-actions]

Mypy Type Checking

✅ No issues found

Metric	Count
📂 Files	300
❌ Errors	0
⚠️ Warnings	0
📝 Notes	0

Full mypy output

Success: no issues found in 300 source files

[github-actions]

Diff Coverage

Diff: origin/main...HEAD, staged and unstaged changes

• streamwise/allocator_bridge.py (69.3%): Missing lines 19,58-62,117-124,137-141,144-148,150-154,157-159,226,232,235,240-247,249,290,292-293,321,331,335
• streamwise/container_config.py (88.2%): Missing lines 16,20
• streamwise/streamwise.py (54.1%): Missing lines 555,573,575-576,580-584,586-592,635,638-639,641-643,645,648-649,652-653,792,823-824,829-832,865-866,886-887,895-898,902-903,914-924,926-930,941-943,975-976,978,983-985,994,997-1004,1006
• tests/streamwise/test_allocator_bridge.py (93.0%): Missing lines 255-259,268-269
• tests/streamwise/test_streamwise.py (100%)
• tests/streamwise/test_streamwise_auto_deploy.py (100%)

Summary

• Total: 585 lines
• Missing: 138 lines
• Coverage: 76%

streamwise/allocator_bridge.py

Lines 15-23

  15 if _HERE not in sys.path:
  16     sys.path.insert(0, _HERE)
  17 _REPO_ROOT = os.path.dirname(_HERE)
  18 if _REPO_ROOT not in sys.path:
! 19     sys.path.insert(0, _REPO_ROOT)
  20 import model_provisioner  # noqa: E402, F401 — adds simulator/ to sys.path
  21 
  22 from dataclasses import dataclass
  23 from typing import Optional

---

Lines 54-66

  54 
  55 
  56 def get_mig_profile(container_name: str, gpu_type: GPUType) -> Optional[str]:
  57     """Return a MIG profile only when MIG is available and the GPU type supports it."""
! 58     if not MIG_AVAILABLE:
! 59         return None
! 60     if gpu_type not in MIG_CAPABLE_GPU_TYPES:
! 61         return None
! 62     return MIG_CONTAINERS.get(container_name)
  63 
  64 
  65 # Mapping from StreamWise app name to simulator workflow key
  66 APP_TO_WORKFLOW: dict[str, str] = {

---

Lines 113-128

  113 
  114 
  115 def _calc_actual_gpus_per_type(specs: list['DeploymentSpec']) -> dict[GPUType, int]:
  116     """Calculate actual GPUs needed per GPUType from deployment specs."""
! 117     result: dict[GPUType, int] = {}
! 118     for spec in specs:
! 119         if spec.mig_profile:
! 120             continue
! 121         gpu_type = _POD_STR_TO_GPU_TYPE.get(spec.gpu_type or "")
! 122         if gpu_type is not None:
! 123             result[gpu_type] = result.get(gpu_type, 0) + spec.gpu
! 124     return result
  125 
  126 
  127 def _trim_specs_for_type(
  128     specs: list['DeploymentSpec'], gpu_type_str: str, excess: int

---

Lines 133-163

  133     Prefers removing replicas of the most-replicated scalable container (typically
  134     realesrgan/upscaler) to minimize impact on pipeline throughput.
  135     """
  136     # Count replicas per container on this GPU type (only scalable ones)
! 137     from collections import Counter
! 138     type_counts: Counter[str] = Counter()
! 139     for spec in specs:
! 140         if spec.gpu_type == gpu_type_str and spec.gpu > 0 and spec.container_name not in COLOCATED_CONTAINERS:
! 141             type_counts[spec.container_name] += 1
  142 
  143     # Prefer trimming containers with most replicas (least impact per removal)
! 144     trimmed = 0
! 145     result_specs = list(specs)
! 146     for container_name, _count in type_counts.most_common():
! 147         if trimmed >= excess:
! 148             break
  149         # Remove replicas from the end of the list
! 150         for i in range(len(result_specs) - 1, -1, -1):
! 151             if trimmed >= excess:
! 152                 break
! 153             spec = result_specs[i]
! 154             if (spec.container_name == container_name
  155                     and spec.gpu_type == gpu_type_str
  156                     and spec.gpu > 0):
! 157                 trimmed += spec.gpu
! 158                 result_specs.pop(i)
! 159     return result_specs
  160 
  161 
  162 def get_available_workflows() -> list[str]:
  163     """Return list of available workflow names for the UI."""

