Add submission pack and MiniCPM5 llama.cpp runtime

FIELD_NOTES.md

@@ -19,6 +19,20 @@ time windows, waiting time, lateness, and a manual baseline comparison.
 - Its model card highlights local deployment, tool use, long context, and
   compact agent workflows, which fit this route-coaching task.
 
+## Competition fit
+
+The project is intentionally small in both model size and product scope. It is
+not a general logistics platform. It handles one common small-business workflow:
+read the daily order sheet, interpret the dispatcher note, and produce a route
+plan that a human can audit.
+
+This directly targets the hackathon signals:
+
+- Backyard AI: practical helper for a local delivery operator.
+- Off the Grid: no cloud LLM API.
+- Llama Champion: MiniCPM5 GGUF is loaded through llama.cpp when available.
+- Sharing is Caring: the planner trace is included as `agent_trace.json`.
+
 ## What the model does
 
 MiniCPM5 receives dispatcher notes such as:
@@ -49,4 +63,3 @@ model invent routes or metrics.
 The demo data is synthetic. The app stores nothing, uses no cloud LLM API, and
 does not require user secrets. Uploaded CSVs are processed only during the
 Gradio session.
-

README.md

@@ -5,6 +5,7 @@ colorFrom: green
 colorTo: yellow
 sdk: gradio
 sdk_version: 6.14.0
+python_version: 3.12
 app_file: app.py
 pinned: false
 license: mit
@@ -17,6 +18,10 @@ tags:
   - openbmb
   - operations-research
   - logistics
+models:
+  - openbmb/MiniCPM5-1B-GGUF
+preload_from_hub:
+  - openbmb/MiniCPM5-1B-GGUF MiniCPM5-1B-Q4_K_M.gguf
 ---
 
 # Tiny Dispatch Coach
@@ -41,16 +46,12 @@ Spaces, and models under 32B parameters.
 - Model repo: `openbmb/MiniCPM5-1B-GGUF`
 - File: `MiniCPM5-1B-Q4_K_M.gguf`
 - Parameter count: `1.08B`
-- Runtime target: local GGUF through `llama-cpp-python` when the Space runtime
-  has a prebuilt llama.cpp wheel or enough memory for the dependency
+- Runtime target: local GGUF through `llama-cpp-python`
 - Cloud LLM APIs: none
 
-The public CPU Basic Space keeps the app responsive by treating the MiniCPM5
-runtime as optional. If `llama-cpp-python` is unavailable during a cold start,
-the app falls back to a deterministic parser and makes that visible in the
-parser trace. On a larger Space or local machine with `llama-cpp-python`
-installed, the same code path downloads `openbmb/MiniCPM5-1B-GGUF` and uses
-MiniCPM5 for note parsing.
+The Space preloads the Q4 MiniCPM5 GGUF file and installs the CPU llama.cpp
+wheel. If a runtime cold start still cannot load the model, the app falls back
+to a deterministic parser and makes that visible in the parser trace.
 
 The route optimizer never depends on hidden model output: every route, time
 window, lateness minute, and baseline delta is computed deterministically.
@@ -59,7 +60,7 @@ window, lateness minute, and baseline delta is computed deterministically.
 
 Included now:
 
-- MiniCPM5-ready text constraint parsing with deterministic CPU fallback.
+- MiniCPM5 text constraint parsing with deterministic CPU fallback.
 - Capacity-safe multi-trip route planning.
 - Manual baseline comparison.
 - Synthetic sample data only.

SUBMISSION.md

@@ -0,0 +1,80 @@
+# Build Small Submission Plan
+
+## Project
+
+- Space: https://huggingface.co/spaces/build-small-hackathon/tiny-dispatch-coach
+- Runtime: https://build-small-hackathon-tiny-dispatch-coach.hf.space/
+- Track: Backyard AI
+- Model: `openbmb/MiniCPM5-1B-GGUF`
+- File: `MiniCPM5-1B-Q4_K_M.gguf`
+- Parameters: 1.08B
+- Runtime path: local GGUF through `llama-cpp-python`
+
+## Why It Fits Build Small
+
+Tiny Dispatch Coach solves a narrow operational problem: turn a small delivery
+sheet and messy dispatcher notes into an auditable route plan. The model does
+not invent routes. MiniCPM5 parses human instructions into a compact constraint
+schema, then deterministic code computes time windows, capacity, route splits,
+late minutes, waiting time, and baseline deltas.
+
+This makes the small model useful because the task is bounded:
+
+- Extract route constraints from natural language notes.
+- Keep all route math deterministic and inspectable.
+- Run without cloud LLM APIs.
+- Use only synthetic demo data.
+
+## Bonus Quest Alignment
+
+- OpenBMB Awards: uses `openbmb/MiniCPM5-1B-GGUF`.
+- Off the Grid: no cloud LLM API or external inference service.
+- Llama Champion: MiniCPM5 runs through `llama-cpp-python` when available.
+- Field Notes: see `FIELD_NOTES.md`.
+- Sharing is Caring: see `agent_trace.json`.
+
+## Demo Video Script
+
+1. Open the Space and point to the OpenBMB MiniCPM5 badges.
+2. Leave the CSV empty so the synthetic sample is used.
+3. Read the default dispatcher note: start time, urgent school/clinic stops,
+   fresh produce before lunch, van capacity 18.
+4. Click **Plan route**.
+5. Show the parser trace: MiniCPM5 path or explicit deterministic fallback.
+6. Show the Dispatch Score:
+   - Manual late minutes: 207.
+   - Tiny Dispatch Coach late minutes: 0.
+   - On-time rate: 100%.
+   - Capacity split: 3 trips.
+7. Show the driver cards and route map.
+8. Close with the privacy stance: synthetic data, no API keys, no customer data,
+   no cloud LLM API.
+
+## Social Post Draft
+
+I built Tiny Dispatch Coach for the Build Small Hackathon:
+
+Small delivery teams often have messy notes, tight windows, and a van capacity
+constraint. This Gradio Space uses OpenBMB MiniCPM5-1B-GGUF to parse dispatch
+notes into route constraints, then a deterministic planner creates auditable
+driver routes.
+
+No cloud LLM API. Synthetic demo data only. 1.08B params.
+
+Space: https://huggingface.co/spaces/build-small-hackathon/tiny-dispatch-coach
+
+#BuildSmallHackathon #HuggingFace #Gradio #OpenBMB #MiniCPM
+
+## Final Submission Checklist
+
+- [x] Public Hugging Face Space.
+- [x] Gradio app.
+- [x] Model under 32B parameters.
+- [x] OpenBMB model listed in README metadata.
+- [x] Synthetic sample data only.
+- [x] No secrets or real customer records.
+- [x] Field notes included.
+- [x] Agent trace included.
+- [ ] Record short demo video.
+- [ ] Publish social post.
+- [ ] Submit Space link, video link, and social post link before the deadline.

app.py

@@ -742,6 +742,12 @@ with gr.Blocks(
 `order_id`, `customer`, `lat`, `lng`, `demand`, `service_min`, `ready_time`, `due_time`, `priority`, `notes`, optional `manual_sequence`.
 
 Leave the file empty to run the included sample route.
+"""
+            )
+            gr.Markdown(
+                """
+### Build Small fit
+OpenBMB MiniCPM5, 1.08B parameters, local GGUF path, no cloud LLM API, synthetic sample data, explicit parser trace.
 """
             )
 

requirements.txt

@@ -1,3 +1,5 @@
 gradio>=6.14.0
 pandas>=2.2.0
 huggingface_hub>=0.34.0
+--extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu
+llama-cpp-python==0.3.28