Upload 3 files

docs/NEXT_STEPS.md

@@ -1,69 +1,33 @@
-# Next steps
+# Next Steps
 
-## Current milestone
+## Current status
 
-V2 validates private Space creation and live Gradio API validation.
+- Phase 1 validates OAuth → Job → Bucket.
+- Phase 2 validates private Space creation/upload → live API validation.
+- Phase 3 validates Pi inside an HF Job editing `app.py` before upload.
+- Phase 4 validates a closer article-style workflow: Pi receives the HF Spaces Agent Quickstart gist, uses `hf` CLI, creates/uploads a private Space, and the wrapper independently validates the live API.
 
-## Phase 3 — Pi inside the Job
+## Next pass: Phase 5
 
-Add Pi as the first agentic runtime:
+Add a narrow model-card workflow:
 
-```text
-HF Job
-→ install Pi
-→ configure HF Inference Providers
-→ give Pi a simple goal
-→ let Pi modify the generated hello Space
-→ upload files
-→ validate live API
-→ collect Pi traces from ~/.pi/agent/sessions
-→ copy raw/private traces into the Bucket
-```
+1. User provides a model ID.
+2. Worker fetches `model_info`, README, pipeline tag, library name, and sibling files.
+3. Worker selects one of a few safe templates.
+4. Pi applies the same gist-style loop to create a private Space.
+5. Wrapper validates the live API.
 
-The first Pi task should be deliberately small:
+Recommended first model families:
 
-```text
-Improve the title/description/examples of the generated Gradio app.
-Do not change the function signature.
-Do not mark success until the live Gradio API test passes.
-```
+- simple Transformers text generation / text classification
+- ASR with `pipeline("automatic-speech-recognition")`
+- Diffusers text-to-image only after the CPU/simple path is stable
 
-## Phase 4 — Apply the gist recipe
+Keep these out of scope for Phase 5:
 
-Give Pi a goal close to the article:
-
-```text
-Read the HF Spaces Agent Quickstart gist.
-The deployed Space is your AI lab.
-Use hf CLI / Hub API.
-Read logs before patching.
-Verify through gradio_client.
-Do not declare success until live API validation passes.
-```
-
-## Phase 5 — First model-card template
-
-Add one narrow model family, ideally a simple Transformers pipeline or a standard Diffusers text-to-image model.
-
-## Phase 6 — ZeroGPU
-
-Add a ZeroGPU-specific Gradio template with `@spaces.GPU` only around inference.
-
-## Phase 2.1 fix
-
-This version fixes private Space validation with `gradio_client>=2.0.0`, whose
-`Client` constructor uses `token=` instead of `hf_token=`. The worker now uses
-signature introspection so older/newer versions can still work. Bucket events
-are also printed to stdout so HF Job logs are readable while the worker runs.
-
-
-## V3 notes
-
-Phase 3 is a controlled Pi smoke test, not the full gist workflow yet. The next pass should make Pi read the HF Spaces Agent Quickstart gist and operate on the already-created target Space with logs → patch → live API validation.
-
-Potential improvements after the smoke test passes:
-
-- replace apt-based Node install with a Python+Node base image to reduce startup time;
-- expose a model/provider picker with known-good defaults;
-- improve Pi trace links for the Hub trace viewer;
-- add stronger secret scanning before showing traces.
+- Docker Spaces
+- multi-GPU
+- fine-tuning
+- ComfyUI
+- video generation
+- public publication automation