---

Lines 222-230

  222     # Load latency data and run allocator
  223     data_dir = _get_data_dir()
  224     latency_data = load_latency_data(data_dir=data_dir)
  225 
! 226     allocator = AutoModelAllocator(
  227         workflow=workflow,
  228         latency_data=latency_data,
  229         policy=STREAMWISE_POLICY,
  230     )

---

Lines 228-253

  228         latency_data=latency_data,
  229         policy=STREAMWISE_POLICY,
  230     )
  231 
! 232     result = allocator.allocate(num_gpus=allocator_gpus, verbose=False)
  233 
  234     # Convert result to deployment specs
! 235     specs = result_to_deployment_specs(result)
  236 
  237     # Trim deployment specs back to the user's actual budget.
  238     # Also handles MIG-unavailable overflow (e.g., OTHERS allocates 1 GPU
  239     # but kokoro+yolo each need a full GPU = 2).
! 240     actual_per_type = _calc_actual_gpus_per_type(specs)
! 241     for gpu_type, budget_count in num_gpus.items():
! 242         actual = actual_per_type.get(gpu_type, 0)
! 243         if actual <= budget_count:
! 244             continue
! 245         excess = actual - budget_count
! 246         gpu_type_str = GPU_TYPE_TO_POD_STR[gpu_type]
! 247         specs = _trim_specs_for_type(specs, gpu_type_str, excess)
  248 
! 249     return DeploymentPlan(
  250         specs=specs,
  251         result=result,
  252         workflow_name=workflow_name,
  253         gpu_budget=gpu_budget,

---

Lines 286-297

  286                         container_name in COLOCATED_CONTAINERS
  287                         and not STREAMWISE_POLICY.disaggregation.get(Model.HF, False)
  288                     )
  289                     if is_colocated:
! 290                         gpu_count = 0
  291                     elif MIG_AVAILABLE and container_name in MIG_CONTAINERS:
! 292                         mig_profile = MIG_CONTAINERS[container_name]
! 293                         gpu_count = 1
  294                     elif container_name in MIG_CONTAINERS:
  295                         # MIG not available: use 1 full GPU instead of a MIG slice
  296                         gpu_count = 1
  297                     else:

---

Lines 317-325

  317     # when MIG is unavailable and services fall back to full GPUs).
  318     actual_gpus: dict[str, int] = {}
  319     for spec in plan.specs:
  320         if spec.mig_profile:
! 321             continue  # MIG slices don't count against full GPU budget
  322         gpu_type_key = spec.gpu_type or "unknown"
  323         actual_gpus[gpu_type_key] = actual_gpus.get(gpu_type_key, 0) + spec.gpu
  324 
  325     total_budget = sum(plan.gpu_budget.values())

---

Lines 327-339

  327     budget_exceeded = total_actual > total_budget
  328 
  329     warnings: list[str] = []
  330     if budget_exceeded:
! 331         mig_hint = (
  332             "Enable MIG to fit lightweight services (kokoro, yolo, realesrgan) "
  333             "on shared GPU slices."
  334         ) if not MIG_AVAILABLE else ""
! 335         warnings.append(
  336             f"Deployment requires {total_actual} full GPUs but budget is "
  337             f"{total_budget}. {mig_hint}"
  338         )

---

streamwise/container_config.py

Lines 12-24

  12 import os
  13 
  14 _HERE = os.path.dirname(os.path.abspath(__file__))
  15 if _HERE not in sys.path:
! 16     sys.path.insert(0, _HERE)
  17 
  18 _REPO_ROOT = os.path.dirname(_HERE)
  19 if _REPO_ROOT not in sys.path:
! 20     sys.path.insert(0, _REPO_ROOT)
  21 
  22 # model_provisioner import adds simulator/ to sys.path
  23 import model_provisioner  # noqa: E402, F401

---

streamwise/streamwise.py

Lines 551-559

  551 
  552 @route("/auto_deploy", methods=["GET"])
  553 async def auto_deploy_page() -> str:
  554     """Render the standalone auto-deploy page. (TODO: enable customization after auto-deploy plan is generated)"""
! 555     return await render_template("auto_deploy.html")
  556 
  557 
  558 @route("/api/pod/<pod_name>", methods=["DELETE"])
  559 async def api_remove_pod(pod_name: str) -> QuartReturn:

---

Lines 569-596

  569 
  570 @route("/api/pods/wrappers", methods=["DELETE"])
  571 async def api_delete_all_wrappers() -> QuartReturn:
  572     """Delete all wrapper pods (non-app, non-system pods) in the namespace."""
! 573     svcs = await get_services(namespace=NAMESPACE, k8s_cluster=k8s_cluster)
  574     # Exclude app pods and the streamwise management pod itself
! 575     excluded = set(STREAMWISE_APPS) | {"streamwise"}
! 576     wrapper_pods = [
  577         svc["pod_name"] for svc in svcs
  578         if svc.get("container_name") not in excluded and svc.get("pod_name")
  579     ]
! 580     deleted = 0
! 581     errors: list[str] = []
! 582     for pod_name in wrapper_pods:
! 583         try:
! 584             await pod_manager.remove_pod(
  585                 pod_name, namespace=NAMESPACE, k8s_cluster=k8s_cluster)
! 586             deleted += 1
! 587         except Exception as e:
! 588             errors.append(f"{pod_name}: {e}")
! 589     result: dict[str, object] = {"deleted": deleted, "total": len(wrapper_pods)}
! 590     if errors:
! 591         result["errors"] = errors
! 592     return jsonify(result), HTTPStatus.OK
  593 
  594 
  595 @route("/api/services", methods=["GET"])
  596 async def api_get_services() -> QuartReturn:

---

Lines 631-657

  631     try:
  632         # Build container_dict from shared constants (CONTAINER_RESOURCES + MIG_CONTAINERS).
  633         # Format: container_name -> (cpu, memory_gib, ephemeral_storage_gib, gpu_info)
  634         # gpu_info is either an int (GPU count) or a MIG profile string.
! 635         container_dict: dict[str, tuple[int, int, int, Union[int, str]]] = {}
  636 
  637         # Services not in CONTAINER_RESOURCES (CPU-only or extra services)
! 638         container_dict["podcasttranscript"] = (1, 4, 16, 0)
! 639         container_dict["slidetranscript"] = (1, 4, 16, 0)
  640 
! 641         for name, (cpu, mem, storage) in CONTAINER_RESOURCES.items():
! 642             if name in MIG_CONTAINERS:
! 643                 container_dict[name] = (cpu, mem, storage, MIG_CONTAINERS[name])
  644             else:
! 645                 container_dict[name] = (cpu, mem, storage, min(2, max_gpus))
  646 
  647         # Additional services not covered by CONTAINER_RESOURCES
! 648         container_dict["fluxkontext"] = (12, 128, 64, 1)
! 649         container_dict["whisper"] = (2, 8, 16, 1)
  650 
  651         # hunyuanframepackvae uses exactly 1 GPU (not scaled by max_gpus)
! 652         cpu, mem, storage = CONTAINER_RESOURCES["hunyuanframepackvae"]
! 653         container_dict["hunyuanframepackvae"] = (cpu, mem, storage, 1)
  654 
  655         for container_name, (cpu, mem_gib, sotrage_gib, gpu_info) in container_dict.items():
  656             num_gpus, mig_profile = parse_gpu_info(gpu_info)
  657             await pod_manager.add_pod(

---

Lines 788-796

  788         if not gpu_budget or not isinstance(gpu_budget, dict):
  789             return jsonify({"error": "Missing or invalid 'gpu_budget' field"}), HTTPStatus.BAD_REQUEST
  790         for gpu_type_name, count in gpu_budget.items():
  791             if isinstance(count, bool) or not isinstance(count, int) or count < 0:
! 792                 return (
  793                     jsonify(
  794                         {
  795                             "error": (
  796                                 "Invalid 'gpu_budget' field: each GPU type count must be a "

---

Lines 819-836

  819         return jsonify(result_json), HTTPStatus.OK
  820 
  821     except ValueError as ve:
  822         return jsonify({"error": str(ve)}), HTTPStatus.BAD_REQUEST
! 823     except AssertionError as ae:
! 824         msg = str(ae) if str(ae) else (
  825             "GPU budget too small. Each GPU type must have at least 8 GPUs "
  826             "(one full server). Use a single GPU type with 8+ GPUs, or "
  827             "ensure each type has at least 8."
  828         )
! 829         return jsonify({"error": msg}), HTTPStatus.BAD_REQUEST
! 830     except Exception as ex:
! 831         logging.exception("Error in auto_deploy: %s", ex)
! 832         return jsonify({"error": str(ex)}), HTTPStatus.INTERNAL_SERVER_ERROR
  833 
  834 
  835 @route("/api/auto_deploy/confirm", methods=["POST"])
  836 async def api_auto_deploy_confirm() -> QuartReturn:

---

Lines 861-870

  861 
  862         for spec in specs:
  863             container_name = spec.get("container_name")
  864             if not container_name:
! 865                 errors.append("Spec missing 'container_name'")
! 866                 continue
  867 
  868             try:
  869                 add_pod_result = await pod_manager.add_pod(
  870                     container_name=container_name,

---

Lines 882-891

  882                 if isinstance(add_pod_result, tuple) and len(add_pod_result) >= 2:
  883                     status_value = add_pod_result[1]
  884                     if isinstance(status_value, HTTPStatus):
  885                         status_code = status_value
! 886                     elif isinstance(status_value, int):
! 887                         status_code = HTTPStatus(status_value)
  888 
  889                 if status_code >= HTTPStatus.BAD_REQUEST:
  890                     msg = f"Failed to deploy '{container_name}' (status={int(status_code)})"
  891                     logging.error(msg)

---

Lines 891-907

  891                     logging.error(msg)
  892                     errors.append(msg)
  893                 else:
  894                     deployed.append(container_name)
! 895             except Exception as pod_ex:
! 896                 msg = f"Failed to deploy '{container_name}': {pod_ex}"
! 897                 logging.error(msg)
! 898                 errors.append(msg)
  899 
  900         # Also deploy the application container if workflow is specified
  901         if workflow and workflow in STREAMWISE_APPS:
! 902             try:
! 903                 add_pod_result = await pod_manager.add_pod(
  904                     container_name=workflow,
  905                     cpu=4,
  906                     memory_gib=16,
  907                     ephemeral_storage_gib=16,

---

Lines 910-934

  910                     mig_profile=None,
  911                     namespace=NAMESPACE,
  912                     k8s_cluster=k8s_cluster,
  913                 )
! 914                 status_code = HTTPStatus.OK
! 915                 if isinstance(add_pod_result, tuple) and len(add_pod_result) >= 2:
! 916                     status_value = add_pod_result[1]
! 917                     if isinstance(status_value, HTTPStatus):
! 918                         status_code = status_value
! 919                     elif isinstance(status_value, int):
! 920                         status_code = HTTPStatus(status_value)
! 921                 if status_code >= HTTPStatus.BAD_REQUEST:
! 922                     msg = f"Failed to deploy app '{workflow}' (status={int(status_code)})"
! 923                     logging.error(msg)
! 924                     errors.append(msg)
  925                 else:
! 926                     deployed.append(workflow)
! 927             except Exception as app_ex:
! 928                 msg = f"Failed to deploy app '{workflow}': {app_ex}"
! 929                 logging.error(msg)
! 930                 errors.append(msg)
  931 
  932         total_deployed = len(deployed)
  933         total_specs = len(specs) + (1 if workflow and workflow in STREAMWISE_APPS else 0)
  934         status = HTTPStatus.OK if not errors else HTTPStatus.MULTI_STATUS

---

Lines 937-947

  937             "errors": errors,
  938             "message": f"Deployed {total_deployed}/{total_specs} containers.",
  939         }), status
  940 
! 941     except Exception as ex:
! 942         logging.exception("Error in auto_deploy/confirm: %s", ex)
! 943         return jsonify({"error": str(ex)}), HTTPStatus.INTERNAL_SERVER_ERROR
  944 
  945 
  946 @route("/api/auto_deploy/workflows", methods=["GET"])
  947 async def api_auto_deploy_workflows() -> QuartReturn:

---

Lines 971-989

  971             gpu_count = node.get("allocatable_resources", {}).get("gpu", 0)
  972             if isinstance(gpu_count, str):
  973                 try:
  974                     gpu_count = int(gpu_count)
! 975                 except ValueError:
! 976                     continue
  977             if gpu_count <= 0:
! 978                 continue
  979             # Map gpu_model label to canonical type name
  980             canonical = _gpu_label_to_canonical(gpu_model)
  981             gpu_counts[canonical] = gpu_counts.get(canonical, 0) + gpu_count
  982         return jsonify({"gpu_budget": gpu_counts}), HTTPStatus.OK
! 983     except Exception as ex:
! 984         logging.exception("Error in cluster_gpus: %s", ex)
! 985         return jsonify({"error": str(ex)}), HTTPStatus.INTERNAL_SERVER_ERROR
  986 
  987 
  988 def _gpu_label_to_canonical(gpu_model: str) -> str:
  989     """Map a GPU product label to a canonical type name for the allocator."""

---

Lines 990-1010

   990     model_upper = gpu_model.upper()
   991     if "H100" in model_upper:
   992         return "H100"
   993     elif "H200" in model_upper:
!  994         return "H200"
   995     elif "A100" in model_upper:
   996         return "A100"
!  997     elif "GB200" in model_upper:
!  998         return "GB200"
!  999     elif "GB300" in model_upper:
! 1000         return "GB300"
! 1001     elif "V100" in model_upper:
! 1002         return "V100"
! 1003     elif "A10" in model_upper:
! 1004         return "A10"
  1005     # Fallback: return as-is
! 1006     return gpu_model
  1007 
  1008 
  1009 @route("/api/node/<node_name>", methods=["DELETE"])
  1010 async def api_remove_node(node_name: str) -> QuartReturn:

---

tests/streamwise/test_allocator_bridge.py

Lines 251-263

  251     plan = run_allocator(
  252         gpu_budget={"A100": 8},
  253         workflow_name="streamcast",
  254     )
! 255     assert isinstance(plan, DeploymentPlan)
! 256     assert len(plan.specs) > 0
! 257     assert plan.result.total_time_s > 0
! 258     assert plan.result.ttff_s > 0
! 259     assert plan.workflow_name == "streamcast"
  260 
  261 
  262 def test_run_allocator_streamchat_8_h100() -> None:
  263     """Run allocator for StreamChat with 8 H100s."""

---

Lines 264-273

  264     plan = run_allocator(
  265         gpu_budget={"H100": 8},
  266         workflow_name="streamchat",
  267     )
! 268     assert isinstance(plan, DeploymentPlan)
! 269     assert len(plan.specs) > 0
  270 
  271 
  272 def test_run_allocator_invalid_workflow() -> None:
  273     """Unknown workflow name raises ValueError."""

---

[github-actions]

Package	Line Rate	Complexity	Health
.	80%	0	✔
apps	67%	0	✔
apps.streamanimate	89%	0	✔
apps.streamcast	91%	0	✔
apps.streamchat	88%	0	✔
apps.streamdub	64%	0	✔
apps.streamedit	79%	0	✔
apps.streamlecture	94%	0	✔
apps.streammovie	89%	0	✔
apps.streampersona	64%	0	✔
apps.streamshort	71%	0	✔
simulator	90%	0	✔
streamwise	67%	0	✔
streamwise.model_provisioner	83%	0	✔
tests	99%	0	✔
tests.simulator	100%	0	✔
tests.streamwise	99%	0	✔
tests.streamwise_app	99%	0	✔
wrapper	67%	0	✔
wrapper.4kagent	94%	0	✔
wrapper.bagel	37%	0	❌
wrapper.cogview	45%	0	➖
wrapper.fantasytalking	55%	0	➖
wrapper.flux	74%	0	✔
wrapper.flux2	100%	0	✔
wrapper.flux2klein	99%	0	✔
wrapper.fluxkontext	100%	0	✔
wrapper.fluxkrea	99%	0	✔
wrapper.fluxupscaler	68%	0	✔
wrapper.hidream	88%	0	✔
wrapper.hunyuanavatar	74%	0	✔
wrapper.hunyuanframepack	52%	0	➖
wrapper.hunyuanframepackf1	59%	0	➖
wrapper.hunyuanframepackvae	63%	0	✔
wrapper.hunyuanimage	84%	0	✔
wrapper.imageresize	100%	0	✔
wrapper.januspro	91%	0	✔
wrapper.kokoro	87%	0	✔
wrapper.llamagen	65%	0	✔
wrapper.mock	86%	0	✔
wrapper.podcasttranscript	45%	0	➖
wrapper.qwenimage	89%	0	✔
wrapper.qwenimageedit	88%	0	✔
wrapper.realesrgan	77%	0	✔
wrapper.slidetranscript	59%	0	➖
wrapper.vibevoice	31%	0	❌
wrapper.vibevoice.schedule	50%	0	➖
wrapper.wan	34%	0	❌
wrapper.wan22	75%	0	✔
wrapper.xtts	75%	0	✔
wrapper.yolo	69%	0	✔
Summary	78% (27214 / 34734)	0	✔

[goiri]

Let's move this to a different PR. It should have a different structure too.

[goiri]

Let's do this in a separate PR.

[goiri]

Change the emoji to the robot?

[goiri]

Can we get these names from services.json?

[goiri]

Where are we getting these second names from? Why not just use the apps? We have the names in some class no?

[goiri]

Don't we have something similar in the add_pod? Can we reuse?

[goiri]

Do we really need this?